jessica dos anjos oliveira -...
TRANSCRIPT
INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA - INPA
PROGRAMA DE PÓS-GRADUAÇÃO EM ECOLOGIA
COMO A PAISAGEM MOLDA O PADRÃO ESPACIAL DE VARIAÇÃO GENÉTICA DOS
QUELÔNIOS AMAZÔNICOS PODOCNEMIS ERYTHROCEPHALA E P.
SEXTUBERCULATA (TESTUDINES, PODOCNEMIDIDAE)?
JESSICA DOS ANJOS OLIVEIRA
Manaus, Amazonas
Abril, 2017
JESSICA DOS ANJOS OLIVEIRA
COMO A PAISAGEM MOLDA O PADRÃO ESPACIAL DE VARIAÇÃO GENÉTICA DOS
QUELÔNIOS AMAZÔNICOS PODOCNEMIS ERYTHROCEPHALA E P. SEXTUBERCULATA
(TESTUDINES, PODOCNEMIDIDAE)?
Orientadora: Dra. FERNANDA DE PINHO WERNECK
Co-orientadores: Dra. Izeni Pires Farias
Dr. Gabriel Corrêa Costa
Manaus, Amazonas
Abril, 2017
Dissertação apresentada ao
Instituto Nacional de Pesquisas
da Amazônia como parte dos
requerimentos para obtenção do
título de Mestre em Biologia
(Ecologia) em abril de 2017.
i
Banca examinadora da defesa oral pública
ii
Ficha catalográfica
O48 Oliveira , Jessica dos Anjos
Como a paisagem molda o padrão espacial de variação genética
dos quelônios amazônicos Podocnemis erythrocephala e P.
sextuberculata (Testudines, Podocnemididae)? / Jessica dos Anjos
Oliveira . --- Manaus: [s.n.], 2017.
112 f.: il., color.
Dissertação (Mestrado) --- INPA, Manaus, 2017.
Orientador: Fernanda de Pinho Werneck
Coorientador: Gabriel Corrêa Costa; Izeni Pires Farias
Área de concentração: Biologia (Ecologia)
1. Quelônios . 2.Genética da paisagem . 3. Isolamento por resistência.
I. Título.
CDD 597.92
Sinopse
Estudou-se a influência de fatores locais e de conectividade da paisagem nos padrões espaciais de diversidade e
diferenciação genéticas de duas espécies de quelônios aquáticos amazônicos.
Palavras-Chave: Genética da Paisagem, Isolamento por Resistência, diversidade genética, diferenciação
genética, Ecologia da Paisagem.
iii
AGRADECIMENTOS
Tenho muitas pessoas a agradecer por contribuições diretas e indiretas ao sucesso desta dissertação.
De fato, na Ciência não se faz nada sozinha, e eu não poderia deixar de reconhecer a todos que me
ajudaram ao longo do processo de fazer Ciência!
Primeiramente, agradeço à minha orientadora Fernanda Werneck por me aceitar como
aluna e apoiar o desenvolvimento deste projeto. Graças ao seu suporte fui aos poucos moldando as
questões e abordagens deste trabalho. Muito obrigada também pela amizade, pelas conversas, pela
compreensão, pela disposição em ajudar e por acompanhar todas as etapas do trabalho. Agradeço
à minha co-orientadora Izeni Farias, que é uma mãezona e sempre se preocupou em me ajudar a
ter sucesso na obtenção dos dados genéticos. Obrigada pelo constante apoio no processo de coleta
dos dados genéticos, bem como ao longo da concepção do projeto, da minha aula de qualificação
e escrita final. Também sou grata ao meu co-orientador Gabriel Costa, cuja disponibilidade no
período que passei em Natal foi fundamental para a análise dos dados. Agradeço por me ensinar
modelagem espacial, gastar dias debatendo as análises espaciais adequadas, seja por Skype ou
pessoalmente, e por todos os inputs na concepção e desenvolvimento do projeto e no texto.
Este projeto foi possível graças a um extenso banco de dados genéticos e de amostras
biológicas provenientes do esforço coletivo de diversos pesquisadores. Agradeço primeiramente à
Maria das Neves Viana por disponibilizar o banco de dados de Podocnemis sextuberculata antes
de estar disponível no GenBank e pelo apoio durante meu trabalho em laboratório com as amostras
adicionais de ambas espécies. Agradeço à Izeni por também disponibilizar o banco de dados de
Podocnemis erythrocephala antes de estar disponível no GenBank. Sou grata a todos os que
coletaram amostras de quelônios deste banco de dados e que me ajudaram na busca pelas
coordenadas de cada localidade, especialmente: Richard Vogt, Paulo Andrade, Rafael Bernhard,
Daniely Félix, Francivane Fernandes, José Erickson e Cleiton Fantin. Também agradeço ao
Richard Vogt e à Fernanda Werneck pelas oportunidades de ida a expedições de campo no Parque
Nacional do Jaú e ao longo do Rio Negro, onde coletei amostras adicionais para o trabalho.
A coleta de dados genéticos rendeu alguns meses no LEGAL (Laboratório de Evolução e
Genética Animal/UFAM). Obrigada Maria Augusta por ter me acompanhado e ensinado tudo no
lab e por ser minha companheira de genética de quelônios! Agradeço ao Giovanni, que aprendeu
iv
rapidamente os protocolos e me auxiliou bastante no laboratório. E obrigada ao pessoal do LEGAL
pela convivência no período de lab e por toda ajuda que me deram quando precisei.
A coleta de dados da paisagem foi a parte do meu trabalho na qual contei com mais ajuda,
insight e dados de diversas pessoas. Agradeço ao Urbano Silva Jr. por disponibilizar os dados de
variação interanual nas cotas dos rios. Sou eternamente grata pela ajuda que tive da Camila
Fagundes e da Camila Ferrara, pensando nas variáveis de paisagem e na interpretação dos
resultados. Conversar com vocês duas sobre o aspecto biológico e espacial do trabalho foi essencial
para eu saber se estava indo no caminho certo e ter novas ideias. À Camila Fagundes agradeço
também pela contribuição na minha aula de qualificação, pela disponibilização dos pontos de
ocorrência das espécies, além de shapefiles e ideias sobre variáveis. Também sou grata ao André
Antunes e à Thaís Morcatty por me ajudarem a pensar na variável de pressão de caça. Agradeço
aos especialistas de quelônios que contribuíram com o questionário de resistência à cor da água:
Richard Vogt, Paulo Andrade, Rafael Bernhard, Augusto Fachín-Terán, Camila Ferrara e Camila
Fagundes. As variáveis de conectividade só puderam ser desenvolvidas graças ao apoio de Felipe
Martello e seus scripts. Sou também extremamente grata à professora Marina Côrtes da UNESP
Rio Claro, pelos ensinamentos na disciplina de Genética da Paisagem ministrada, além das
importantes contribuições dadas pessoalmente e via Skype sobre as variáveis de paisagem e
análises espaciais.
Agradeço aos amigos e colegas do grupo Evolução e Biogeografia da Biota Amazônica
(EBBA) e à Ariane pelas discussões de artigos, reuniões, confraternizações e amizade. Também
agradeço as sugestões de Igor Kaefer, Waleska Gravena e Camila Fagundes, membros da banca da
minha aula de qualificação.
Muito obrigada às maravilhosas pessoas da turma da Ecologia de 2015 por estes dois anos!
Sou grata também aos moradores da república Viracopos e moradores das kitnets rosinhas pela
convivência. Um obrigada especial à Sulamita, minha irmã de Amazônia. E nada disso teria sido
possível sem o apoio incondicional da minha família e amigos de Brasília, vocês foram
fundamentais!
Finalmente, agradeço ao CNPq por conceder a bolsa de Mestrado. E ao CNPq e à FAPEAM
pelo financiamento dos campos e laboratório.
v
Resumo
Associações entre fatores da paisagem e processos ecológicos como dispersão, reprodução e
sobrevivência de organismos podem afetar processos microevolutivos como fluxo gênico, deriva
e seleção. A Genética da Paisagem surgiu como um campo de pesquisa que combina genética
populacional, ecologia de paisagens e análises espaciais para quantificar explicitamente os efeitos
da qualidade da matriz, da composição e configuração da paisagem nos processos microevolutivos.
Em sistemas fluviais, como a bacia Amazônica, o processo de isolamento por distância é
normalmente forte e pode mascarar a importância de barreiras, resistência da paisagem e fatores
ambientais locais em moldar padrões genéticos. Portanto, nesta dissertação eu avaliei a importância
de variáveis locais e de conectividade em moldar os padrões espaciais de variação genética de dois
quelônios aquáticos Amazônicos com diferentes capacidades dispersoras. Meus objetivos foram:
1) avaliar se a espécie com maior capacidade de dispersão (Podocnemis sextuberculata) possui
menor estrutura genética especial que a espécie com baixa capacidade dispersora (P.
erythrocephala); 2) testar se fatores de conectividade estão relacionados à diferenciação genética
para a espécie de baixa capacidade dispersora (P. erythrocephala) mas não para a de alta
capacidade dispersora (P. sextuberculata); e 3) testar se fatores locais estão mais fortemente
associados à diversidade genética intrapopulacional da espécie de baixa capacidade de dispersão
(P. erythrocephala). Com ampla amostragem pela distribuição geográfica das espécies na bacia
Amazônica, eu estimei os parâmetros genéticos para P. erythrocephala em 14 localidades (273
amostras) e para P. sextuberculata em 20 localidades (336 amostras). Apliquei seleção de modelos
em modelos associando a diversidade genética a variáveis locais representando hipóteses de clima
e produtividade, instabilidade interanual de níveis da água dos rios, pressão de caça e aumento de
diversidade genética a jusante dos rios. Usei General Dissimilarity Modelling (GDM) para modelar
a relação entre diferenciação genética e variáveis de conectividade representando hipóteses de
isolamento por distância (IBD), isolamento por resistência (IBR) e isolamento por barreira (IBB).
Diferentemente do esperado, variáveis locais foram mais importantes em explicar a diversidade
genética intrapopulacional da espécie com maior capacidade de dispersão (P. sextuberculata) que
de P. erythrocephala, com melhores modelos incluindo produtividade, distância da localidade mais
a jusante, densidade de vilas humanas e adequabilidade climática histórica. Fatores de
conectividade em geral não foram importantes em explicar a diferenciação genética para nenhuma
das espécies, entretanto, como esperando, os modelos GDM explicaram uma maior parte da
variação para a espécie de menor capacidade dispersora, P. erythrocephala. Além disso, modelos
de IBB e IBR explicaram mais diferenciação genética que IBD, revelando a importância em incluir
a complexidade ambiental e da paisagem quando estudar padrões genéticos espaciais. Nessa
dissertação mostro que, apesar de variáveis locais serem frequentemente desconsideradas em
estudos de Genética da Paisagem, elas podem influenciar a diversidade genética intrapopulacional
de espécies aquáticas, inclusive daquelas com alta capacidade dispersora. Ao usar um método
inédito de modelos de resistência no contexto de Genética da Paisagem de Rios (Riverscape
Genetics) e ao usar fatores da paisagem relevantes no contexto Amazônico, forneço uma
abordagem para o estudo dos papéis de variáveis locais e de conectividade em moldar os padrões
genético-espaciais de vertebrados aquáticos em sistemas fluviais.
vi
Abstract
How does the riverscape shapes the spatial pattern of genetic variation of Amazonian river
turtles Podocnemis erythrocephala and P. sextuberculata (Testudines, Podocnemididae)?
Associations between landscape factors and ecological processes such as dispersal, reproduction
and survival of organisms can ultimately affect microevolutionary processes such as gene flow,
drift and selection. Landscape Genetics emerged as a research field that combines population
genetics, landscape ecology, and spatial analyses to explicitly quantify the effects of landscape
composition, configuration, and matrix quality on microevolutionary processes. In fluvial systems,
such as Amazon basin, the process of isolation by distance is often strong and can mask the
importance of barriers, landscape resistance and local environmental factors on shaping genetic
patterns. Therefore, in this dissertation I assessed the importance of local and connectivity variables
in shaping the spatial genetic variation patterns of two Amazonian river turtle species with distinct
dispersal abilities. My objectives were: 1) assess whether the high-dispersal species (Podocnemis
sextuberculata) has less spatial genetic structure than the low-dispersal species (P.
erythrocephala); 2) test whether connectivity factors are related to genetic differentiation for the
low-dispersal (P. erythrocephala) species but not for the high-dispersal species (P. sextuberculata);
and 3) test whether local factors are more strongly associated to intrapopulational genetic
diversity for the low dispersal species (P. erythrocephala). With broad sampling throughout their
distribution in Amazon basin, I estimated genetic diversity and differentiation for 14 localities
totaling 273 samples of P. erythrocephala and for 20 localities totaling 336 samples of P.
sextuberculata. I applied model selection on models associating genetic diversity to local variables
representing hypothesis of climate and productivity, instability of inter-annual water levels, hunting
pressure and downstream increase in genetic diversity. I used General Dissimilarity Modelling
(GDM) to model the relationship of genetic differentiation with connectivity variables representing
hypothesis of isolation by distance (IBD), isolation by resistance (IBR) and isolation by barrier
(IBB). Differently from the expected, local variables were more important in explaining genetic
diversity of the high-dispersal species (P. sextuberculata) than of P. erythrocephala, with best
models including productivity, distance from downstream locality, density of human villages and
historical climatic suitability. Connectivity factors in general were not important in explaining
genetic differentiation turnover for either species, but as expected, the GDM models explained a
larger amount of deviance for the low-dispersal species, P. erythrocephala. Also, IBB and IBR
models explained more genetic differentiation turnover than IBD, revealing the importance of
including the environmental and landscape complexity when studying spatial genetic patterns. I
showed that, although local variables are often overlooked in Landscape Genetics studies, they can
influence intrapopulacional genetic diversity of aquatic species, even those with high dispersal
ability. By applying a novel resistance-model framework in Riverscape Genetics and by using
riverscape factors relevant in Amazonian context, I provide an approach to study the roles of local
and connectivity variables in shaping genetic patterns of aquatic vertebrates in fluvial systems.
vii
Sumário
Introdução geral ............................................................................................................................. 1
Objetivos .......................................................................................................................................... 7
Capítulo I. – Model-based riverscape genetics: disentangling the roles of local and
connectivity factors in shaping spatial genetic patterns of two Amazon River turtles with
different dispersal abilities ............................................................................................................. 9
INTRODUCTION ...................................................................................................................... 11
METHODS ................................................................................................................................. 15
RESULTS ................................................................................................................................... 21
DISCUSSION ............................................................................................................................. 23
CONCLUSIONS AND PERSPECTIVES .................................................................................. 32
REFERENCES ........................................................................................................................... 33
TABLES ..................................................................................................................................... 46
FIGURE LEGENDS ................................................................................................................... 51
FIGURES .................................................................................................................................... 53
Conclusões ..................................................................................................................................... 59
Apêndices ....................................................................................................................................... 60
1
Introdução geral
Associações entre fatores da paisagem e processos ecológicos como dispersão, reprodução e
sobrevivência dos organismos podem afetar processos microevolutivos como fluxo gênico, deriva
genética e seleção (Sork e Waits, 2010). Compreender estas associações e seus efeitos é essencial
para a conservação das espécies, uma vez que impactos na variação genética de populações é um
dos principais fatores que podem levar à extinção de espécies (Spielman et al., 2004). A Genética
da Paisagem surgiu como um campo de pesquisa que combina genética populacional, ecologia da
paisagem e análises espaciais para quantificar explicitamente os efeitos da composição e
configuração da paisagem e qualidade da matriz nos processos microevolutivos (Balkenhol et al.,
2016). Desde que o termo foi cunhado (Manel et al., 2003), o campo evoluiu de métodos descritivos
para abordagens que testam hipóteses explícitas e modelam respostas genéticas em relação às
variáveis de paisagem (Cushman et al., 2006; Storfer et al., 2010). Apesar de apenas 15% das
pesquisas de genética da paisagem serem conduzidas em ambientes de água doce (Storfer et al.,
2010), há diversas evidências de estrutura genética em espécies associadas a estes hábitats (Hughes
et al., 2009; Ozerov et al., 2012; Hand et al., 2015). Entretanto, em ambientes de água doce,
especialmente em sistemas fluviais, o processo de isolamento por distância-IBD (Wright, 1943) é
frequentemente forte e pode mascarar a importância de outras variáveis da paisagem em moldar
padrões genéticos (Selkoe et al., 2015). Portanto, em sistemas de rios é essencial implementar
abordagens que separem os efeitos da distância geográfica de outros fatores ambientais.
Apesar do IBD ser responsável por parte da estrutura genética encontrada em populações
de diversos táxons (Jenkins et al., 2010), a heterogeneidade ambiental de paisagens pode afetar a
sincronização dos processos de migração e reprodução entre populações, modificando padrões de
fluxo gênico e assim aumentando a diferenciação genética entre elas (Sexton et al., 2014; Wang e
Bradburd, 2014). Estudos de genética da paisagem de rios (em inglês “Riverscape genetics”)
frequentemente testam barreiras discretas (isolamento por barreira, IBB) como cachoeiras e
represas (Wofford et al., 2005; Deiner et al., 2007; Leclerc et al., 2008; Kanno et al., 2011), mas
fatores menos conspícuos também podem agir como barreiras ao fluxo gênico e causar
diferenciação detectável. Por exemplo, paisagens dendríticas estão hierarquicamente estruturadas
por elevação e portanto o fluxo gênico é tipicamente assimétrico (Selkoe et al., 2015), com a
declividade muitas vezes determinando a variação genética espacial de espécies de água doce
2
(Hughes et al., 2009; Cook et al., 2011; Kanno et al., 2011). Além disso, estudos com diversas
espécies de peixes com diferentes histórias de vida ilustram que dissimilaridades físico-químicas
de massas d’água causam divergência genética (Leclerc et al., 2008; Cooke et al., 2014;
Beheregaray et al., 2015). Entretanto, essas dissimilaridades ambientais são raramente – ou não
são – avaliadas em termos de resistência oferecida à migração entre populações, resultando em
uma falta de estudos empíricos com modelos de resistência da paisagem de rios. Para espécies
terrestres, o uso de caminhos de menor custo (LCPs) e superfícies de resistência tem se mostrado
mais eficaz para prever padrões de fluxo gênico entre localidades do que medidas diretas de
dissimilaridades ou distâncias (Cushman et al., 2006; McRae, 2006; Spear et al., 2010; Wang et
al., 2013). Ainda, a integração de modelos de adequabilidade climática em análises de LCPs pode
melhorar o entendimento sobre conectividade da paisagem, rotas potenciais para dispersão e
distribuição de hábitats adequados às espécies (Wang et al., 2008; Ortego et al., 2015).
Variáveis de conectividade sozinhas normalmente não explicam os padrões espaciais de
estruturação genética observados em populações de água doce, sendo que processos locais podem
também influenciar padrões genéticos neutros (Murphy et al., 2010; Ozerov et al., 2012; Kovach
et al., 2015). Enquanto fatores de conectividade moldam taxas de migração e fluxo gênico, fatores
locais podem determinar tamanhos populacionais efetivos (Ne) e, por deriva genética, deixar um
sinal mais forte na diversidade genética intrapopulacional (Wright, 1931; Frankham, 1996; Wagner
e Fortin, 2013; DiLeo e Wagner, 2016). Apesar da importância da diversidade genética em manter
o fitness de populações e reduzir os riscos de extinção, poucos estudos de genética da paisagem
consideram os efeitos de variáveis locais, em nível de pontos amostrais, na diversidade genética
intrapopulacional (DiLeo e Wagner, 2016). Em redes de rios, diversos fatores contribuem para a
manutenção de padrões genéticos locais por meio de processos neutros (Thomaz et al., 2016). Por
exemplo, muitos táxons de rios apresentam um padrão de acúmulo local de alelos (e, portanto, de
diversidade genética) em regiões a jusante, devido a um fluxo gênico enviesado a jusante
(Downstream Increase in Intraspecific Genetic Diversity – DIGD, Paz‐Vinas et al., 2015). Em
sistemas fluviais sujeitos à dinâmica de áreas alagáveis, outro fator local que pode ter efeitos sobre
a diversidade genética é a instabilidade interanual nos níveis do rio, que impactam a qualidade
anual e localização de hábitats para desova e nidificação de animais (Ouellet‐Cauchon et al., 2014;
Bermudez-Romero et al., 2015). Flutuações interanuais do nível da água resultam em migração
3
forçada por longas distâncias para desovar ou eventos frequentes de extinção/recolonização,
causando reduzida estrutura genética e aumento de deriva genética (Østergaard et al., 2003;
Ouellet‐Cauchon et al., 2014). Além disso, variáveis climáticas e de produtividade também estão
associadas à variação genética neutra e adaptativa da fauna de água doce, provavelmente porque
condições locais adequadas aumentam a persistência populational e oferecem resiliência frente às
mudanças climáticas (Murphy et al., 2010; Vincent et al., 2013; Hand et al., 2015; Kovach et al.,
2015). Adicionalmente, a persistência populacional de animais silvestres pode ser afetada pela
pressão antrópica de caça, por meio de mudanças genéticas como alterações na subdivisão
populacional, perda de variação genética e mudanças genéticas seletivas (Allendorf et al., 2008).
Estudos comparativos elucidam como diferenças ecológicas intrínsecas entre espécies
podem resultar em efeitos distintos de fatores locais e de conectividade da paisagem em padrões
genéticos, além de sugerir efeitos consistentes de uma dada paisagem em mais de uma espécie
(Storfer, 2013). Por exemplo, ao examinarmos espécies proximamente relacionadas com
distribuições geográficas parcial ou totalmente sobrepostas, podemos investigar se diferentes
capacidades dispersivas correlacionam com padrões genéticos distintos exibidos por cada espécie
(Steele et al., 2009). Espécies animais de baixa capacidade de dispersão, comparadas às de maior
capacidade, frequentemente exibem maior divergência genética, menor diversidade genética e
estrutura genética espacial mais acentuada (Gomez‐Uchida et al., 2009; Steele et al., 2009;
Richardson, 2012). Isto pode levar a uma maior resposta genética aos fatores locais e de
conectividade para espécies de baixa capacidade dispersora devido à alta deriva genética e menor
fluxo gênico entre localidades (Gomez‐Uchida et al., 2009). Comparações interespecíficas são
particularmente úteis para guiar estratégias de manejo para organismos ameaçados habitando
paisagens heterogêneas, como quelônios (Reid et al., 2017). Apesar de viverem na interface entre
terra e água e possuírem diversos traços de história de vida influenciados por fatores da paisagem,
quelônios têm sido amplamente ignorados em estudos de genética da paisagem em geral. Existem
poucos estudos de genética da paisagem com jabutis terrestres de desertos (Hagerty et al., 2011,
Latch et al., 2011), e apenas um com espécies aquáticas e semi-aquáticas (Reid et al., 2017), todos
em regiões temperadas. Tal lacuna de conhecimento é preocupante, considerando que quelônios
estão entre os grupos de vertebrados mais ameaçados no mundo (Rhodin et al., 2008) e que a
4
maioria de espécies de quelônios de água doce ocorrem em zonas tropicais e subtropicais
megadiversas (Bour, 2008), como a bacia Amazônica.
A bacia Amazônica é a maior bacia hidrográfica do mundo, composta por um sistema
complexo e ambientalmente heterogêneo formado por rios, riachos e florestas alagáveis com ampla
variação na geomorfologia, dinamismo do pulso de inundação e propriedades físico-químicas da
água (Sioli, 1984). Esta complexidade influencia a movimentação, reprodução e sobrevivência dos
organismos, moldando padrões genéticos populacionais de diversos vertebrados aquáticos (Farias
et al., 2004; De Thoisy et al., 2006; Escalona et al., 2009; Farias et al., 2010; Beheregaray et al.,
2015; Gravena et al., 2015), incluindo tartarugas de rios (Pearse et al., 2006; Santos et al., 2016).
Entretanto, ao meu conhecimento, até o momento nenhum estudo usou uma abordagem
espacialmente explícita baseada em modelos para testar quais fatores de paisagem de rios da bacia
Amazônica podem estar por trás dos padrões genéticos observados. No sistema de estudo,
Podocnemis erythrocephala (conhecida como “irapuca”) é a menor espécie do gênero Podocnemis
ocorrendo na bacia Amazônica. É também a espécie com distribuição geográfica mais restrita,
ocorrendo no Brasil, na Colômbia e na Venezuela, principalmente em rios de águas pretas e seus
tributários (Mittermeier e Wilson, 1974; Pritchard, 1979; Ernst e Barbour, 1989), mas também em
rios e lagos de águas claras (Pritchard, 1979; Hoogmoed e de Avila-Pires, 1990; Vogt et al., 1991;
Iverson, 1992). A segunda menor espécie, Podocnemis sextuberculata (conhecida como “iaçá”), é
amplamente distribuída na drenagem do Rio Amazonas no Peru, na Colômbia e no Brasil (Ernst e
Barbour, 1989; Iverson, 1992), principalmente em grandes rios de águas brancas e claras (Pezzuti
e Vogt, 1999; Pezzuti et al., 2000; Bernhard, 2001; Fachín-Terán et al., 2003). A distribuição
geográfica das duas espécies sobrepõe em algumas regiões de tributários do rio Amazonas (Figura
1). P. sextuberculata é uma espécie de alta capacidade dispersora cujas fêmeas migram longas
distâncias em grupo para nidificar em grandes praias arenosas (Pezzuti e Vogt, 1999), com registro
de até 60 km percorridos por uma fêmea em um ano (Fachín-Terán et al., 2006). Por outro lado,
fêmeas de P. erythrocephala nidificam sozinhas ou em pequenos grupos em regiões de solos
arenosos e vegetação arbustiva (campinas e campinaranas) e praias (Rueda-Almonacid et al., 2007;
Batistella e Vogt, 2009). P. erythrocephala possui movimentos mais curtos (Bernhard, 2010 dados
não publicados) e portanto parece ter menor capacidade dispersora, sendo comumente encontrada
em pequenos riachos e lagos em vez de no canal principal dos grandes rios (Rhodin et al., 2015).
5
Nessa dissertação busquei avaliar a importância de variáveis locais e de conectividade da
paisagem para moldar a variação genética espacial de duas espécies de tartarugas de rio
Amazônicas com diferentes capacidades dispersoras e preferências de hábitat (tipos de água e
substratos de nidificação). Para isso, usei variáveis locais biologicamente relevantes representando
hipóteses de produtividade e clima, instabilidade interanual de níveis de água, pressão de caça e
aumento de diversidade genética a jusante do rio (DIGD). A expectativa é de que essas variáveis
locais reduzam ou aumentem os tamanhos populacionais efetivos (Ne), consequentemente afetando
a taxa de deriva genética e diversidade genética intrapopulacional. Usei também variáveis de
conectividade ambiental representando hipóteses de isolamento por distância (IBD), isolamento
por resistência (IBR) e isolamento por barreira (IBB). Os modelos de IBR incluem resistência
oferecida pelo tipo de rio (custo de águas brancas, pretas e claras para a movimentação de cada
espécie), por hábitats climaticamente inadequados (atual e histórico) e por declividade. Estas
variáveis de conectividade devem restringir a dispersão e padrões reprodutivos entre localidades,
reduzindo o fluxo gênico e aumentando a diferenciação genética entre populações.
6
Figura 1. Distribuição geográfica potencial de Podocnemis erythrocephala (máscara vermelho
escura) e P. sextuberculata (máscara amarela) estimada por Fagundes et al., 2015. A sobreposição
na distribuição das duas espécies está destacada pela máscara laranja. Fotos: P. erythrocephala -
Jessica dos Anjos Oliveira; P. sextuberculata - Claudia Keller.
7
Objetivos
Objetivo geral
Investigar a importância de variáveis ambientais locais e de conectividade para determinar a
variação genética espacial de duas espécies de tartarugas de rio Amazônicas (Podocnemis
erythrocephala e P. sextuberculata) com diferentes capacidades dispersoras.
Objetivos específicos
1. Avaliar se a espécie com maior capacidade dispersora (P. sextuberculata) possui menor
estruturação genética espacial que a espécie que dispersa menos (P. erythrocephala).
2. Testar se fatores de conectividade que reduzem o fluxo gênico estão relacionados à
diferenciação genética para a espécie com menor capacidade dispersora (P. erythrocephala)
mas não para P. sextuberculata (alta dispersão).
3. Testar se fatores locais estão relacionados à diversidade genética intrapopulacional de
ambas espécies, mas com efeito mais forte na diversidade genética da espécie de baixa
dispersão (P. erythrocephala).
8
Capítulo I.
Oliveira, J. A., Farias, I. P., Costa, G. C. & Werneck, F. P. Model-based riverscape
genetics: disentangling the roles of local and connectivity factors in shaping
spatial genetic patterns of two Amazonian turtles with different dispersal
abilities. Manuscrito submetido para Ecography.
9
ORIGINAL RESEARCH
Model-based riverscape genetics: disentangling the roles of local and connectivity factors in
shaping spatial genetic patterns of two Amazonian turtles with different dispersal abilities
Jessica dos Anjos Oliveira1,2*, Izeni Pires Farias2, Gabriel C. Costa3 and Fernanda P. Werneck4
1 Programa de Pós-Graduação em Ecologia, Instituto Nacional de Pesquisas da Amazônia,
69080-971 Manaus, Amazonas, Brazil
2 Laboratório de Evolução e Genética Animal, Departamento de Genética, Universidade Federal
do Amazonas, 69077-000 Manaus, Amazonas, Brazil
3 Department of Biology, Auburn University at Montgomery, Montgomery AL 36124
4 Programa de Coleções Científicas Biológicas, Coordenação de Biodiversidade, Instituto
Nacional de Pesquisas da Amazônia, 69060-000 Manaus, Amazonas, Brazil
Corresponding author: Jessica dos Anjos Oliveira ([email protected])
10
ABSTRACT
In fluvial systems, the process of isolation by distance is often strong and can mask the
importance of barriers, landscape resistance and local environmental factors on shaping genetic
patterns. By comparing two Amazonian river turtle species with distinct dispersal abilities, we
assessed how differently and which local and connectivity variables influence, respectively, their
genetic diversity and differentiation. With broad sampling throughout their distribution in the
Amazon basin, we estimated genetic diversity and differentiation for 14 localities totaling 273
samples of Podocnemis erythrocephala and for 20 localities totaling 336 samples of P.
sextuberculata. We applied model selection on models associating genetic diversity to local
variables representing hypothesis of climate and productivity, instability of inter-annual water
levels, hunting pressure and downstream increase in genetic diversity. We used General
Dissimilarity Modelling to model the relationship of genetic differentiation with connectivity
variables representing hypothesis of isolation by distance (IBD), isolation by resistance (IBR) and
isolation by barrier (IBB). Local variables were more important in explaining genetic diversity of
the high-dispersal species (P. sextuberculata) than of P. erythrocephala, with best models
including productivity, distance from downstream locality, density of human villages and
historical climatic suitability. Connectivity factors in general were not important in explaining
genetic differentiation turnover for either species, but GDM models explained a larger amount of
deviance for the low-dispersal species, P. erythrocephala. Also, IBB and IBR models explained
more genetic differentiation turnover than IBD. We showed that, although local variables are
often overlooked in Landscape/Riverscape Genetics studies, they can influence intrapopulacional
genetic diversity of aquatic species, even those with high dispersal ability. By applying a novel
resistance-model framework in Riverscape Genetics and by using riverscape factors relevant in
Amazonian context, we provide an approach to study the roles of local and connectivity variables
in shaping genetic patterns of aquatic vertebrates in fluvial systems.
Key words: landscape genetics, resistance model, genetic differentiation, genetic diversity,
Amazon basin, Podocnemis erythrocephala, Podocnemis sextuberculata.
11
INTRODUCTION
Associations between landscape factors and ecological processes such as dispersal, reproduction
and survival of organisms can ultimately affect microevolutionary processes such as gene flow,
drift and selection (Sork and Waits 2010). Understanding these associations and their effects is
essential for species conservation because factors that negatively impact the genetic diversity of
populations can eventually drive species extinction (Spielman et al. 2004). Landscape genetics
emerged as a research field that combines population genetics, landscape ecology, and spatial
analyses to explicitly quantify the effects of landscape composition, configuration, and matrix
quality on microevolutionary processes (Balkenhol et al. 2016). Since the term was coined
(Manel et al. 2003), the field evolved from descriptive approaches to explicit hypothesis testing
framework and modeling of genetic responses in response to predictive landscape variables
(Cushman et al. 2006, Storfer et al. 2010). Although only 15% of landscape genetic studies were
conducted in freshwater habitats (Storfer et al. 2010), there is mounting evidence for complex
spatial genetic structure in these habitats (Hughes et al. 2009, Ozerov et al. 2012, Hand et al.
2015). However, in freshwater environments, especially in fluvial systems, the process of
isolation by distance-IBD (Wright 1943) can often overwhelm the importance of other processes
that might shape genetic patterns (Selkoe et al. 2015). As such, in river systems, it is especially
necessary to implement approaches that are able to disentangle the confounding effects of
geographical distance and other environmental factors.
While IBD is responsible for part of the populational genetic structure found in several
taxa (Jenkins et al. 2010), landscape environmental heterogeneity can affect synchronization of
migration and mating processes among populations, modifying gene flow patterns and increasing
genetic differentiation (Wang and Bradburd 2014). Riverscape genetics studies usually test for
discrete barriers (isolation by barrier, IBB) such as waterfalls and dams (Wofford et al. 2005,
Deiner et al. 2007, Kanno et al. 2011), but less conspicuous factors may also act as barriers to
gene flow and cause detectable differentiation. For example, dendritic landscapes are structured
hierarchically by elevation and therefore gene flow is typically asymmetric (Selkoe et al. 2015),
with stream gradient often determining the spatial genetic variation of freshwater species
12
(Hughes et al. 2009, Cook et al. 2011). In addition, studies with a broad range of fish species with
different life-histories showed that physical-chemical dissimilarities of water masses can cause
genetic divergence (Leclerc et al. 2008, Beheregaray et al. 2015). However, these environmental
dissimilarities are rarely – if at all – assessed in terms of resistance to migration between
populations, resulting in a lack of empirical studies with riverscape resistance models. For
terrestrial species, least-cost paths (LCPs) and resistance surfaces have been shown to be better
predictors of gene flow patterns among localities than direct measures of dissimilarity or
distances (Cushman et al. 2006, McRae 2006, Spear et al. 2010, Wang et al. 2013). Also,
integration of climatic suitability models into LCP analyses can improve our understanding on
landscape connectivity, potential routes of dispersal and distribution of suitable habitats for the
species (Wang et al. 2008, Ortego et al. 2015).
Connectivity variables alone often do not explain the observed spatial genetic structure of
freshwater populations, and local processes may also influence neutral genetic patterns (Murphy
et al. 2010, Ozerov et al. 2012, Kovach et al. 2015). While connectivity factors shape migration
and gene flow rates affecting genetic differentiation among populations, local factors can
determine effective population sizes (Ne) and through genetic drift leave a stronger signal in
genetic diversity within populations (Wright 1931, Frankham 1996, Wagner and Fortin 2013,
DiLeo and Wagner 2016). Regardless of the importance of genetic diversity on maintaining
population fitness and reducing extinction risk, very few landscape genetics studies consider the
effects of site-based, local variables on intrapopulational genetic diversity (DiLeo and Wagner
2016). In river networks, several factors can contribute to the maintenance of local genetic
patterns through neutral processes (Thomaz et al. 2016). For example, a broad variety of riverine
taxa show a pattern of local accumulation of genetic diversity in downstream regions due to
downstream-biased gene flow (Downstream Increase in Intraspecific Genetic Diversity – DIGD,
Paz‐Vinas et al. 2015). In river systems subject to floodplain dynamics, another important local
factor is the instability of inter-annual water level, which impacts on yearly quality and
localization of spawning or nesting habitats for animals (Ouellet‐Cauchon et al. 2014, Bermudez-
Romero et al. 2015). The inter-annual water level fluctuations result in forced migration over
13
longer distances to spawn or frequent extinction/recolonization events, causing reduced genetic
structure and increased genetic drift (Østergaard et al. 2003, Ouellet‐Cauchon et al. 2014). Also,
at-site climate and productivity variables are associated to neutral and adaptive genetic variation
for freshwater fauna, likely because suitable conditions can enhance population persistence and
offer resiliency in the face of climate change (Murphy et al. 2010, Vincent et al. 2013, Hand et al.
2015, Kovach et al. 2015). In addition, population persistence of wild animals can be affected by
local human harvest, through genetic changes such as alteration of population subdivision, loss of
genetic variation, and selective genetic changes (Allendorf et al. 2008).
Comparative studies are essential to elucidate how intrinsic ecological differences among
species can generate distinct effects of local and connectivity landscape factors on genetic
patterns (Storfer 2013). For instance, differences in dispersal ability among closely related
species correlates with distinct genetic patterns (Steele et al. 2009). Low-dispersal species,
compared to species with high-dispersal capacities, often exhibit higher genetic divergence, lower
genetic diversity and more pronounced spatial genetic structure (Gomez‐Uchida et al. 2009,
Steele et al. 2009, Richardson 2012). This may lead to stronger genetic response to local and
connectivity factors for poor-dispersers due to increased drift and lower gene flow among
localities (Gomez‐Uchida et al. 2009). These comparisons are particularly useful in guiding
management strategies for threatened organisms inhabiting heterogeneous landscapes, such as
turtles (Reid et al. 2017). Despite living in the land-water interface and having variable life
history traits influenced by landscape factors, turtles have been largely ignored in landscape
genetics studies in general. There are two studies with terrestrial desert tortoises (Hagerty et al.
2011, Latch et al. 2011) and only one with aquatic and semi-aquatic species (Reid et al. 2017), all
in temperate regions. This knowledge gap is concerning, given that turtles are among the most
threatened vertebrate species in the world (Rhodin et al. 2008) and the majority of freshwater
turtle species occur in megadiverse tropical and subtropical zones (Bour 2008), such as the
Amazon basin.
The Amazon basin is the largest hydrographic basin in the world, composed by a complex
and environmentally heterogeneous system formed by rivers, streams and floodplain forests with
14
varying geomorphology, flood pulse dynamism and physical-chemical water properties (Sioli
1984). This complexity influences the movement, mating and survival of organisms, shaping
population genetic patterns of several aquatic vertebrates (Farias et al. 2004, De Thoisy et al.
2006, Escalona et al. 2009, Farias et al. 2010, Beheregaray et al. 2015, Gravena et al. 2015),
including river turtles (Pearse et al. 2006, Santos et al. 2016). However, to our knowledge, no
study attempted to use a spatially explicit model-based framework to test which Amazon basin
riverscape factors may be behind the observed genetic patterns. In our study system, Podocnemis
erythrocephala (Red-headed Amazon River turtle) is the smallest Podocnemis species occurring
in the Amazon basin, reaching a maximum of 32.2 cm of carapace length. It is also the least
broadly distributed, occurring in Brazil, Colombia and Venezuela, mainly in black water rivers
and their tributaries (Mittermeier and Wilson 1974, Pritchard 1979, Ernst and Barbour 1989), but
also in clear water lakes and rivers (Pritchard 1979, Hoogmoed and de Avila-Pires 1990, Vogt et
al. 1991, Iverson 1992). The second smallest species, reaching a maximum of 34 cm of carapace
length, Podocnemis sextuberculata (Six-tubercled Amazon River Turtle), is broadly distributed in
the Amazon River drainage in Peru, Colombia and in Brazil (Ernst and Barbour 1989, Iverson
1992), mainly in large white water and clear water rivers (Pezzuti and Vogt 1999, Pezzuti et al.
2000, Fachín-Terán et al. 2003). The geographical distribution of both species overlaps in a few
regions in Amazon River tributaries. P. sextuberculata is a high-dispersal species whose females
migrate long distances to nest in group in large sandy beaches (Pezzuti and Vogt 1999), with
records of up to 60 km moved by a female in a year (Fachín-Terán et al. 2006). On the other
hand, females of P. erythrocephala nest alone or in small groups in sandy shrub lands or forests
and beaches (Rueda-Almonacid et al. 2007, Batistella and Vogt 2008). Podocnemis
erythrocephala is commonly found in smaller streams and lakes instead of the main river
channels (Mittermeier et al. 2015) and therefore seems to have lower dispersal potential.
Here we assessed the importance of local and connectivity variables in shaping the spatial
genetic variation of two Amazon river turtle species differing in their dispersal abilities and
habitat preferences. For this, we used biologically meaningful local variables representing
hypothesis of climate and productivity, instability of inter-annual water levels, hunting pressure
15
and downstream increase in intraspecific genetic diversity (see Table 1). These local variables are
hypothesized to reduce or increase the effective population sizes (Ne), consequently affecting the
rate of genetic drift and genetic diversity of populations (Figure 1). The connectivity variables we
used represent hypothesis of isolation by distance (IBD), isolation by resistance (IBR) and
isolation by barrier (IBB). The IBR models include resistance offered by river type (cost of white,
black or clear waters for each species movement), by climatically unsuitable habitats (current and
historical) and by slope (Table 1). These connectivity variables are hypothesized to restrict the
dispersal and mating patterns among localities, reducing the gene flow and increasing the genetic
differentiation between populations (Figure 1). We therefore tested the hypotheses that 1)
connectivity factors that reduce gene flow are related to genetic differentiation for P.
erythrocephala, which has lower dispersal ability, but not for P. sextuberculata (higher dispersal
capacity); and 2) local factors are related to intraspecific genetic diversity of both species, but
leave a stronger effect on the diversity of the low-dispersal species, P. erythrocephala. By using
this model-based riverscape genetics approach, we can gain insights on broader patterns and
processes taking place at the Amazon basin and make use of recently available macro-scale and
high-resolution variables biologically relevant for freshwater vertebrates. In addition to being the
first landscape/riverscape genetics approach with tropical freshwater turtles, our study is the first
with an Amazon aquatic vertebrate that takes in consideration explicit resistance models to test
for IBR.
METHODS
Study region and genetic sampling
We used samples from 14 localities for P. erythrocephala and 20 localities for P. sextuberculata
(Figures 2 and 3; Appendix A in Supporting Information), covering a large portion of their
respective geographic distributions. Since our genetic sampling covers a large portion of the
species’ distributions and the predictor variables represent potential historical (rather than
contemporary) effects on the populations, we used the mtDNA control region (CR) as molecular
marker for both species. We choose this marker because of the high polymorphism in the CR
16
reported for Podocnemis species (Pearse et al. 2006, Santos et al. 2016, Viana et al. under
review). For P. erythrocephala we had a total of 273 sequences (503 bp), from which 246 were
sequenced by Santos et al. (2016, GenBank KY702009–KY702254). Following the same
laboratory procedures as Santos et al. (2016) we extracted, amplified and sequenced the CR for
additional 27 samples related to four new localities (5-BCL, 8-JAU, 13-JUR and 14-PAR; Figure
2; Appendix A; GenBank KY713319–KY713345). For P. sextuberculata we had a total of 336
sequences (605 bp), from which 319 were sequenced in a recent study by Viana et al. (under
review, GenBank KY702255–KY702573). We extracted, amplified and sequenced the CR of
additional 17 samples from three new localities (6-IPX, 10-PPP and 12-CAP; Figure 3; Appendix
A; GenBank KY713302–KY713318), following the same laboratory methods of Viana et al.
(under review).
Genetic metrics
We implemented descriptive genetic analyses in our data set to assess genetic structure by
constructing a haplotype network, performing an AMOVA and a Bayesian analysis of population
admixture. The details and results of these descriptive analyses are available in Appendix B.
We calculated for each sampled locality two intraspecific genetic diversity indices,
haplotypic diversity (Hd, Nei 1987) and nucleotide diversity (π, Nei 1987), in DNASP v.5.10.1
(Librado and Rozas 2009). We also estimated pairwise φST between sampling sites using the
software ARLEQUIN v. 3.5.2.2 (Excoffier and Lischer 2010), testing for significance by
randomization with 1,000 permutations. We used the diversity metrics as response variable for
node-level analysis and the pairwise φST for the link-level analysis.
Landscape data
We collected several landscape metrics for each analytical approach (nodes and links) in order to
represent non-mutually exclusive hypothesis that may explain diversity and differentiation
patterns for the species. We describe the hypotheses and mechanisms linking the local (nodes)
17
and connectivity (links) factors to the expected effects on, respectively, diversity and
differentiation indices of populations on Table 1.
Node-level local variables
For energy availability hypothesis, we used the mean Net Primary Productivity (NPP) from 2000
to 2015 (NASA 2016) to represent the availability of food (fruits, seeds and leaves) for turtles in
each locality sampled.
To represent the environmental stability hypotheses, we used Ecological Niche Modeling
(ENM) to predict the climatic suitability for each species. We used the maximum entropy
machine-learning algorithm, MAXENT, implemented in the R package dismo (Hijmans et al.
2015) to construct the models. The projection to present conditions was the variable for current
environmental stability. In addition, to enable a continuous view of historical climatic suitability,
we projected the models to 62 climatic reconstructions covering the last 120 kyr at small time
intervals (1 to 4 kyr) using the Hadley Centre Climate model (HadCM3; Singarayer and Valdes
2010, Fuchs et al. 2013, Carnaval et al. 2014). We calculated the mean value of suitability for the
62 layers of time and used the resulting mean raster layer as the variable of historical
environmental stability.
To represent high variability of extreme water levels, which can potentially decrease
predictability of available nesting beaches annually (Bermudez-Romero et al. 2015) or cause
nests flooding (Pantoja-Lima et al. 2009), we used two raster maps created by Silva-Junior
(2015) representing extremes of river flows. The rasters was generated from the coefficient of
variation of high (CVmax) and low river flows (CVmin) for 5 thousand points in Amazon basin
for the period of 1998 to 2009.
To represent the subsistence consumption of turtles by rural/riverine human villages, we
generated a kernel-density map of human villages occurring along the distribution of samples for
both species. We used the kernel values of each sampling locality as a surrogate for subsistence
hunting pressure. In addition, because urban centers are the final destination for illegally caught
18
turtles (Fachín-Terán et al. 2004), we measured for each sampling locality the distance (by river
way) to the closest urban center as a mean to characterize illegal commercial hunting.
Finally, to assess if there is a pattern of Downstream Increase in Intraspecific Genetic
Diversity (DIGD), we defined the mouth of Amazon River as the ultimate downstream point and
extracted for each locality the distance by river way to the Amazon River mouth.
Link-level connectivity variables
To test the hypothesis of isolation by distance (IBD) we measured the river distance between
localities using the R package gdistance (van Etten 2012).
For the links analytical level, we used resistance models, a novel approach for riverscape
genetics that can increase our understanding of gene flow patterns as it tests specifically for
migration complexity and resistance between populations. The least-cost paths (LCPs) are
calculated by searching for the path that minimizes the total cumulative cost (or resistance)
between two points (Wang et al. 2009). A difference on the approach within a riverscape genetics
framework (compared to terrestrial habitats) is that, for species using exclusively river ways to
move – the case for the species here studied –, the only path possible is the river path. Therefore,
the LCPs between two localities will always be the same regardless of the variable under
consideration. However, the cost values of each pixel (and therefore the accumulated-cost of
LCP) will be distinct for different variables. To characterize isolation by resistance (IBR) we
used slope, river types (colors) and climatic suitability (Table 1).
We calculated LCPs of average upstream slope (Domisch et al. 2015) between localities
as a surrogate for the presence of topographic barriers (e.g., rapids or waterfalls) or increased
topographic resistance to turtles’ movement.
The rivers in Amazon basin are classified in three types (black, white and clear waters)
based on different origins and physical and chemical properties of their waters (Sioli 1984). Since
there is a lack of biological data on movement preference related to water types for turtles, we
used expert parameterization of resistance values (Zeller et al. 2012). We sent a questionnaire
(Appendix C) to six Amazon turtle experts, asking them to assign different costs to each water
19
type representing how costly they are to the movement of each species. The cost values would
range from 1 (low or no cost to animal movement) to 5 (high cost or barrier to movement).
Because the responses varied among experts (Appendix C), we used the mean cost value of their
opinions to calculate the LCPs between localities.ir
To assess resistance to movement offered from present and past climatic unsuitable
habitats for turtles we clipped the historical and current ENM maps generated (see node-level
local variables section) to the river courses. We used the reverse of suitability values (1 –
suitability) to assign resistances to each pixel in the river network for each species, because
places with lower suitability should represent higher resistance to the movement (Wang et al.
2013). The resulting raster maps have resistance values ranging from 0 (no resistance to species
movement) to 1 (complete resistance to species movement). We then calculated LCPs between
localities for historical and current resistance imposed by unsuitable habitats for each species.
Additionally, the Amazon River was proposed as a potential barrier to the dispersal of P.
erythrocephala (Santos et al. 2016) because the large extension and whitewaters of the Amazon
River may represent a barrier to this species. With samples from two additional locations on the
right-margin of Amazon River, we tested for isolation by barrier (IBB), only for P.
erythrocephala. For this, we attributed binary codes for localities from the same (0) or opposite
(1) sides of Amazon River.
In Appendix D we: describe in further details how we obtained local landscape variables;
show the results for model performance of ENM; and display maps for variables NPP, CVmax,
CVmin, distCITY, villages, distMOUTH, resistance from current and historical suitability, and
river type categories and upstream slope used to generate the LCPs.
Landscape genetic analyses
Because the genetic diversity and differentiation metrics used here can be affected by sampling
sizes (Goodall-Copestake et al. 2012), for both node and link-level analyses we only used
localities for which we had at least 10 individuals sampled (N ≥ 10). This reduced our number of
sites from 14 to 11 for P. erythrocephala and from 20 to 17 for P. sextuberculata.
20
For node-level analysis, we modeled the genetic response variables (Hd and π) in relation
to the predictor landscape variables using generalized linear models (GLMs). To avoid
multicollinearity we only included non-correlated predictor variables in mixed models (r < 0.6;
Appendix E). We also tested for the presence of spatial autocorrelation in the response variables
to ensure the relationships between genetic and landscape are not an artefact of spatial structure
(Wagner and Fortin 2015). We therefore constructed a Moran’s I correlograms but since no
autocorrelation was detected on the response variables or model residuals, we did not built spatial
models (Appendix E). We built GLMs comprising all combinations of one to two predictors
(except when they were collinear). We used a maximum of two predictor covariates per model
because of the limited sample size (N=11 for P. erythrocephala and N=17 for P. sextuberculata).
We also included a null model without predictors to compete with the set of models. To perform
model selection we calculated AIC corrected for small sample sizes (AICc) and Akaike’s weight
of evidence (wAICc) as the relative contribution of models (Burnham and Anderson 2003). We
considered models with ΔAIC (the difference between each model and the best model) ≤ 2 as
equally plausible to explain the observed pattern. To run the AIC-based analyses, we used the R
package AICcmodavg (Mazerolle and Mazerolle 2016).
To assess the importance of each landscape factor in link-level analysis, we controlled for
the geographic distance in the LCPs (IBR models) by dividing the accumulated-costs of LCPs by
the riverway distance among pairs of localities. We believe that by doing so we are representing
in each hypothesis solely the environmental dissimilarity of resistance among localities, despite
longer or shorter geographic distances (see Figure 4). After this control, all correlations between
predictor matrices (IBD, IBR/distance and IBB) were < 0.7 (Appendix F), enabling the test of
non-mutually exclusive hypotheses in multiple regression models (Wagner and Fortin 2015). To
model genetic differentiation (φST) in relation to the geographic and environmental
dissimilarities we applied a generalized dissimilarity modelling (GDM). GDM is a nonlinear
extension of permutational matrix regression that models pairwise biological (in this case
genetic) dissimilarity between sites (Ferrier et al. 2007). GDM accounts for the two types of non-
linearity often encountered in ecological modelling of biological traits: (1) since φST values are
21
scaled between 0–1, the population divergence cannot extend beyond φST = 1, even if habitat
differentiation or geographic distance keep increasing; (2) the rate of change in response
variables along environmental gradients is often not constant (Thomassen et al. 2010). The two
main advantages of using GDM in a landscape genetics approach are its particular suitability for
genetic data (pairwise differentiation) and the possibility of using resistance/LCP models along
with true measures of geographic distances (Thomassen et al. 2010). We therefore applied GDM,
including five predictor variables for P. sextuberculata and six for P. erythrocephala (Table 3),
using the R package gdm (Manion et al. 2014). We assessed the relationship among φST and
each predictor by examining the response curves generated for variables for which I-spline basis
functions could be calculated (i.e., presented non-zero coefficients). In these response curves, the
maximum height represents the relative importance of the variables retained in the model and the
slopes indicate the rate of change in the response variable along the environmental gradient
concerned (Ferrier et al. 2007). We also performed a test of variable importance using an iterative
process that adds and removes the variables to determine the significance by computing the
difference in deviance explained by a model with and a model without the variable concerned
(Fitzpatrick et al. 2013). Although model selection would be the best approach to compare node
and link-level analyses, because the residuals of matrix regressions are not independent of each
other, information-theoretic indices commonly used for model selection (AIC, AICc, or BIC) are
not applicable to distance matrices (Wagner and Fortin 2015). In fact, model selection on link-
level analyses remains to be developed, thus currently the influential observations are best
identified with leave-one-out jackknife methods (Wagner and Fortin 2015), as the one
implemented here.
RESULTS
Genetic metrics
Detailed results for descriptive genetic analyses are available in Appendix B. We recovered
overall moderate haplotype diversity for both species (P. erythrocephala: Hd = 0.627; P.
sextuberculata: Hd = 0.776) and high nucleotide diversity (Appendix A) compared to other
22
studies using control region of mtDNA for Podocnemis species (see Pearse et al. 2006), 0.00234
for P. erythrocephala and 0.00458 for P. sextuberculata. The localities were significantly
differentiated for both species, with pairwise φST between localities ranging from 0 to 0.898 for
P. erythrocephala and from 0 to 0.937 for P. sextuberculata (see tables in Appendix B).
Landscape genetic analyses
Node-level analysis
For P. erythrocephala, two of our competing models explained haplotype and nucleotide
diversity (Table 2), but with the null model being equally plausible to explain the observed
patterns (ΔAICc < 2). Although not differentiated from the null model, the two best models
explaining the genetic diversity of P. erythrocephala are the distance to the nearest urban center
(distCITY) and a combined effect of this distance and the coefficient of variation of high river
flow (CVmax). The relationships among these predictor variables and the response variables are
according to our expectations: increased genetic diversity on localities farther from cities
(positive relationship with distCITY) and on localities with lower variability in maximum flows
(negative relationship with CVmax). For P. sextuberculata, six competing models explained the
two diversity metrics (Table 2): the site productivity (NPP) alone, the distance to Amazon River
mouth (distMOUTH) alone, and the combined effects of each of these variables with density of
rural human communities (NPP+villages and distMOUTH+villages) and historical climatic
suitability (NPP+suit_past and distMOUTH+suit_past). The two most important variables, NPP
and distMOUTH, are highly correlated (r = 0.93; p < 0.001), being difficult to determine which
of the two are influencing genetic diversity. In addition, relationships between distMOUTH,
villages and suit_past with genetic diversity are opposed to the expected: increased genetic
diversity on upstream localities (positive relationship with distMOUTH; Figure 6), on localities
near higher density of human settlements (positive relationship with villages; Figure 6), and on
localities with lower climatic suitability (negative relationship with suit_past; Figure 6). The
relationship for NPP was as expected: higher genetic diversity at more productive sites (higher
NPP; Figure 6). The cumulative contribution (wAICc) of the models to the observed pattern were
23
moderate, 0.53 for Hd and 0.52 for π (Table 2). Full tables of AIC models are available at
Appendix E.
Link-level analysis
The full GDM model explained 20.44% of the deviance in φST turnover for P. erythrocephala
and derived I-spline basis functions for four out of the six variables (Table 3; see Appendix G for
GDM response curves). Summing the coefficients of I-spline basis functions as a measure of
relative variable importance (i.e., the height of each curve; Fitzpatrick and Keller 2015), the main
predictor for genetic differentiation of P. erythrocephala was the Amazon River (0.387),
followed by resistance from current climatic suitability (0.189), resistance from historical
climatic suitability (0.136) and riverway distance (0.110) (Appendix G). For P. sextuberculata
the full GDM model explained only 6.49% of the deviance in φST turnover and derived I-splines
for three out of five variables (Table 3; Appendix G). The most important variable to predict
genetic differentiation of P. sextuberculata was the resistance from river color (0.953), followed
by resistance from current climatic suitability (0.226) and riverway distance (0.187) (Appendix
G).
Although the response curves and I-splines coefficients can elucidate the most important
variables to φST turnover, we detected no significance for models or variables in terms of
variable importance testing by permutations (P. erythrocephala: Full model-2, p = 0.12; P.
sextuberculata: Full model-2, p = 0.11; Table 3). The correction of IBR models by geographic
distance allowed us to disentangle the effects of riverway distance and environmental resistance,
being IBD only retained as a potential predictor after this correction (results not shown).
DISCUSSION
Here we investigated which local and connectivity factors from the Amazon basin riverscape
influenced genetic diversity and differentiation of two Amazonian River turtle species with
different dispersal abilities. We found a relationship between genetic patterns and biologically
meaningful local variables, relevant in the Amazonia context, such as hunting pressure,
24
productivity, and river flow variation. We also found, but in a minor extent, influence of
connectivity variables on genetic differentiation of the turtles: barrier, riverway distance, and
resistance offered by river types and climatically unsuitable habitats. Opposed to our initial
expectations, our results show a stronger influence of local factors on the intrapopulational
genetic diversity of the high-dispersal species, Podocnemis sextuberculata, than on the genetic
diversity of P. erythrocephala (lower dispersal potential). Although in general connectivity
factors are less important in shaping genetic structure of both species, connectivity factors as
expected explain a higher percentage of genetic differentiation of P. erythrocephala than of P.
sextuberculata. Our results therefore demonstrate the importance of assessing the effects of local
variables in riverscape genetics studies, even when dealing with high-dispersal species without
apparent discrete genetic structure.
Influence of local factors on genetic diversity
For both species, the best-fit models associated to genetic diversity patterns contained the proxies
for hunting pressure. For P. erythrocephala, although not differentiated from the null model, the
distance to nearest urban center (distCITY) was included in the best model along with variability
of high river flow (CVmax), and ranked alone as second best model to explain nucleotide
diversity. This may be an evidence of the potential impacts of illegal commercial hunting, or
urban centers per se, on turtles. Illegal harvest by professional fishermen is common and
widespread, being characterized by removal of large quantities of adult turtles to be sold in urban
centers (Pantoja-Lima et al. 2014). For example, a study in the Xingu River found a decrease in
abundance and density of P. unifilis with increasing proximity to urban centers (Alcântara et al.
2013). Human rural communities may also pose a threat to Podocnemis species if exploitation
occurs in an unsustainable manner, causing population declines (Conway-Gómez 2007,
Bernardes et al. 2014) and ultimately affecting genetic patterns (Allendorf et al. 2008). For P.
sextuberculata, the density of rural human settlements (villages) was included in the second best
models for nucleotide and haplotype diversity, combined with primary productivity (NPP) and
distance from the Amazon River mouth (distMOUTH), respectively. However, the relationship is
25
opposed to the predicted, as the intraspecific genetic diversity was higher in places with higher
density of human communities. This is unexpected given the historical use of turtles since the
18th century and high rates of consumption of P. sextuberculata by villagers reported along the
Amazon basin (Smith 1979, Fachín-Terán et al. 2004, Kemenes and Pezzuti 2007, Pantoja-Lima
et al. 2014). The case may be that density of villages per se do not represent turtle consumption,
since feeding habits and consumption rates vary among places (Pezzuti et al. 2010). Although
human settlements also often represent habitat loss for species (Turtle Conservation Fund 2002),
the higher genetic diversity of P. sextuberculata where there is more human villages may be a
consequence of human settlements often establishing in productive sites offering protein
resources, where people hunt in the proximities (Peres 2000). The model NPP + villages supports
the hypothesis that human villages may be established in more productive sites, which in turn
harbor larger population sizes of P. sextuberculata, therefore maintaining higher nucleotide
diversity where productivity and number of villages are higher. Yet, we need to be cautious when
interpreting effects of recent events on mtDNA genetic diversity (Wang 2010) because while
population declines due to harvesting in turtles occurs over years, genetic variation is lost over
generations (Marsack and Swanson 2009). Nonetheless, we do recommend localized efforts to
assess current consumption rate in villages in relation to availability/abundance of turtles as
resource, along with a local genetic study, to measure the direct impacts of preference and
harvesting on these species.
The positive influence of NPP on genetic diversity of P. sextuberculata but not in P.
erythrocephala could be due to their different geographical distributions. Because P.
sextuberculata occurs mainly in white water rivers, known to be very productive compared to the
black waters of the Negro River (main occurrence of P. erythrocephala; Sioli 1984) this factor
may be more relevant to the establishment and growth of populations of P. sextuberculata. While
NPP is the most relevant variable explaining nucleotide diversity of P. sextuberculata, the best
model determining haplotype diversity was distance from Amazon River mouth (distMOUTH).
NPP and distMOUTH are highly correlated, and their correlation is probably due to a west-east
gradient of decreasing primary productivity (Malhi et al. 2004) and distance to Amazon River
26
mouth. For distMOUTH we found a downstream decrease in genetic diversity of P.
sextuberculata, as opposed to the expected pattern of Downstream Increase in Genetic Diversity
(DIGD). This reverse pattern may occur because floodplains and wetlands, which serve as
feeding and movement habitat for P. sextuberculata (Fachin-Terán and Vogt 2014), are more
abundant in western compared to the eastern portion of Amazon basin (Junk et al. 2011). Also,
upstream sites (i.e., mostly western localities in our sampling) are less affected by deforestation,
urbanization, and other anthropogenic alterations of habitats widespread on eastern localities
closer to Amazon River mouth (Laurance et al. 2001). Hence, these conditions, along with
productivity of upstream sites, could harbor larger effective population sizes and larger genetic
diversity in P. sextuberculata across the basin. DIGD is often modelled in dendritic-like river
systems and is more widespread across species with exclusive aquatic dispersal (Paz‐Vinas et al.
2015). We believe DIGD does not describe broad genetic patterns for Amazon River turtles
because 1) turtles also use land environments to nest and bask; 2) Amazon basin is not a true
dendritic network (as large portions are floodplains and wetlands); 3) the scale of Amazon basin
is too coarse to capture the processes underlying the pattern. Further investigation within a sub-
basin (engaging intensive sampling) is necessary to determine whether processes of downstream-
biased dispersal, increase in habitat availability downstream, and upstream-directed colonization
generate a pattern of DIGD for turtles at smaller spatial scales.
The best model for nucleotide diversity of P. erythrocephala, although not differentiated
from the null model, included variability of inter-annual highest water levels (CVmax) in
combination with distance to the nearest urban center (distCITY). Under this model, populations
of P. erythrocephala which are further from urban centers (higher distCITY) and in places with
more stable high river flows (lower CVmax) would maintain larger effective population sizes and
be less affected by genetic drift. Since extremes of flood pulse in the Amazon basin influence
composition of fish assemblages (Sousa and Freitas 2008, Correia et al. 2015, Röpke et al. 2017)
and the dynamics of flooded/non-flooded areas (Junk 1997), it probably also affects populations
of species directly depending on these factors. During falling water periods, while P.
sextuberculata is known to select higher spots in large sandy beaches to nest (Pezzuti and Vogt
27
1999), P. erythrocephala nests more distant from river margins, in a variety of vegetated
substrates such as campinas and savannahs (Batistella and Vogt 2008). Although both species
seem to adopt strategies to minimize flooding of nests by rising water levels, this is still the most
important natural cause of nest losses for these species (Pezzuti and Vogt 1999, Pantoja-Lima et
al. 2009, Carvalho Jr et al. 2011), being particularly detrimental for P. erythrocephala nests
(Castaño-Mora et al. 2003, Batistella and Vogt 2008). Besides this, variability on flooding
extremes was not related to genetic patterns of P. sextuberculata but was included in the best
model for nucleotide diversity of P. erythrocephala. Therefore, it may be that P. erythrocephala
is more susceptible to nests flooding or have recruitment and population sizes more affected by
these nest losses than P. sextuberculata. However, this relationship is tenuous, and an
investigation of genetic diversity related to variability of flooding extremes across larger
temporal scales would benefit this subject.
Opposed to our expectations, there is more influence of local variables on the genetic
diversity of the high dispersal species, P. sextuberculata, than on genetic diversity of P.
erythrocephala (lower dispersal ability). However, several life-history traits other than dispersal
also influence intraspecific genetic diversity, among which generation time and habitat
specialization (Ellegren and Galtier 2016). We cannot precise whether generation times would
influence the genetic-landscape relationships, once this information is unknown for Podocnemis
species. Considering that both species have similar sizes, we assume they also have similar
generation times and mutation rates (Martin and Palumbi 1993) and thus we believe this trait
does not justify the different responses between species. Regarding specialization, although P.
sextuberculata disperses large distances to nest in sandy beaches, its nests are only found in high
points of sandy beaches (Vogt 2008). P. erythrocephala, on the other hand, nests in a wider
variety of substrates, including sandy beaches, but also shrub lands (known as campinas and
campinaranas, (Vogt 2008) and savannas (Carvalho Jr et al. 2011). This wider variety of nesting
substrates may therefore reduce the influence of local variables on recruitment and population
sizes of P. erythrocephala and counterbalance its low-dispersal ability.
28
Influence of connectivity factors on genetic differentiation
The GDMs revealed that 20% of φST turnover in P. erythrocephala is explained by connectivity
variables, while for P. sextuberculata this value is only 6%. The percent deviance explained is
used as a measure of model fit in GDM, and compared to other studies using genetic data, our
results indicate an overall poor model fit (27-88% for AFLP-neutral and mtDNA Freedman et al.
2010, e.g., 24-63% for SNPs Fitzpatrick and Keller 2015, 57% for AFLP markers Shryock et al.
2016). In addition, the variable importance permutation test did not recover significance for any
variable or model. This is an indicative that connectivity variables are less important in shaping
genetic patterns of Amazon river turtles than local variables. Nevertheless, the height of spline
curves can still serve as a measure of which are the most important variables influencing genetic
differentiation in these species. Our GDM analyses fitted I-spline functions for four connectivity
variables for P. erythrocephala and three for P. sextuberculata. The asymptotic shapes of these
curves demonstrate the usefulness of GDM to model non-linear relationships commonly found in
link-level landscape genetics analyses (Spear et al. 2015).
Amazon River was the most important variable explaining genetic differentiation of P.
erythrocephala. The role of Amazon River as a potential barrier to dispersal of P. erythrocephala
was suggested in a previous population genetics study (Santos et al. 2016). By adding samples
from two localities on the right margin of Amazon River to their dataset and employing a
riverscape genetics approach, we corroborate the idea that the river is the most important
predictor of genetic differentiation, at least among the set of variables here tested. Surprisingly,
resistance offered by different river types was not important to explain the genetic differentiation
of P. erythrocephala, which is restricted to black and clear waters. Therefore, we cannot precise
whether Amazon River works as a barrier due to its large width (Hayes and Sewlal 2004), its
whitewaters per se (Beheregaray et al. 2015) or a historical process of the river dynamics. On the
other hand, resistance from river type was the most important variable for P. sextuberculata,
which can be found in all three types of water. This suggests that populations of P.
sextuberculata are increasingly divergent along paths containing costlier water types (for this
species, costlier waters are black > clear > white; Appendix C), despite total distance to be
29
travelled among sites. However, due to low percent of variation explained by GDM (6%) and
lack of significance, this pattern remains only as a suggestion of future investigation for the
species.
Even when species have different dispersal abilities and distinct spatial genetic structure,
it is possible to detect influence from common waterscape features (Liggins et al. 2016). Here we
detected a minor but congruent influence of two riverscape features on genetic differentiation of
both species: current resistance to unsuitable climates and riverway distance. For both species the
IBR models explained more of the genetic differentiation turnover than IBD. Riverway distance
has often a minor or no role in explaining the genetic differentiation of aquatic vertebrates in
Amazon basin, potentially because of high connectivity offered by flooded habitats (Cantanhede
et al. 2005, Hrbek et al. 2005, Pearse et al. 2006, Santos et al. 2016). Our results emphasize the
utility of adding resistance-based models to classical IBD and IBB models when studying
riverscape genetics. We also reinforce the usefulness of expert’s opinion to parameterize LCPs in
systems for which empirical resistance evidence is lacking (Zeller et al. 2012). In addition,
dividing the cost-weighted distances of each variable by riverway distance allowed us to assess
the accumulative cost of traversing costly environments despite the total distance to be travelled.
This control by distance allowed us to disentangle IBR models from IBD and test whether
distance by itself or resistance by itself increased genetic differentiation on these species. We
suggest this approach when dealing with species that move exclusively through linear habitats
(i.e., rivers), for which there is only one path possible between populations, but environmental
dissimilarity may be more determinant to dispersal than distance.
Although we recovered broad unsuitable conditions for both species in the past compared
to present conditions (Figure 5), in both cases the resistance offered by current unsuitable habitats
explained more differentiation turnover than historical. It is necessary to be cautious when
interpreting patterns of current landscape connectivity with genetic data because of the risk of
underestimating current genetic connectivity (Samarasin et al. 2016). However, suitability values
for both species did not vary much from present (0 kya) to the past 12,000 years (12 kya; data not
shown). We argue that the relationship of current suitability with genetic differentiation, if real,
30
established from gene flow levels occurring in climatic conditions from the past 12,000 years and
not necessarily from present (0 time). Another interesting outcome is that climatic suitability does
not influence local genetic diversity of P. erythrocephala and is included in the least important
model explaining genetic diversity of P. sextuberculata, while the resistance offered by
unsuitable climate along connectivity paths does have a minor role in explaining genetic
differentiation patterns from both species. By using ecological niche modeling, Ortego et al.
(2015) similarly found that genetic diversity was not associated with habitat suitability or
stability, while dispersal routes defined by stability of suitable habitats was the primary driver of
genetic differentiation in canyon live oak populations. Nonetheless, the weak (and lack of)
relationship with genetic diversity and lack of significance in the test of variable importance for
genetic differentiation seen here corroborate the idea that environmental stability (i.e., climatic
suitability) is overall less important in structuring genetic variation in aquatic organisms (Thomaz
et al. 2015) than in terrestrial species (Carnaval et al. 2009, Ortego et al. 2015). Thomaz et al.
(2015) found that palaeodrainages influence the genetic patterns of a freshwater fish dependent
upon forest habitat, while habitat stability (as measured by climatic suitability) do not. Similarly,
our results highlight that other aspects of the riverscape are more important to both genetic
diversity and genetic differentiation patterns for river turtles.
Further considerations
The lack of significance in relationships among connectivity variables and φST may be
associated to three potential drawbacks in this study. First, mitochondrial DNA (mtDNA) used
here has lower mutation rate in comparison to microsatellite and SNPs data usually employed in
landscape genetic studies (Storfer et al. 2010, Wang 2010). However, mtDNA is particularly well
suited to questions investigating the historical effects of landscape factors across broad spatial
scales (Anderson et al. 2010, Bowen et al. 2014, Mitchell et al. 2015, Thomaz et al. 2015, Liggins
et al. 2016), as our approach here. Also, a population genetics study with Podocnemis expansa
found positive correlations among microsatellite and mtDNA diversity indices within
populations, discussing that observed genetic patterns from both markers potentially result from
31
common historical demographic effects (Pearse et al. 2006). We believe that the genetic variation
in mtDNA control region across the basin recovered here is adequate to make population
comparisons. Nevertheless, microsatellite and SNP markers should be more variable (Whittaker
et al. 2003, Morin et al. 2004), and remain to be developed and tested for P. erythrocephala and
P. sextuberculata. Second, the temporal mismatch between landscape effects and genetic
responses is a general critic to landscape genetic studies (Epps and Keyghobadi 2015), especially
when using a historical marker to assess contemporary landscape changes (Anderson et al. 2010).
Here we employed connectivity variables that likely represent the configuration of the landscape
across several past decades (current suitability) and millennia (riverway distances, slope, river
types, historical suitability and Amazon River). Accordingly, we believe their effect is historical
and could have been reinforcing potential gene flow restrictions until recent times. Therefore, the
lack of significance was probably not an artefact of temporal mismatch between our connectivity
variables and genetic differentiation measured through mtDNA. Third, turtles, as long-lived
organisms with delayed maturation time, are expected to have longer time to manifest changes in
genetic patterns (Kuo and Janzen 2004, Marsack and Swanson 2009, Ortego et al. 2015). But the
high genetic diversity found in highly hunted Podocnemis species (this work, Pearse et al. 2006,
Escalona et al. 2009) are evidence that turtles longevity, along with overlapping generations and
multiple paternity (Valenzuela 2000, Fantin et al. 2008, Fantin et al. 2010, Fantin et al. 2015)
may be buffering potential bottlenecks from past centuries. In general, measures of genetic
diversity approach equilibrium more slowly than genetic differentiation metrics (Varvio et al.
1986). Yet, it does not mean that signals of landscape change always take longer time to emerge
in measures of genetic diversity compared to genetic differentiation (DiLeo and Wagner 2016),
as illustrated by significant associations among local factors and genetic diversity but not
between genetic differentiation and connectivity recovered here. In addition, studies with turtles
and other long-lived organisms detected effects of landscape factors on genetic divergence within
few generations (Epps et al. 2005, Moore et al. 2008, Reid et al. 2017). Thus, lack of relationship
between genetic differentiation and connectivity factors may be an evidence of high migration,
but also a consequence of several other factors affecting the rate at which neutral genetic
32
differentiation reach equilibrium, such as effective population sizes and population dynamics
(Epps and Keyghobadi 2015).
CONCLUSIONS AND PERSPECTIVES
Overall, our study shows that local variables can be important factors determining genetic
diversity patterns of river turtles, despite major attention often given to connectivity variables.
We show that for Podocnemis sextuberculata, a high-dispersal species, genetic diversity was
associated to higher primary productivity, density of human villages and distance from Amazon
River mouth, and to lower historical climatic suitability. Besides a lack of strong relationship
among genetic differentiation and connectivity variables, there was a tendency of higher
influence from connectivity factors on genetic differentiation of the low-dispersal species, P.
erythrocephala. In addition, the models explaining more genetic differentiation turnover were
isolation by barrier (Amazon River) for P. erythrocephala and isolation by resistance offered by
costly river types for P. sextuberculata. We assessed variables biologically relevant for other
Amazonian riverine species in a basin-wide context and hope this work can stimulate research on
which riverscape factors in Amazon basin are potential drivers of genetic patterns for aquatic
vertebrates. Our study is the first to engage empirical model-based riverscape genetics in Amazon
basin and to develop resistance models in a riverscape genetics context. Therefore, it should
provide a framework to investigate spatial genetic patterns of other high-dispersal riverine
species in drainage systems. Finally, by examining broad patterns, we identified potential factors
that could receive a deeper local investigation in future studies.
ACKNOWLEDGEMENTS
We thank M. N. S. Viana for contributing with the additional biological samples used in this
work. We also thank P. C. A. Machado, R. C. Vogt, J. Erickson and F. Fernandes for collecting
samples. JAO received her Master's fellowship from National Council for Scientific and
Technological Development (CNPq). FPW thanks financial support from CNPq (475559/2013-
4), Fundação de Amparo à Pesquisa do Amazonas-FAPEAM (062.00665/2015) and Partnerships
33
for Enhanced Engagement in Research from the U.S. National Academy of Sciences and U.S.
Agency of International Development (PEER NAS/USAID PGA-2000005316). GCC thanks
CNPq grant (302297/2015-4). IPF thanks for financial support from CNPq-SISBIOTA grant
(563348/2010-0).
REFERENCES
Alcântara, A. S. et al. 2013. Effects of the hydrological cycle and human settlements on the
population status of Podocnemis unifilis (Testudines: Podocnemididae) in the Xingu
River, Brazil. — Chelonian Conservation and Biology 12: 134-142.
Allendorf, F. W. et al. 2008. Genetic effects of harvest on wild animal populations. — Trends in
Ecology & Evolution 23: 327-337.
Anderson, C. D. et al. 2010. Considering spatial and temporal scale in landscape‐genetic studies
of gene flow. — Molecular Ecology 19: 3565-3575.
Balkenhol, N. et al. 2016. Introduction to Landscape Genetics–Concepts, Methods, Applications.
— Landscape Genetics: Concepts, Methods, Applications 1-8.
Batistella, A. M. and Vogt, R. C. 2008. Nesting ecology of Podocnemis erythrocephala
(Testudines, Podocnemididae) of the Rio Negro, Amazonas, Brazil. — Chelonian
Conservation and Biology 7: 12-20.
Beheregaray, L. B. et al. 2015. Ecological speciation in the tropics: Insights from comparative
genetic studies in Amazonia. — Frontiers in Genetics 5: 1-19.
Bermudez-Romero, A. L. et al. 2015. Nesting habitat of the "cupiso" Podocnemis sextuberculata
(Testudines: Podocnemididae) in Erepecu lake (Pará - Brazil). — Acta Biológica
Colombiana 20: 183-191.
Bernardes, V. C. D. et al. 2014. Abundance and population structure of Podocnemis
erythrocephala (Testudines, Podocnemididae) in the Unini River, Amazonas. —
Chelonian Conservation and Biology 13: 89-95.
Bour, R. 2008. Global diversity of turtles (Chelonii; Reptilia) in freshwater. Freshwater animal
diversity assessment. Springer, pp. 593-598.
34
Bowen, B. W. et al. 2014. Phylogeography unplugged: comparative surveys in the genomic era.
— Bulletin of Marine Science 90: 13-46.
Burnham, K. P. and Anderson, D. R. 2003. Model selection and multimodel inference: a practical
information-theoretic approach. — Springer Science & Business Media.
Cantanhede, A. et al. 2005. Phylogeography and population genetics of the endangered
Amazonian manatee, Trichechus inunguis Natterer, 1883 (Mammalia, Sirenia). —
Molecular Ecology 14: 401-413.
Carnaval, A. C. et al. 2009. Stability predicts genetic diversity in the Brazilian Atlantic forest
hotspot. — Science 323: 785-789.
Carnaval, A. C. et al. 2014. Prediction of phylogeographic endemism in an environmentally
complex biome. — Proceedings of the Royal Society of London B: Biological Sciences
281: 20141461.
Carvalho Jr, E. A. et al. 2011. Podocnemis erythrocephala Nests in the Lower Tapajós River,
Central Amazonia, Brazil. — Chelonian Conservation and Biology 10: 146-148.
Castaño-Mora, O. et al. 2003. Reproductive ecology of Podocnemis erythrocephala (Testudines:
Podocnemididae) in the Lower Inírida River, Colombia. — Chelonian Conservation and
Biology 4: 664-670.
Conway-Gómez, K. 2007. Effects of human settlements on abundance of Podocnemis unifilis and
P. expansa turtles in northeastern Bolivia. — Chelonian Conservation and Biology 6:
199-205.
Cook, B. D. et al. 2011. Landscape genetic analysis of the tropical freshwater fish Mogurnda
mogurnda (Eleotridae) in a monsoonal river basin: importance of hydrographic factors
and population history. — Freshwater Biology 56: 812-827.
Correia, G. B. et al. 2015. Intra-and inter-annual changes in the condition factors of three
Curimatidae detritivores from Amazonian floodplain lakes. — Biota Neotropica 15: 1-7.
Cushman, S. A. et al. 2006. Gene flow in complex landscapes: testing multiple hypotheses with
causal modeling. — The American Naturalist 168: 486-499.
35
De Thoisy, B. et al. 2006. Genetic structure, population dynamics, and conservation of Black
caiman (Melanosuchus niger). — Biological Conservation 133: 474-482.
Deiner, K. et al. 2007. Population structure and genetic diversity of trout (Oncorhynchus mykiss)
above and below natural and man-made barriers in the Russian River, California. —
Conservation Genetics 8: 437-454.
DiLeo, M. F. and Wagner, H. H. 2016. A Landscape Ecologist’s Agenda for Landscape Genetics.
— Current Landscape Ecology Reports 1: 115-126.
Domisch, S. et al. 2015. Near-global freshwater-specific environmental variables for biodiversity
analyses in 1 km resolution. — Scientific data 2: 1-13.
Ellegren, H. and Galtier, N. 2016. Determinants of genetic diversity. — Nature Reviews Genetics
17: 422-433.
Epps, C. W. and Keyghobadi, N. 2015. Landscape genetics in a changing world: disentangling
historical and contemporary influences and inferring change. — Molecular Ecology 24:
6021-6040.
Epps, C. W. et al. 2005. Highways block gene flow and cause a rapid decline in genetic diversity
of desert bighorn sheep. — Ecology Letters 8: 1029-1038.
Ernst, C. H. E. and Barbour, R. W. 1989. Turtles of the World. — Smithsonian Institution Press.
Escalona, T. et al. 2009. Population genetics of the endangered South American freshwater turtle,
Podocnemis unifilis, inferred from microsatellite DNA data. — Conservation Genetics 10:
1683-1696.
Excoffier, L. and Lischer, H. E. 2010. Arlequin suite ver 3.5: a new series of programs to perform
population genetics analyses under Linux and Windows. — Molecular Ecology Resources
10: 564-567.
Fachin-Terán, A. and Vogt, R. C. 2014. Alimentação de Podocnemis sextuberculata (Testudines:
Podocnemididae) na Reserva Mamirauá, Amazonas, Brasil. — Revista Colombiana de
Ciencia Animal 6: 286-309.
36
Fachín-Terán, A. et al. 2003. Estrutura populacional, razão sexual e abundância de Podocnemis
sextuberculata (Testudines, Podocnemididae) na reserva de desenvolvimento sustentável
Mamirauá, Amazonas, Brasil. — Phyllomedusa 2: 43-63.
Fachín-Terán, A. et al. 2004. Patterns of Use and Hunting of Turtles in the Mamirauá Sustainable
Development Reserve, Amazonas, Brazil. — In: Silvius, K. M. et al. (eds), People in
Nature: Wildlife Conservation in South and Central America. Columbia University Press,
pp. 362-377.
Fachín-Terán, A. et al. 2006. Seasonal movements of Podocnemis sextuberculata (Testudines:
Podocnemididae) in the Mamiraua sustainable development reserve, Amazonas, Brazil.
— Journal Information 5: 18-24.
Fantin, C. et al. 2010. Polyandry in the red-headed river turtle Podocnemis erythrocephala
(Testudines, Podocnemididae) in the Brazilian Amazon. — Genetics and Molecular
Research 9: 435-440.
Fantin, C. et al. 2015. Evidence of multiple paternal contribution in Podocnemis sextuberculata
(Testudines: Podocnemididae) detected by microsatellite markers. — Phyllomedusa:
Journal of Herpetology 14: 89-97.
Fantin, C. et al. 2008. Polyandry in Podocnemis unifilis (Pleurodira; Podocnemididae), the
vulnerable yellow-spotted Amazon River turtle. — Amphibia-Reptilia 29: 479-486.
Farias, I. P. et al. 2004. Genetic diversity and population structure of Amazonian crocodilians. —
Animal Conservation 7: 265-272.
Farias, I. P. et al. 2010. Are rapids a barrier for floodplain fishes of the Amazon basin? A
demographic study of the keystone floodplain species Colossoma macropomum
(Teleostei: Characiformes). — Molecular Phylogenetics and Evolution 56: 1129-1135.
Ferrier, S. et al. 2007. Using generalized dissimilarity modelling to analyse and predict patterns
of beta diversity in regional biodiversity assessment. — Diversity and Distributions 13:
252-264.
37
Fitzpatrick, M. C. and Keller, S. R. 2015. Ecological genomics meets community‐level modelling
of biodiversity: mapping the genomic landscape of current and future environmental
adaptation. — Ecology Letters 18: 1-16.
Fitzpatrick, M. C. et al. 2013. Environmental and historical imprints on beta diversity: insights
from variation in rates of species turnover along gradients. — Proceedings of the Royal
Society of London B: Biological Sciences 280: 20131201.
Frankham, R. 1996. Relationship of genetic variation to population size in wildlife. —
Conservation Biology 10: 1500-1508.
Freedman, A. H. et al. 2010. Genomic signals of diversification along ecological gradients in a
tropical lizard. — Molecular Ecology 19: 3773-3788.
Fuchs, J. et al. 2013. Extending ecological niche models to the past 120 000 years corroborates
the lack of strong phylogeographic structure in the Crested Drongo (Dicrurus forficatus
forficatus) on Madagascar. — Biological Journal of the Linnean Society 108: 658-676.
Gomez‐Uchida, D. et al. 2009. Interaction of landscape and life history attributes on genetic
diversity, neutral divergence and gene flow in a pristine community of salmonids. —
Molecular Ecology 18: 4854-4869.
Goodall-Copestake, W. et al. 2012. On the comparison of population-level estimates of haplotype
and nucleotide diversity: a case study using the gene cox1 in animals. — Heredity 109:
50-56.
Gravena, W. et al. 2015. Living between rapids: genetic structure and hybridization in botos
(Cetacea: Iniidae: Inia spp.) of the Madeira River, Brazil. — Biological Journal of the
Linnean Society 114: 764-777.
Hagerty, B. E. et al. 2011. Making molehills out of mountains: landscape genetics of the Mojave
desert tortoise. — Landscape Ecology 26: 267-280.
Hand, B. K. et al. 2015. Climate Variables Explain Neutral and Adaptive Variation within
Salmonid Metapopulations: The Importance of Replication in Landscape Genetics. —
Molecular Ecology 25: 689-705.
38
Hayes, F. E. and Sewlal, J. A. N. 2004. The Amazon River as a dispersal barrier to passerine
birds: effects of river width, habitat and taxonomy. — Journal of Biogeography 31: 1809-
1818.
Hijmans, R. J. et al. 2015. dismo: Species distribution modeling. R package version 1.0-12.
Hoogmoed, M. S. and de Avila-Pires, T. C. 1990. New distribution data for Podocnemis
erythrocephala (Spix) with remarks on some other turtle taxa (Reptilia: Chelonia:
Pelomedusidae). — Zoologische Mededelingen 64: 21-24.
Hrbek, T. et al. 2005. Population genetic analysis of Arapaima gigas, one of the largest
freshwater fishes of the Amazon basin: implications for its conservation. — Animal
Conservation 8: 297-308.
Hughes, J. M. et al. 2009. Genes in streams: using DNA to understand the movement of
freshwater fauna and their riverine habitat. — BioScience 59: 573-583.
Iverson, J. 1992. A revised checklist with distribution maps of the turtle of the world.
Jenkins, D. G. et al. 2010. A meta‐analysis of isolation by distance: relic or reference standard for
landscape genetics? — Ecography 33: 315-320.
Junk, W. J. 1997. General aspects of floodplain ecology with special reference to Amazonian
floodplains. The Central Amazon Floodplain. Springer, pp. 3-20.
Junk, W. J. et al. 2011. A classification of major naturally-occurring Amazonian lowland
wetlands. — Wetlands 31: 623-640.
Kanno, Y. et al. 2011. Fine‐scale population structure and riverscape genetics of brook trout
(Salvelinus fontinalis) distributed continuously along headwater channel networks. —
Molecular Ecology 20: 3711-3729.
Kemenes, A. and Pezzuti, J. C. B. 2007. Estimate of trade traffic of Podocnemis (Testudines,
pedocnemididae) from the middle Purus River, Amazonas, Brazil. — Chelonian
Conservation and Biology 6: 259-262.
Kovach, R. P. et al. 2015. Genetic diversity is related to climatic variation and vulnerability in
threatened bull trout. — Global Change Biology 21: 2510-2524.
39
Kuo, C. and Janzen, F. J. 2004. Genetic effects of a persistent bottleneck on a natural population
of ornate box turtles (Terrapene ornata). — Conservation Genetics 5: 425-437.
Latch, E. K. et al. 2011. Fine-scale analysis reveals cryptic landscape genetic structure in desert
tortoises. — PLoS one 6: e27794.
Laurance, W. F. et al. 2001. The future of the Brazilian Amazon. — Science 291: 438-439.
Leclerc, E. et al. 2008. The landscape genetics of yellow perch (Perca flavescens) in a large
fluvial ecosystem. — Molecular Ecology 17: 1702-1717.
Librado, P. and Rozas, J. 2009. DnaSP v5: a software for comprehensive analysis of DNA
polymorphism data. — Bioinformatics 25: 1451-1452.
Liggins, L. et al. 2016. Seascape features, rather than dispersal traits, predict spatial genetic
patterns in co‐distributed reef fishes. — Journal of Biogeography 43: 256-267.
Malhi, Y. et al. 2004. The above‐ground coarse wood productivity of 104 Neotropical forest
plots. — Global Change Biology 10: 563-591.
Manel, S. et al. 2003. Landscape genetics: combining landscape ecology and population genetics.
— Trends in Ecology & Evolution 18: 189-197.
Manion, G. et al. 2014. gdm: Functions for Generalized Dissimilarity Modeling.
Marsack, K. and Swanson, B. J. 2009. A genetic analysis of the impact of generation time and
road-based habitat fragmentation on eastern box turtles (Terrapene c. carolina). —
Copeia 2009: 647-652.
Martin, A. P. and Palumbi, S. R. 1993. Body size, metabolic rate, generation time, and the
molecular clock. — Proceedings of the National Academy of Sciences 90: 4087-4091.
Mazerolle, M. J. and Mazerolle, M. M. J. 2016. Package ‘AICcmodavg’.
McRae, B. H. 2006. Isolation by resistance. — Evolution 60: 1551-1561.
Mitchell, M. W. et al. 2015. Environmental variation and rivers govern the structure of
chimpanzee genetic diversity in a biodiversity hotspot. — BMC Evolutionary Biology 15:
1-13.
Mittermeier, R. and Wilson, R. 1974. Redescription of Podocnemis erythrocephala (Spix, 1824),
an amazonian pelomedusid turtle. — Papéis Avulsos Zoologia, São Paulo 28: 147-162.
40
Mittermeier, R. A. et al. 2015. Podocnemis erythrocephala (Spix 1824) – Red-headed Amazon
River Turtle, Irapuca. — In: Rhodin, A. et al. (eds), Conservation Biology of Freshwater
Turtles and Tortoises: A Compilation Project of the IUCN/SSC Tortoise and Freshwater
Turtle Specialist Group. Chelonian Research Monographs, pp. 087.1-10.
Moore, J. et al. 2008. Fine‐scale genetic structure of a long‐lived reptile reflects recent habitat
modification. — Molecular Ecology 17: 4630-4641.
Morin, P. A. et al. 2004. SNPs in ecology, evolution and conservation. — Trends in Ecology &
Evolution 19: 208-216.
Murphy, M. et al. 2010. Landscape genetics of high mountain frog metapopulations. —
Molecular Ecology 19: 3634-3649.
NASA 2016. Moderate Resolution Imaging Spectroradiometer (MODIS), NASA.
Nei, M. 1987. Molecular evolutionary genetics. — Columbia University Press.
Ortego, J. et al. 2015. Climatically stable landscapes predict patterns of genetic structure and
admixture in the Californian canyon live oak. — Journal of Biogeography 42: 328-338.
Østergaard, S. et al. 2003. Long‐term temporal changes of genetic composition in brown trout
(Salmo trutta L.) populations inhabiting an unstable environment. — Molecular Ecology
12: 3123-3135.
Ouellet‐Cauchon, G. et al. 2014. Landscape variability explains spatial pattern of population
structure of northern pike (Esox lucius) in a large fluvial system. — Ecology and
Evolution 4: 3723-3735.
Ozerov, M. Y. et al. 2012. “Riverscape” genetics: river characteristics influence the genetic
structure and diversity of anadromous and freshwater Atlantic salmon (Salmo salar)
populations in northwest Russia. — Canadian Journal of Fisheries and Aquatic Sciences
69: 1947-1958.
Pantoja-Lima, J. et al. 2014. Chain of commercialization of Podocnemis spp. turtles (Testudines:
Podocnemididae) in the Purus River, Amazon basin, Brazil: current status and
perspectives. — Journal of Ethnobiology and Ethnomedicine 10: 8.
41
Pantoja-Lima, J. et al. 2009. Seleção de locais de desova e sobrevivência de ninhos de quelônios
Podocnemis no baixo Rio Purus, Amazonas, Brasil. — Revista Colombiana de Ciencia
Animal 1: 37-59.
Paz‐Vinas, I. et al. 2015. Evolutionary processes driving spatial patterns of intra‐specific genetic
diversity in river ecosystems. — Molecular Ecology 24: 4586–4604.
Pearse, D. E. et al. 2006. Estimating population structure under nonequilibrium conditions in a
conservation context: continent-wide population genetics of the giant Amazon river turtle,
Podocnemis expansa (Chelonia; Podocnemididae). — Molecular Ecology 15: 985-1006.
Peres, C. A. 2000. Effects of subsistence hunting on vertebrate community structure in
Amazonian forests. — Conservation Biology 14: 240-253.
Pezzuti, J. and Vogt, R. 1999. Nesting ecology of Podocnemis sextuberculata (Testudines,
Pelomedusidae) in the Japurá River, Amazonas, Brazil. — Chelonian Conservation and
Biology 3: 419-424.
Pezzuti, J. et al. 2000. Nesting ecology of pelomedusid turtles in the Purus River, Amazonas,
Brazil. Report of the 80th Annual Meeting of the American Society of Icthyologists and
Herpetologists. La Paz, Universidad Autónoma de Baja California Sur. p 294.
Pezzuti, J. C. et al. 2010. Uses and taboos of turtles and tortoises along Rio Negro, Amazon
Basin. — Journal of Ethnobiology 30: 153-168.
Pritchard, P. 1979. Encyclopedia of Turtles. Neptune, New Jersey: TFH Publ. — Inc. Ltd
Reid, B. N. et al. 2017. Genetic effects of landscape, habitat preference, and demography on three
co‐occurring turtle species. — Molecular Ecology 781-798.
Rhodin, A. G. et al. 2008. Conservation Biology of Freshwater Turtles and Tortoises: A
Compilation Project of the IUCN/SSC Tortoise and Freshwater Turtle Specialist Group.
— Chelonian Research Foundation.
Richardson, J. L. 2012. Divergent landscape effects on population connectivity in two co‐
occurring amphibian species. — Molecular Ecology 21: 4437-4451.
Röpke, C. P. et al. 2017. Simultaneous abrupt shifts in hydrology and fish assemblage structure in
a floodplain lake in the central Amazon. — Scientific Reports 7: 40170.
42
Rueda-Almonacid, J. V. et al. 2007. Las tortugas y los crocodilianos de los paises andinos del
trópico. — In: Internacional, C. (ed), Serie de guias tropicales de campo n. 6.
Samarasin, P. et al. 2016. The problem of estimating recent genetic connectivity in a changing
world. — Conservation Biology.
Santos, R. C. et al. 2016. Testing the effects of barriers on the genetic connectivity in Podocnemis
erythrocephala (Red-headed Amazon River Turtle): implications for management and
conservation. — Chelonian Conservation and Biology 15: 12-22.
Selkoe, K. A. et al. 2015. Waterscape genetics–applications of landscape genetics to rivers, lakes,
and seas. — Landscape genetics: concepts, methods, applications. Hoboken, NJ: John
Wiley & Sons.
Shryock, D. F. et al. 2016. Landscape genetic approaches to guide native plant restoration in the
Mojave Desert. — Ecological Applications 27: 429-445.
Singarayer, J. S. and Valdes, P. J. 2010. High-latitude climate sensitivity to ice-sheet forcing over
the last 120kyr. — Quaternary Science Reviews 29: 43-55.
Sioli, H. 1984. The Amazon: limnology and landscape ecology of a mighty tropical river and its
basin. — Springer Science & Business Media.
Smith, N. J. 1979. Aquatic turtles of Amazonia: an endangered resource. — Biological
Conservation 16: 165-176.
Sork, V. L. and Waits, L. 2010. Contributions of landscape genetics–approaches, insights, and
future potential. — Molecular Ecology 19: 3489-3495.
Sousa, R. G. and Freitas, C. E. d. C. 2008. The influence of flood pulse on fish communities of
floodplain canals in the Middle Solimões River, Brazil. — Neotropical Ichthyology 6:
249-255.
Spear, S. F. et al. 2010. Use of resistance surfaces for landscape genetic studies: considerations
for parameterization and analysis. — Molecular Ecology 19: 3576-3591.
Spear, S. F. et al. 2015. Resistance Surface Modeling in Landscape Genetics. — Landscape
Genetics: Concepts, Methods, Applications 129-148.
43
Spielman, D. et al. 2004. Most species are not driven to extinction before genetic factors impact
them. — Proceedings of the National Academy of Sciences of the United States of
America 101: 15261-15264.
Steele, C. et al. 2009. Influence of life‐history variation on the genetic structure of two sympatric
salamander taxa. — Molecular Ecology 18: 1629-1639.
Storfer, A. 2013. Landscape Genetics. — In: Levin, S. A. (ed), Encyclopedia of Biodiversity
(Second Edition). Academic Press, pp. 508-523.
Storfer, A. et al. 2010. Landscape genetics: where are we now? — Molecular Ecology 19: 3496-
3514.
Thomassen, H. A. et al. 2010. Spatial modelling and landscape‐level approaches for visualizing
intra‐specific variation. — Molecular Ecology 19: 3532-3548.
Thomaz, A. T. et al. 2016. The architecture of river networks can drive the evolutionary
dynamics of aquatic populations. — Evolution 70: 731-739.
Thomaz, A. T. et al. 2015. Testing the effect of palaeodrainages versus habitat stability on
genetic divergence in riverine systems: study of a Neotropical fish of the Brazilian coastal
Atlantic Forest. — Journal of Biogeography 42: 2389-2401.
Turtle Conservation Fund 2002. A global action plan for conservation of tortoises and freshwater
turtles: strategy and funding prospectus 2002-2007. — Conservation International and
Chelonian Research Foundation.
Valenzuela, N. 2000. Multiple paternity in side‐neck turtles Podocnemis expansa: evidence from
microsatellite DNA data. — Molecular Ecology 9: 99-105.
van Etten, J. 2012. gdistance: Distances and routes on geographical grids. R package version 1.1–
4.
Varvio, S.-L. et al. 1986. Genetic variation in subdivided populations and conservation genetics.
— Heredity 57: 189-198.
Viana, M. N. S. et al. under review. Genetic connectivity of the threatened Amazon river turtle
Podocnemis sextuberculata (Testudines, Podocnemididae): implications for conservation.
— Chelonian Conservation and Biology.
44
Vincent, B. et al. 2013. Landscape genomics in Atlantic salmon (Salmo salar): searching for
gene–environment interactions driving local adaptation. — Evolution 67: 3469-3487.
Vogt, R. et al. 1991. Geographic distribution: Podocnemis erythrocephala. — Herpetological
Review 22: 25.
Vogt, R. C. 2008. Amazon turtles. — INPA-Instituto Nacional de Pesquisas da Amazônia.
Wagner, H. H. and Fortin, M.-J. 2013. A conceptual framework for the spatial analysis of
landscape genetic data. — Conservation Genetics 14: 253-261.
Wagner, H. H. and Fortin, M. J. 2015. Basics of spatial data analysis: linking landscape and
genetic data for landscape genetic studies. — Landscape Genetics: Concepts, Methods,
Applications 77-98.
Wang, I. J. 2010. Recognizing the temporal distinctions between landscape genetics and
phylogeography. — Molecular Ecology 19: 2605-2608.
Wang, I. J. and Bradburd, G. S. 2014. Isolation by environment. — Molecular Ecology 23: 5649-
5662.
Wang, I. J. et al. 2013. Quantifying the roles of ecology and geography in spatial genetic
divergence. — Ecology Letters 16: 175-182.
Wang, I. J. et al. 2009. Landscape genetics and least‐cost path analysis reveal unexpected
dispersal routes in the California tiger salamander (Ambystoma californiense). —
Molecular Ecology 18: 1365-1374.
Wang, Y.-H. et al. 2008. Habitat suitability modelling to correlate gene flow with landscape
connectivity. — Landscape Ecology 23: 989-1000.
Whittaker, J. C. et al. 2003. Likelihood-based estimation of microsatellite mutation rates. —
Genetics 164: 781-787.
Wofford, J. E. et al. 2005. Influence of barriers to movement on within-watershed genetic
variation of coastal cutthroat trout. — Ecological Applications 15: 628-637.
Wright, S. 1931. Evolution in Mendelian populations. — Genetics 16: 97-159.
Wright, S. 1943. Isolation by distance. — Genetics 28: 114.
45
Zeller, K. A. et al. 2012. Estimating landscape resistance to movement: a review. — Landscape
Ecology 27: 777-797.
46
TABLES
Table 1. Local and connectivity variables hypothesized to influence genetic diversity (Hd and π) and differentiation (φST) for P.
erythrocephala and P. sextuberculata.
Response
variables Hypotheses Predictor variables Pred. Mechanism References
Genetic
Diversity
Energy availability Net primary
productivity +
More productive locals provide more energy
(resources, food) for populations, enhancing
population persistance and population sizes
Wright, 1983; Murphy et al.
2010; Data source: MODIS 17,
NASA (2016).
Current
environmental
stability
Current climatic
suitability +
Locals with current suitable climatic
conditions favor higher survival and larger
population sizes
Hand et al. 2016; Kovach et al.
2015; Data source: WorldClim
v. 1.4.
Historical
environmental
stability
Historical climatic
suitability (average
from last 120 kyrs)
+
Locals with historical climatic suitability
sheltered higher survival and persistence of
populations through time
Carnaval et al. 2009; Graham et
al. 2006; Data source:
WorldClim v. 1.4; Carnaval et
al. 2014.
Variability of high
river flow (wet
season)
Coefficient of variation
of interannual high
river flow
- Annual variability on highest water levels
decreases recruitment by nests inundation
Batistella & Vogt, 2008;
Pantoja-Lima et al. 2009; Data
source: Silva Jr. et al. 2015.
Variability of low
river flow (dry
season)
Coefficient of variation
of interannual low river
flow
-
Annual variability on lowest water levels
decreases predictability of available nesting
beaches for local populations, hampering
nesting and reproductive success
Alho et al. 1982; Bermudez-
Romero et al. 2014; Ouellet-
Cauchon et al. 2014
Data source: Silva Jr. et al. 2015
47
Commercial
hunting
Distance to nearest
urban center +
Populations distant from urban centers are less
subjected to large removal of adult individuals
for traffic and illegal consumption, sheltering
larger population sizes
Alcântara et al. 2013; Schneider
et al. 2011; Smith, 1979
Subsistence
hunting
Kernel density of
riverine and rural
human communities
locations
-
Populations densely surrounded by human
rural settlements are more exposed to
predation of eggs (lower recruitment) and
nesting females (smaller population sizes)
Bernardes et al. 2014; Schneider
et al. 2011
Data source: IBGE
Downstream
Increase in
intraspecific
Genetic Diversity
(DIGD)
Distance from Amazon
river mouth -
Downstream locals have biased accumulation
of alleles and immigrants
Paz‐Vinas & Blanchet, 2015;
Paz-Vinas et al. 2015
Data source: Venticinque et al.
2016
Genetic
Differentiation
Isolation By
Distance (IBD) Riverway distance +
Reduced migration and mating between
geographically distant populations Wright, 1943
Isolation By
Resistance (IBR)
Cost distance of
resistance from river
types
+ Reduced migration and mating between
populations separated by costly water types
Beheregaray et al. 2015; de
Thoisy et al. 2006
Data source: Venticinque et al.
2016
Cost distance of
resistance from slope +
Reduced migration between populations
separated by topographically complex
pathways offering resistance from slope
Caldera & Bolnick, 2008; Cook
et al. 2011; Kanno et al. 2011
Data source: Domisch et al.
2015
Cost distance of
resistance from current
climatic suitability
+
Reduced migration and mating between
populations separated by current climatically
unsuitable pathways
Mitchell et al. 2015
48
Cost distance of
resistance from
historical climatic
suitability (average
from last 120 kyrs)
+
Reduced migration and mating through time
between populations separated by climatically
unsuitable historical pathways
Ortego et al. 2015
Isolation By
Barrier (IBB)
Presence of Amazon
river (only for P.
erythrocephala)
+
Reduced migration and mating between
populations in opposite margins of Amazon
river
Santos et al. 2016
For each hypothesis, we reviewed the literature for evidences and potential biological mechanisms linking the variables to the
predicted effects (“Pred.”) on genetic diversity or genetic differentiation. The predictor variables for genetic diversity describe
variation within localities (local variables) while the variables for genetic differentiation describe environmental variation between
localities (connectivity variables). References include both the literature reviewed and the data sources used to produce each predictor
variable.
49
Table 2. Model selection for genetic diversity of P. erythrocephala and P. sextuberculata. Best
models selected based on ΔAICc > 2 are bolded.
π Hd
Species Models K Δ AICc wAICc K Δ AICc wAICc
P.
erythrocephala
CVmax + distCITY 4 0.00 0.34 4 0.85 0.14
distCITY 3 1.06 0.20 3 0.35 0.18
NULL 2 1.96 0.13 2 0.00 0.22
P.
sextuberculata
NPP 3 0 0.19 3 1.89 0.08
NPP + villages 4 0.7 0.14 4 2.44 0.06
distMOUTH 3 1.22 0.11 3 0 0.21
NPP + suit_past 4 1.87 0.08 4 4.49 0.02
villages + distMOUTH 4 2.57 0.05 4 0.64 0.15
suit_past + distMOUTH 4 3.6 0.03 4 1.79 0.09
Abbreviations: π – nucleotide diversity; Hd – haplotype diversity; K – number of parameters estimated for
each model; ΔAICc – Akaike values corrected for small samples; wAICc – Akaike’s weight of evidence;
CVmax – coefficient of variation of interannual high river flow; distCITY – distance to nearest urban
center; NULL – null model representing the absence of an effect; NPP – net primary productivity; villages
– kernel density of human villages; distMOUTH – distance from Amazon River mouth; suit_past – mean
historical suitability.
50
Table 3. Model fit and relative importance of predictor variables (representing IBD, IBR and
IBB hypothesis) for link-level GDM analyses of genetic differentiation of P. erythrocephala and
P. sextuberculata. In “Model” the numbers are relative to the variables (below) included in that
model. Variable importance is the sum of I-splines coefficients. Dashes indicate zero coefficients
of I-splines. NA = not assessed. No variable was significant after 1,000 permutations.
Best model P. erythrocephala P. sextuberculata
Model 1 + 4 + 5 + 6 1 + 2 + 4
Model deviance 20.12 57.07
Percent deviance explained 20.45 6.49
p-value 0.12 0.11
Variable importance: P. erythrocephala P. sextuberculata
1. distance 0.110 0.187
2. river color resistance - 0.953
3. slope resistance - -
4. current suitability resistance 0.189 0.226
5. historical suitability resistance 0.136 -
6. Amazon river 0.387 NA
51
FIGURE LEGENDS
Figure 1. Scheme depicting the predicted genetic responses to demographic and neutral genetic
processes affected by landscape factors. The local landscape factors (node-level variables) can
increase or decrease population sizes, affecting the effective population sizes (Ne). Populations
with smaller Ne’s are subject to higher rates of genetic drift, leading to lower levels of genetic
diversity in these locals. Connectivity factors (link-level variables), on the other hand, will
enhance or diminish migration and mating success, having an effect on gene flow rates. Reduced
gene flow between populations will then lead to higher genetic differentiation among them. In
our study, we measured biologically meaningful landscape factors for river turtles and related
them to genetic responses to understand and discuss the processes generating this relationship.
Adapted with permission from DiLeo and Wagner (2016).
Figure 2. Sampling localities for P. erythrocephala. Pie charts represent percentage of
individuals belonging to three biological clusters identified by Bayesian analysis of population
structure in BAPS. BAPS graphs are depicted in Appendix B. Pink mask: potential geographic
distribution of P. erythrocephala as estimated by Fagundes et al 2015. Illustration: Karl Mokros.
Figure 3. Sampling localities for P. sextuberculata. Pie charts represent percentage of individuals
belonging to four biological clusters identified by Bayesian analysis of population structure in
BAPS. BAPS graphs are depicted in Appendix B. Yellow mask: potential geographic distribution
of P. sextuberculata as estimated by Fagundes et al 2015. Illustration: Karl Mokros.
Figure 4. Hypothetical scenario illustrating how to represent more accurately the environmental
dissimilarity among localities in isolation by resistance (IBR) models. Colors indicate resistance
by river color (water types): green – low cost (1), yellow – medium cost (2), and red – high cost
(5). Slope symbols indicate resistance by upstream slope: large symbols – high slope (100), small
symbols – medium slope (50), and absence of symbols – low slope (10). The hypothetical grid
has cells with 1 km of resolution. Due to geographic distances being too large or too small, three
paths would receive the same cost distance for river color (1-2, 1-4 and 1-5) and for slope (1-2, 1-
3, 1-5) despite the large environmental differences among them. By dividing by riverway
distance, we are able to separate the effects of geography and environment. We can therefore
obtain values (LCP/dist) that represent the environmental dissimilarity of pathways between pairs
52
of localities and assess the sole effect of that variable in our response variable (genetic
differentiation), despite geographical distance. LCP = least cost path. dist = riverway distance
(km).
Figure 5. Climate suitability from Ecological Niche Modelling for P. erythrocephala (a, c) and
P. sextuberculata (b, d). Red and yellow dots are the occurrence records for P. erythrocephala
and P. sextuberculata, respectively. We estimated current suitability (a, b) in MAXENT using six
World Clim variables. Historical suitability (c, d) is the average suitability of 62 points of time
from 120 kya to present (available in Carnaval et al 2014). For both species, we recovered low
mean climatic suitability for the past compared to present suitability.
Figure 6. Relationship among nucleotide diversity (π) of P. sextuberculata and local variables
included in best models selected with AICc: NPP (a), distMOUTH (b), villages (c) and suit_past
(d). Relationships of genetic diversity with distMOUTH, villages and suit_past are opposed to the
expected.
53
FIGURES
Figure 1
54
Figure 2
55
Figure 3
56
Figure 4
57
Figure 5
58
Figure 6.
59
Conclusões
De modo geral, meu trabalho mostra que variáveis locais podem ser fatores importantes
determinando padrões de diversidade genética de quelônios aquáticos, apesar da grande
importância comumente dada a variáveis de conectividade. Para Podocnemis sextuberculata, uma
espécie de grande capacidade de dispersão, fatores locais explicam metade da diversidade genética
intraespecífica. Altos valores de diversidade genética de P. sextuberculata estavam associados a
uma maior produtividade primária, maior densidade de vilas humanas, maior distância da foz do
Rio Amazonas e menor adequabilidade climática histórica. A falta de fortes relações entre
diferenciação genética e variáveis de link podem sugerir alta conectividade aquática na bacia
Amazônica, aumentando o fluxo gênico para quelônios. Isto é congruente com a fraca estrutura
populacional espacial recuperada para as duas espécies. Apesar da não-significância entre
diferenciação genética e variáveis de conectividade, houve uma tendência de maior efeito dos
fatores de conectividade na diferenciação genética da espécie de menor capacidade de dispersão,
P. erythrocephala, como evidenciado por maior porcentagem de variação explicada.
Adicionalmente, os modelos explicando maior variação na diferenciação genética foram
isolamento por barreira (i.e., rio Amazonas) para P. erythrocephala e isolamento por resistência
oferecida por tipos de rios com maiores custos (i.e., diferentes cores) para P. sextuberculata. Os
padrões recuperados neste trabalho também podem ser úteis para outros táxons aquáticos
Amazônicos, dado que as variáveis analisadas são biologicamente relevantes para espécies usando
ambientes de rios a nível de bacia. Eu espero que esta pesquisa possa estimular outros estudos sobre
quais aspectos de paisagens de rios na bacia Amazônica são potenciais fatores influenciando a
diferenciação e diversidade genéticas de vertebrados aquáticos em geral e quelônios aquáticos em
particular. O presente estudo é o primeiro a aplicar genética da paisagem de rios (i.e., Riverscape
Genetics) na bacia Amazônica e a usar modelos de resistência empíricos no contexto de Riverscape
Genetics. Portanto, deve prover uma abordagem para a investigação de padrões espaciais genéticos
de outras espécies aquáticas com grande capacidade de dispersão em sistemas de rios. Finalmente,
ao examinar padrões amplos, identifiquei potenciais fatores que poderiam receber uma
investigação mais local e profunda em futuros estudos.
60
Apêndice – Material suplementar do manuscrito submetido para Ecography
SUPPLEMENTARY MATERIAL
Oliveira, J. A., Farias, I. P., Costa, G. C. & Werneck, F. P. Model-based riverscape genetics: disentangling the roles of local and
connectivity factors in shaping spatial genetic patterns of two Amazonian turtles with different dispersal abilities. – Ecography 000:
000–000.
Appendix A. Table of samples and diversity measures by locality
Table A1. Localities sampled for each species. Localities numbers and codes are accordingly to map of samples. Coordinates are in
decimal degrees, datum WGS 84. N: number of individuals. S: number of polymorphic sites. h: number of haplotypes. Hd: haplotype
diversity. π: nucleotide diversity. State abbreviations: AM – Amazonas; RR – Roraima; PA – Pará; AC – Acre. Bolded locality codes
indicate localities included in landscape genetic analyses (N ≥ 10).
# Species Locality Water body County Code Latitude Longitude N S h Hd π
1 P. erythrocephala São Gabriel da Cachoeira Negro river AM SGC 0.3787 -67.3084 18 6 8 0.797 0.00290
2 P. erythrocephala Santa Isabel do Rio Negro Ayuanã river AM SIS -0.5424 -64.9231 34 2 3 0.116 0.00023
3 P. erythrocephala Barcelos Cumicurí river AM CUM -0.6831 -63.2007 29 8 7 0.377 0.00122
4 P. erythrocephala Barcelos Itú river AM ITU -0.4006 -63.4609 29 12 11 0.695 0.00309
5 P. erythrocephala Barcelos, Comun. Carvoeiro Negro river AM BCL -1.3948 -61.9810 4 0 1 - -
6 P. erythrocephala Parque Nacional do Viruá Iruá river RR PNV 0.9871 -61.2572 10 1 2 0.200 0.00040
7 P. erythrocephala Parque Nacional do Jaú Jaú river AM PNJ -1.9025 -61.7511 34 7 9 0.813 0.00316
8 P. erythrocephala Jaú river mouth Jaú river AM JAU -1.9031 -61.4261 4 0 1 - -
61
9 P. erythrocephala Nhamundá Paracatu river AM NHA -2.0215 -57.0620 23 6 6 0.458 0.00119
10 P. erythrocephala Terra Santa Jamary stream PA TSA -2.0775 -56.5164 19 6 6 0.468 0.00144
11 P. erythrocephala Oriximiná Sapucuá lake PA ORI -1.7941 -56.2023 17 2 3 0.324 0.00067
12 P. erythrocephala Barreirinha Andirá river AM BAR -3.1368 -57.1375 33 7 8 0.657 0.00185
13 P. erythrocephala Juruti Juruti Velho lake PA JUR -2.4106 -56.1979 17 1 2 0.118 0.00023
14 P. erythrocephala Parintins Uiacurapá river AM PAR -2.7736 -56.7797 2 0 1 - -
All 273 38 48 0.627 0.00234
1 P. sextuberculata São Paulo de Olivença Camatiã river AM SPO -3.4711 -69.0244 24 13 10 0.844 0.00323
2 P. sextuberculata ESEC Juami-Japurá Japurá river AM JAP -1.6470 -68.1560 36 6 6 0.713 0.00191
3 P. sextuberculata RESEX do Alto Juruá Juruá river AC AJU -8.1265 -72.8079 10 9 6 0.911 0.00544
4 P. sextuberculata RESEX do Médio Juruá Juruá river AM MJU -5.4537 -67.4449 31 12 9 0.583 0.00351
5 P. sextuberculata RESEX do Baixo Juruá Juruá river AM BJU -3.4752 -66.0902 12 12 8 0.909 0.00581
6 P. sextuberculata RESEX Catuá-Ipixuna Solimões river AM IPX -3.7590 -64.0668 2 0 1 - -
7 P. sextuberculata Branco river mouth Branco river AM BRA -1.4075 -61.6539 2 1 2 - -
8 P. sextuberculata Parque Nacional do Viruá Anauá river RR PNV 0.9783 -61.2730 10 10 5 0.756 0.00375
9 P. sextuberculata REBIO do Abufari Purus river AM ABU -5.3853 -63.0778 19 6 4 0.614 0.00188
10 P. sextuberculata RDS Piagaçu-Purus Ayapuá stream AM PPP -4.4391 -62.2944 2 3 2 - -
11 P. sextuberculata RDS Piagaçu-Purus Purus river AM PUR -4.2896 -61.8658 19 10 7 0.544 0.00259
12 P. sextuberculata Manicoré Capanã lake AM CAP -6.0288 -61.8946 13 7 7 0.872 0.00288
13 P. sextuberculata Manicoré Madeira river AM MAD -5.6553 -61.2485 18 4 5 0.405 0.00073
14 P. sextuberculata Parintins Amazonas river AM PAR -2.6341 -57.1842 22 7 4 0.658 0.00191
15 P. sextuberculata Barreirinha Andirá river AM BAR -2.8747 -57.0835 16 14 10 0.867 0.00507
16 P. sextuberculata Nhamundá Nhamundá river AM NHA -1.9466 -56.9646 14 1 2 0.143 0.00024
17 P. sextuberculata Terra Santa Piraruacá lake PA TSA -2.0117 -56.3233 12 1 2 0.303 0.00050
62
18 P. sextuberculata REBIO Trombetas Jacaré lake PA TRO -1.3588 -56.8655 21 13 9 0.681 0.00439
19 P. sextuberculata Oriximiná Sapucuá lake PA ORI -1.8081 -56.0031 13 4 4 0.603 0.00178
20 P. sextuberculata Vitória do Xingu Xingu river PA XIN -2.7137 -52.0225 40 9 6 0.279 0.00181
All 336 42 61 0.776 0.00467
63
Appendix B. Descriptive genetic analyses: haplotype network, Bayesian analysis of
population admixture, pairwise φST and AMOVA
Methods - Descriptive genetic analyses
To describe the genealogical relationships among localities we constructed a haplotype network
for each species on HAPLOVIEWER (Salzburger et al. 2011) using maximum likelihood
phylogenetic trees estimated in RAXML (STAMATAKIS 2006). To assess patterns of population
structure at the broad scale for each species we inferred the most probable number of genetic
clusters (K) and individual’s assignment to each cluster with a Bayesian analysis of population
admixture implemented in BAPS v. 6.0 (Corander et al. 2006). To characterize intraspecific
genetic differentiation among localities we performed an analysis of molecular variance
(AMOVA) (Excoffier et al. 1992) using pairwise φST between sampling sites using the software
ARLEQUIN v. 3.5.2.2 (Excoffier and Lischer 2010), testing for significance by randomization with
1,000 permutations.
Results - Descriptive genetic analyses
Our data set for P. erythrocephala (N = 273) included 38 polymorphic sites, resulting in 48
haplotypes of which 34 were unique. For P. sextuberculata (N = 336), we found 42 polymorphic
sites and 61 haplotypes of which 40 were unique. Both species have a high proportion of shared
haplotypes among localities. We recovered overall moderate haplotype diversity for both species
(P. erythrocephala: Hd = 0.627; P. sextuberculata: Hd = 0.776), with localities values ranging
from 0.116 to 0.813 for P. erythrocephala and from 0.143 to 0.911 for P. sextuberculata. We
found high nucleotide diversity compared to other studies using control region of mtDNA for
Podocnemis species (see Pearse et al. 2006), 0.00234 for P. erythrocephala (range of 0.00023 to
0.00316) and 0.00458 for P. sextuberculata (range of 0.00024 to 0.00581).
The analysis of population admixture implemented in BAPS recovered three genetic
clusters (Ln likelihood = -713.0684) for P. erythrocephala and four genetic clusters (Ln
likelihood = -1007.2423) for P. sextuberculata. The clustering of individuals for both species did
not correspond to geographical locations, except for the clusters including mainly individuals
from São Gabriel da Cachoeira (1 – SGC) for P. erythrocephala and from Xingu River (20 –
64
XIN) for P. sextuberculata. The populations (i.e., localities) were significantly differentiated for
both species (P. erythrocephala: φST = 0.34060, p < 0.0001; P. sextuberculata: φST = 0.45353,
p < 0.0001), with pairwise φST between localities ranging from 0 to 0.898 for P. erythrocephala
and from 0 to 0.937 for P. sextuberculata.
65
Haplotype networks for P. erythrocephala and P. sextuberculata
Figure B1. Genealogical relationships among mtDNA Control Region (CR, 503 bp) haplotypes
from 273 samples of Podocnemis erythrocephala in 14 localities in Amazon basin.
Figure B2. Genealogical relationships among mtDNA Control Region (CR, 605 bp) haplotypes
from 336 samples of Podocnemis sextuberculata in 20 localities in Amazon basin.
66
Population structure: BAPS graphs, pairwise φST and AMOVA table
Figure B3. Biological population groups of Podocnemis erythrocephala estimated by Bayesian analysis of population structure in
BAPS software. BAPS partitioned the populations into three biological groups of populations. Localities codes and numbers are
according to Table A1.
Figure B4. Biological population groups of Podocnemis sextuberculata estimated by Bayesian analysis of population structure in
BAPS software. BAPS partitioned the populations into four biological groups of populations. Localities codes and numbers are
according to Table A1.
67
Table B1. Pairwise φST (lower diagonal) and riverway distance (km; upper diagonal) between localities of Podocnemis
erythrocephala studied. We did not included localities 5 – BCL, 8 – JAU and 14 – PAR in landscape genetic analyses due to possible
sampling bias (N < 10) on φST estimates. Bold values indicate significant φST values (p < 0.05).
1 - SGC 2 - SIS 3 - CUM 4 – ITU 5 - BCL 6 - PNV 7 - PNJ 8 - JAU 9 - NHA 10 - TSA 11 - ORI 12 - BAR 13 - JUR
1 - SGC 0 356.2 578.1 600.6 754.0 1083.8 921.7 866.2 1735.1 1653.2 1730.6 1669.9 1667.0
2 - SIS 0.757 0 225.0 247.5 400.9 730.7 568.5 513.1 1381.9 1300.1 1377.5 1316.8 1313.9
3 - CUM 0.635 0.011 0 67.6 184.4 514.2 352.0 296.6 1165.4 1083.6 1161.0 1100.2 1097.4
4 - ITU 0.440 0.065 0.019 0 243.6 573.4 411.2 355.7 1224.6 1142.8 1220.2 1159.4 1156.6
5 - BCL 0.567 -0.141 -0.135 -0.096 0 329.8 167.6 112.1 981.0 899.2 976.6 915.8 913.0
6 - PNV 0.630 0.022 -0.023 -0.002 -0.122 0 470.2 414.7 1283.6 1201.8 1279.2 1218.4 1215.6
7 - PNJ 0.534 0.215 0.166 0.141 0.035 0.121 0 55.5 939.5 857.7 935.1 874.3 871.5
8 - JAU 0.567 -0.141 -0.135 -0.096 0.000 -0.122 0.035 0 884.0 802.2 879.6 818.9 816.0
9 - NHA 0.644 0.023 0.010 0.040 -0.129 -0.016 0.127 -0.129 0 110.4 197.6 261.5 151.9
10 - TSA 0.620 0.038 0.014 0.033 -0.125 -0.013 0.132 -0.125 -0.012 0 115.8 179.7 70.0
11 - ORI 0.660 0.053 0.012 0.037 -0.103 0.015 0.158 -0.103 0.019 0.022 0 257.1 147.4
12 - BAR 0.626 0.175 0.125 0.114 0.005 0.091 0.182 0.005 0.105 0.099 0.129 0 193.5
13 - JUR 0.759 0.882 0.669 0.470 0.898 0.857 0.496 0.898 0.690 0.669 0.796 0.597 0
14 - PAR 0.509 -0.333 -0.325 -0.275 0.000 -0.324 -0.109 0.000 -0.319 -0.316 -0.290 -0.147 0.885
Table B1. (continuation)
14 - PAR
1 - SGC 1597.4
2 - SIS 1244.3
3 - CUM 1027.8
68
4 - ITU 1087.0
5 - BCL 843.4
6 - PNV 1146.0
7 - PNJ 801.9
8 - JAU 746.4
9 - NHA 189.1
10 - TSA 107.3
11 - ORI 184.7
12 - BAR 84.0
13 - JUR 121.1
14 - PAR 0
Table B2. Pairwise φST (lower diagonal) and riverway distance (km; upper diagonal) between localities of Podocnemis sextuberculata
studied. We did not included localities 6 – IPX, 7 – BRA and 10 – PPP in landscape genetic analyses due to possible sampling bias (N
< 10) on φST estimates. Bold values indicate significant φST values (p < 0.05).
1 – SPO 2 - JAP 3 - AJU 4 - MJU 5 - BJU 6 - IPX 8 - BRA 9 - PNV 10 - ABU 11 - PPP 12 - PUR 13 - CAP
1 - SPO 0 1225.1 1993.1 1067.7 663.9 794.1 1637.7 1971.5 1583.4 1314.1 1244.6 2042.9
2 - JAP 0.083 0 2180.0 1254.6 850.8 652.9 1496.5 1830.3 1442.3 1172.9 1103.4 1901.8
3 - AJU 0.080 0.152 0 926.0 1329.2 1749.0 2592.6 2926.4 2538.4 2269.0 2199.5 2997.8
4 - MJU 0.011 0.075 0.036 0 403.8 823.6 1667.2 2001.0 1613.0 1343.6 1274.1 2072.5
5 - BJU 0.033 0.159 0.027 0.015 0 419.8 1263.4 1597.2 1209.2 939.8 870.3 1668.7
6 - IPX -0.273 -0.105 -0.142 -0.228 -0.206 0 843.6 1177.4 789.4 520.0 450.5 1248.8
7 - BRA -0.177 -0.282 -0.218 -0.213 -0.175 0.000 0 333.8 944.6 675.3 605.8 1017.0
8 - PNV -0.015 0.026 -0.010 -0.042 -0.006 -0.269 -0.31 0 1278.4 1009.1 939.6 1350.9
69
9 - ABU 0.043 0.013 0.073 0.016 0.088 -0.100 -0.34 -0.024 0 336.7 338.9 1349.9
10 - PPP 0.051 0.340 0.027 0.090 -0.063 0.000 0.00 0.052 0.268 0 69.5 1080.5
11 - PUR -0.008 0.081 0.074 -0.007 0.032 -0.264 -0.18 -0.024 0.017 0.016 0 1011.0
12 - CAP 0.019 0.145 0.096 0.057 0.044 -0.201 -0.08 0.016 0.122 0.124 0.053 0
13 - MAD 0.023 0.110 0.204 0.054 0.131 -0.330 0.14 0.056 0.132 0.505 0.037 0.092
14 - PAR 0.072 0.185 0.209 0.106 0.136 -0.183 0.06 0.095 0.184 0.247 0.085 0.095
15 - BAR 0.028 0.067 0.004 -0.014 0.008 -0.209 -0.28 -0.045 -0.003 0.004 0.009 0.058
16 - NHA 0.579 0.541 0.528 0.518 0.533 0.937 0.85 0.597 0.647 0.884 0.641 0.685
17 - TSA 0.031 0.147 0.194 0.051 0.107 -0.237 0.32 0.077 0.173 0.542 0.047 0.116
18 - TRO 0.016 0.099 0.044 -0.021 0.021 -0.235 -0.20 -0.035 0.043 0.048 0.008 0.062
19 - ORI 0.040 0.057 0.101 0.042 0.100 -0.204 -0.19 0.012 0.044 0.296 0.039 0.107
20 - XIN 0.781 0.828 0.726 0.731 0.718 0.843 0.82 0.774 0.815 0.829 0.797 0.813
Table B2. (continuation)
14 – MAD 15 – PAR 16 - BAR 17 – NHA 18 - TSA 19 - TRO 20 - ORI 21 - XIN
1 - SPO 1940.2 1735.2 1867.4 1950.3 1888.3 2136.9 2009.5 2686.7
2 - JAP 1799.0 1594.0 1726.2 1809.1 1747.2 1995.7 1868.3 2545.6
3 - AJU 2895.1 2690.1 2822.3 2905.2 2843.2 3091.8 2964.4 3641.6
4 - MJU 1969.7 1764.7 1896.9 1979.8 1917.9 2166.4 2039.0 2716.3
5 - BJU 1565.9 1360.9 1493.1 1576.0 1514.1 1762.6 1635.2 2312.5
6 - IPX 1146.1 941.1 1073.3 1156.2 1094.2 1342.8 1215.4 1892.6
7 - BRA 914.3 709.3 841.5 924.4 862.4 1111.0 983.6 1660.8
8 - PNV 1248.1 1043.1 1175.3 1258.2 1196.3 1444.8 1317.4 1994.7
9 - ABU 1247.1 1042.1 1174.3 1257.2 1195.3 1443.8 1316.4 1993.7
10 - PPP 977.8 772.8 905.0 987.9 925.9 1174.5 1047.1 1724.3
70
11 - PUR 908.3 703.2 835.5 918.4 856.4 1105.0 977.6 1654.8
12 - CAP 102.8 796.7 928.9 1011.8 949.8 1198.4 1071.0 1748.2
13 - MAD 0 693.9 826.1 909.0 847.1 1095.6 968.2 1645.5
14 - PAR 0.092 0 132.2 215.1 153.2 401.7 274.3 951.6
15 - BAR 0.094 0.132 0 214.3 152.3 400.9 273.5 950.7
16 - NHA 0.857 0.731 0.460 0 113.0 375.6 248.2 925.5
17 - TSA 0.033 0.106 0.096 0.903 0 201.0 48.0 863.5
18 - TRO 0.067 0.106 -0.006 0.510 0.06 0 159.0 903.1
19 - ORI 0.076 0.137 0.046 0.732 0.11 0.052 0 775.7
20 - XIN 0.867 0.841 0.710 0.876 0.87 0.717 0.838 0
71
Table B3. Results of AMOVA for P. erythrocephala and P. sextuberculata assessing whether
there is structure among localities studied (14 localities for P. erythrocephala and 20 for P.
sextuberculata).
P. erythrocephala
Source of variation d.f. Sum of
squares
Variance
components
Percentage
of variation
Among populations 13 56.281 0.20655 Va 34.06
Within populations 259 103.565 0.39987 Vb 65.94
Total 272 159.846 0.60641 -
Fixation Index (FST): 0.341 p < 0.0001
P. sextuberculata
Source of variation d.f. Sum of
squares
Variance
components
Percentage
of variation
Among populations 19 221.939 0.66054 Va 45.35
Within populations 316 251.502 0.79589 Vb 54.65
Total 335 473.440 1.45643 -
Fixation Index (FST): 0.453 p < 0.0001
Appendix C. Assessing resistance costs offered by water types
We sent the closed format questionnaire bellow to six turtle specialists, explaining the objective
of the assessment and asking them to assign a cost value associated to water types for P.
erythrocephala and P. sextuberculata. We prepared the questionnaire in Google Forms© and
send to experts by e-mail.
72
Questionnaire sent to experts:
Cost for movement of turtles in different water types
Thank you for your contribution!
In my study I will assess, for each pair of populations (localities), whether the cost of moving
between them is associated to the degree of pairwise genetic differentiation (e.g., the costlier is
the movement between two locals, the highest will be the genetic differentiation between
individuals of these locals). Here I am only assessing the cost of movement related to the
different water types: white, clear and blackwaters. The cost of movement is a value representing
the resistance of an animal to cross a particular environment, the physiological cost for the animal
to cross that environment, a reduction in the survival of the organism or an integration of all these
factors. For the study, the cost of movement through different water types will be measured from
a raster layer of the rivers in which each pixel (1 km x 1 km) will have a cost value associated to
the water color (how costly it is for individuals of that species to move on that water type?). Since
there are no studies empirically quantifying whether these turtles present preference, lower
survival or differential mobility between water masses of different colors, it is here that I ask your
expert opinion. The cost values for movement of each species on the different water types will be
based on your opinion ("expert opinion"). To this, I ask you to consider, besides the known
distribution of both species, your knowledge regarding the habitat and microhabitat preferences,
relative abundance in different water types (e.g., if population studies indicate higher abundance
in a particular water type comparing to other), occasional records and your own experience in
general. On the next few pages, I will ask about the cost of movement in white, clear and black
waters for Podocnemis erythrocephala and P. sextuberculata. You may answer only to the
species you are familiar to, in case it is not both of them.
73
Amazon River: example of whitewater river
Tapajós River: example of clearwater river
Negro River: example of blackwater river
74
The objective is to attribute a value, from 1 to 5, which represents how costly you think it is, to
irapuca/iaçá, to move in that particular water type. The value of 1 means that that water type
offers little to any cost for the animal movement. The value of 5 means that that water type is
almost impassable for irapuca/iaçá, offering a high cost of movement for the animal.
Do you have any additional comment/suggestion about water types for irapuca/iaçá?
______________________________________________________________________________
______________________________________________________________________________
Results from questionnaire: cost values based on expert’s opinion
Since the expert's opinions varied, we used the mean cost values attributed to each water type for
each species as the resistance cost.
75
Figure C1. Cost values attributed by experts to each water type for each species studied. Circles:
iaçá (Podocnemis sextuberculata). Triangles: irapuca (Podocnemis erythrocephala). n = number
of opinions. The gray star symbols are the average of values attributed by experts to each water
type for each species.
Appendix D. Predictor variables
Landscape variables
Node-level local variables
For energy availability hypothesis, we used the mean Net Primary Productivity (NPP) from 2000
to 2015 (NASA 2016) to represent the availability of food (fruits, seeds and leaves) for turtles in
each locality sampled. For each locality, we obtained the mean NPP in a buffer of 5 km of radius
76
for P. erythrocephala and of 12 km of radius for P. sextuberculata. We selected these buffer radii
based on mark-and-recapture and movement studies describing mean linear distances for
individuals of each species (Fachín-Terán et al. 2006, Bernhard 2010 unpublished data).
To represent the historical and current environmental stability hypothesis, we used
Ecological Niche Modeling (ENM) to predict the climatic suitability for each species. We used a
total of 158 occurrence records for P. erythrocephala and 329 for P. sextuberculata. These
occurrences were compiled in a recent study by Fagundes et al. (2015) and include data from
literature review, Brazilian scientific collections and museum specimens, and unpublished data
from turtle specialists. We first modeled the climate-based habitat suitability for each species
using the maximum entropy machine-learning algorithm, MAXENT, implemented in the R
package dismo (Hijmans et al. 2015). To construct the models we used seven bioclimatic
variables (BIO1, BIO4, BIO10, BIO11, BIO 12, BIO15, BIO16 AND BIO17) from the
WorldClim database (http://www.worldclim.org) interpolated to 1 km resolution (Hijmans et al.
2005), removing highly correlated variables (r > 0.8). We produced 20 replicate model runs to
statistically evaluate the models, using 75% of the records for training and 25% for testing. We
evaluated model performance using the area under the curve (AUC) of the receiver operating
characteristic (ROC) plot, which ranges from 0.5 (random prediction) to 1 (maximum prediction).
To enable a continuous view of historical climatic suitability we projected the models to 62
climatic reconstructions covering the last 120 kyr at small time intervals (1 to 4 kyr) using the
Hadley Centre Climate model (HadCM3; Singarayer and Valdes 2010, Fuchs et al. 2013,
Carnaval et al. 2014). The output raster layers have an index of suitability for each cell ranging
from 0 to 1, being low values indicative of unsuitable conditions for species occurrence and high
values indicative of suitable conditions. In the R package raster (Hijmans and Van Etten 2014)
we calculated the mean value of suitability for the 62 layers of time. We used the resulting mean
raster layer as the historical climatic suitability map and the present projection as the current
suitability map. Finally, we extracted the historical and current suitability values for each locality.
Variability on the lowest river flow (dry season) can affect the reproductive behavior of
turtles by decreasing the predictability of available nesting beaches annually (Bermudez-Romero
et al. 2015) while variability on the highest river flow (wet season) can cause flooding of nests
(Pantoja-Lima et al. 2009). We therefore used two raster maps created by Silva-Junior (2015)
77
through calculation of the coefficient of variation of the high river flows (CVmax) and low river
flows (CVmin) for 5 thousand points in the Amazon basin for the period of 1998 to 2009. To
represent high variability of extreme water levels we extracted for each locality the CVmax and
CVmin values.
Harvest of turtles in the Amazon occurs historically either through subsistence
consumption or through illegal commercial hunting (Smith 1979, Fachin-Terán and Vogt 2014,
Pantoja-Lima et al. 2014). Thus, we developed metrics to characterize both types of hunting. To
represent the probable influence of the presence of rural/riverine human villages and
communities on turtle stocks (subsistence consumption), we generated a kernel-density map of
human communities. We considered all the locations of villages, isolated rural settlements and
indigenous villages in an area comprehending the distribution of samples for both species
(available at IBGE 2015). We used the heatmap tool in QGIS 2.14.2 (QGIS Development Team
2014), which implements a kernel function to fit a smooth surface to each point and calculate the
density of points in a region (cell size: 0.0251, radius: 80,000). The output raster gives kernel
values, which are relative values of the density of points (in this case, villages). We used the
kernel values of each sampling locality as a surrogate for subsistence hunting pressure. In
addition, because urban centers are the final destination for illegally caught turtles (Fachín-Terán
et al. 2004), we measured for each sampling locality the distance (by river way) to the closest
urban center as a mean to characterize illegal commercial hunting. We considered all the cities
and capitals recognized as urban centers by IBGE (2015).
Finally, to assess if there is a pattern of Downstream Increase in Intraspecific Genetic
Diversity (DIGD), we defined the mouth of Amazonas River as the ultimate downstream point,
considering the broad distribution of both species in the Amazon River basin. We extracted for
each locality the distance by river way to the Amazon River mouth from a vector layer depicting
these values, created by Venticinque et al. (2016).
Results of Ecological Niche Modelling (ENM)
The climatic variables included in the final ENMs for both species were: temperature seasonality
(BIO4), mean temperature of warmest quarter (BIO10), annual precipitation (BIO12),
precipitation seasonality (BIO15) and precipitation of wettest quarter (BIO16). The average
78
training AUC for the replicate runs was high for both species (P. erythrocephala: 0.969, SD
0.004; P. sextuberculata: 0.933, SD 0.006), indicating high model fit. The estimated mean
historical suitability (from 0 to 120 kya) for the species indicates that, in average, past climatic
conditions were mostly unsuitable for species occurrence (Figure 5). The resistance pathways for
current and historical climatic conditions for each species can be visualized in the maps at
Appendix D.
Figure D1. Net primary productivity (NPP) map. Average annual NPP from 2000-2015. Source:
USGS NASA - MODIS 17 (https://modis.gsfc.nasa.gov).
79
Figure D2. Coefficient of variation of interannual high river flow (CVmax). Source: Silva Jr.
2015.
Figure D3. Coefficient of variation of interannual low river flow (CVmin). Source: Silva Jr.
2015.
80
Figure D4. Brazilian urban centers (capitals and cities) within the Amazon Basin. Source: IBGE
(http://www.ibge.gov.br).
Figure D5. Kernel density of Brazilian human villages (rural settlements, villages and indigenous
lands) within the area (rectangle) containing our study samples. Source of villages locations:
IBGE (http://www.ibge.gov.br)
81
Figure D6. Vector containing distances to the Amazon River mouth. Source: Venticinque et al.
2016.
Figure D7. Raster of upstream slope. Source: Domisch et al. 2015.
82
Figure D8. Resistance from current climatic suitability for Podocnemis erythrocephala.
Resistance was calculated as (1 – current suitability). Current suitability was estimated using
MaxEnt (see Methods).
Figure D9. Resistance from current climatic suitability for Podocnemis sextuberculata.
Resistance was calculated as (1 – current suitability). Current suitability was estimated using
MaxEnt (see Methods).
83
Figure D10. Resistance from historical climatic suitability for Podocnemis erythrocephala.
Resistance was calculated as (1 – historical suitability). Historical suitability refers to mean
suitability (average of 62 layers of time covering the past 120 kyr). Historical suitability was
estimated using MaxEnt (see Methods).
Figure D11. Resistance from historical climatic suitability for Podocnemis sextuberculata.
Resistance was calculated as (1 – historical suitability). Historical suitability refers to mean
84
suitability (average of 62 layers of time covering the past 120 kyr). Historical suitability was
estimated using MaxEnt (see Methods).
Figure D12. Vector containing river types (colors) for the Amazon Basin. Source: Venticinque et
al. 2016.
Appendix E. Exploratory node-level analyses and full AIC tables for node-level landscape
genetics analyses
Exploratory node-level analyses
We first tested for multicollinearity among predictor variables (Tables E1 and E2).
85
Table E1. Coefficient of Pearson’s correlation (r) among predictor variables for P.
erythrocephala. Predictors with r > 0.6 (bold values) were not included in the same mixed
models.
NPP suitPRE suitPAST CVmin CVmax distCITY villages distFOZ
NPP
suitPRE -0.354
suitPAST -0.295 0.666
CVmin 0.824 -0.631 -0.647
CVmax 0.473 -0.497 -0.364 0.707
distCITY 0.512 -0.438 0.068 0.264 0.361
villages -0.163 0.336 0.456 -0.189 -0.324 -0.425
distFOZ 0.797 -0.487 -0.367 0.756 0.216 0.348 -0.075
Table E2. Coefficient of Pearson’s correlation (r) among predictor variables for P.
sextuberculata. Predictors with r > 0.6 (bold values) were not included in the same mixed
models.
NPP suit_pre suit_past CVmin CVmax distCITY kernel distFOZ
NPP
suit_pre -0.581
suit_past -0.577 0.705
CVmin 0.395 -0.241 -0.009
CVmax 0.189 -0.028 0.120 0.584
distCITY 0.045 0.097 0.042 0.013 0.163
kernel -0.056 0.160 0.422 -0.098 -0.375 -0.309
distFOZ 0.926 -0.637 -0.580 0.308 -0.041 0.107 -0.007
We then tested for spatial autocorrelation (Moran’s I correlograms) of response variables
performed in the software SAM v. 4.0 (Rangel et al. 2010). Tables E3 – E6 and Figures E1 – E4.
86
Table E3. Moran’s I values (10 distance classes) and significance (P) for Hd (Haplotype
diversity) of P. erythrocephala.
D.Class Count DistCntr Moran's I P I (max) I/I(max)
1 18 56.151 -0.474 0.196 1.337 -0.355
2 16 109.173 0.104 0.739 1.173 0.089
3 16 216.174 -0.009 0.995 0.828 -0.01
4 16 349.051 -0.173 0.533 0.707 -0.245
5 16 491.724 0.264 0.392 0.739 0.358
6 16 582.205 -0.072 0.834 0.632 -0.115
7 16 635.316 -0.258 0.397 0.836 -0.309
8 16 713.835 -0.232 0.437 0.975 -0.238
9 16 839.127 -0.422 0.221 1.164 -0.363
10 18 1106.73 -0.021 0.899 0.304 -0.069
Expected: -0.077
Table E4. Moran’s I values (10 distance classes) and significance (P) for π (Nucleotide diversity)
of P. erythrocephala.
D.Class Count DistCntr Moran's I P I (max) I/I(max)
1 18 56.151 -0.362 0.241 1.414 -0.256
2 16 109.173 0.143 0.673 1.156 0.123
3 16 216.174 0.032 0.925 1.037 0.031
4 16 349.051 -0.337 0.251 0.7 -0.481
5 16 491.724 0.238 0.508 0.439 0.543
6 16 582.205 0.013 0.955 0.708 0.019
7 16 635.316 -0.162 0.603 0.911 -0.178
8 16 713.835 -0.255 0.402 0.896 -0.284
9 16 839.127 -0.511 0.106 1.173 -0.435
10 18 1106.73 -0.151 0.538 0.741 -0.204
Expected: -0.077
87
Table E5. Moran’s I values (10 distance classes) and significance (P) for Hd (Haplotype
diversity) of P. sextuberculata.
D.Class Count DistCntr Moran's I P I (max) I/I(max)
1 38 98.624 -0.098 0.538 1.084 -0.09
2 36 258.639 -0.204 0.276 0.982 -0.207
3 36 458.71 0.253 0.171 1.142 0.222
4 36 574.276 -0.2 0.342 1.158 -0.173
5 36 642.884 0.044 0.804 0.965 0.046
6 36 723.798 -0.066 0.764 0.871 -0.076
7 36 822.173 0.365 0.06 1.194 0.305
8 36 1020.69 -0.295 0.151 0.941 -0.314
9 36 1245.19 0.091 0.663 0.823 0.111
10 36 1861.34 -0.338 0.121 1.242 -0.272
Expected: -0.053
Table E6. Moran’s I values (10 distance classes) and significance (P) for π (Nucleotide diversity)
of P. sextuberculata.
D.Class Count DistCntr Moran's I P I (max) I/I(max)
1 38 98.624 -0.249 0.206 0.972 -0.257
2 36 258.639 -0.307 0.151 1.081 -0.284
3 36 458.71 0.178 0.422 0.718 0.247
4 36 574.276 -0.084 0.683 0.8 -0.105
5 36 642.884 0.076 0.774 0.789 0.096
6 36 723.798 -0.035 0.879 0.609 -0.058
7 36 822.173 0.047 0.824 0.862 0.054
8 36 1020.69 -0.071 0.688 0.506 -0.14
9 36 1245.19 0.033 0.839 0.472 0.07
10 36 1861.34 -0.25 0.146 0.89 -0.281
Expected: -0.053
88
Figure E1. Moran’s I correlogram for Hd (Haplotype diversity) of P. erythrocephala.
Figure E2. Moran’s I correlogram for π (Nucleotide diversity) of P. erythrocephala.
89
Figure E3. Moran’s I correlogram for Hd (Haplotype diversity) of P. sextuberculata.
Figure E4. Moran’s I correlogram for π (Nucleotide diversity) of P. sextuberculata.
90
Full tables with models tested for model selection, containing all models tested, number of
parameters (K), Akaike Information Criterion corrected for small samples (AICc), AICc
difference from best model (Δ AICc), Akaike’s weight of evidence (wAICc) and cumulative
wAICc (Cum. Wt.).
Table E7. All models associating nucleotide diversity and local variables for P. erythrocephala.
Bolded values are considered the best models (Δ AICc < 2).
Podocnemis erythrocephala - Nucleotide diversity (pi)
K AICc Δ AICc wAICc Cum.Wt
pi ~ CVmax + distCITY 4 -115.7 0 0.34 0.34
pi ~ distCITY 3 -114.6 1.06 0.2 0.54
pi ~ NULL 2 -113.7 1.96 0.13 0.67
pi ~ distMOUTH 3 -111.13 4.53 0.04 0.71
pi ~ CVmin + distCITY 4 -110.76 4.9 0.03 0.74
pi ~ CVmax 3 -110.67 4.99 0.03 0.77
pi ~ villages 3 -110.4 5.26 0.02 0.79
pi ~ NPP + distCITY 4 -110.06 5.6 0.02 0.81
pi ~ suit_pre 3 -110.04 5.63 0.02 0.83
pi ~ suit_past 3 -109.99 5.67 0.02 0.85
pi ~ NPP 3 -109.97 5.69 0.02 0.87
pi ~ CVmin 3 -109.91 5.75 0.02 0.89
pi ~ distCITY + distMOUTH 4 -109.71 5.95 0.02 0.91
pi ~ suit_pre + distCITY 4 -109.61 6.05 0.02 0.93
pi ~ suit_past + distCITY 4 -109.54 6.12 0.02 0.94
pi ~ distCITY + villages 4 -109.37 6.29 0.01 0.96
pi ~ CVmax + distMOUTH 4 -107.65 8.01 0.01 0.96
pi ~ suit_past + distMOUTH 4 -106.96 8.7 0 0.97
pi ~ CVmax + villages 4 -106.95 8.71 0 0.97
pi ~ suit_pre + CVmax 4 -106.87 8.79 0 0.98
91
pi ~ NPP + CVmax 4 -106.57 9.09 0 0.98
pi ~ villages + distMOUTH 4 -106.46 9.2 0 0.98
pi ~ suit_past + villages 4 -106.11 9.55 0 0.99
pi ~ suit_pre + distMOUTH 4 -105.89 9.77 0 0.99
pi ~ CVmin + villages 4 -105.46 10.2 0 0.99
pi ~ suit_past + CVmax 4 -105.45 10.21 0 0.99
pi ~ NPP + villages 4 -105.27 10.39 0 0.99
pi ~ suit_pre + villages 4 -105.24 10.42 0 1
pi ~ NPP + suit_past 4 -105.14 10.52 0 1
pi ~ NPP + suit_pre 4 -104.88 10.78 0 1
Table E8. All models associating haplotype diversity and local variables for P. erythrocephala.
Bolded values are considered the best models (Δ AICc < 2).
Podocnemis erythrocephala - Haplotype diversity (Hd)
K AICc Δ AICc wAICc Cum.Wt
hd ~ NULL 2 5.78 0 0.22 0.22
hd ~ distCITY 3 6.13 0.35 0.18 0.4
hd ~ CVmax + distCITY 4 6.63 0.85 0.14 0.54
hd ~ CVmax 3 8.82 3.04 0.05 0.59
hd ~ NPP + distCITY 4 9.07 3.29 0.04 0.63
hd ~ CVmin + distCITY 4 9.12 3.34 0.04 0.67
hd ~ CVmin 3 9.19 3.42 0.04 0.71
hd ~ distMOUTH 3 9.21 3.43 0.04 0.75
hd ~ suit_past 3 9.23 3.45 0.04 0.79
hd ~ villages 3 9.43 3.66 0.03 0.82
hd ~ suit_pre 3 9.57 3.79 0.03 0.86
hd ~ NPP 3 9.68 3.9 0.03 0.89
hd ~ suit_past + distCITY 4 10.91 5.14 0.02 0.9
hd ~ suit_pre + distCITY 4 11.1 5.32 0.02 0.92
92
hd ~ distCITY + villages 4 11.28 5.5 0.01 0.93
hd ~ distCITY + distMOUTH 4 11.35 5.58 0.01 0.95
hd ~ suit_pre + CVmax 4 13.01 7.24 0.01 0.95
hd ~ CVmax + distMOUTH 4 13.1 7.32 0.01 0.96
hd ~ CVmax + villages 4 13.21 7.44 0.01 0.96
hd ~ suit_past + distMOUTH 4 13.32 7.55 0 0.97
hd ~ suit_past + villages 4 13.52 7.75 0 0.97
hd ~ suit_past + CVmax 4 13.9 8.13 0 0.98
hd ~ NPP + CVmax 4 13.94 8.17 0 0.98
hd ~ CVmin + villages 4 13.96 8.19 0 0.98
hd ~ villages + distMOUTH 4 14.22 8.44 0 0.99
hd ~ suit_pre + distMOUTH 4 14.44 8.67 0 0.99
hd ~ NPP + suit_past 4 14.46 8.69 0 0.99
hd ~ NPP + villages 4 14.61 8.83 0 0.99
hd ~ suit_pre + villages 4 14.63 8.85 0 1
hd ~ NPP + suit_pre 4 14.71 8.93 0 1
Table E9. All models associating nucleotide diversity and local variables for P. sextuberculata.
Bolded values are considered the best models (Δ AICc < 2).
Podocnemis sextuberculata - Nucleotide diversity (pi)
K AICc Δ AICc wAICc Cum.Wt
pi ~ NPP 3 -167.78 0 0.19 0.19
pi ~ NPP + villages 4 -167.08 0.7 0.14 0.33
pi ~ distMOUTH 3 -166.56 1.22 0.11 0.44
pi ~ NPP + suit_past 4 -165.9 1.87 0.08 0.51
pi ~ NPP + suit_pre 4 -165.44 2.34 0.06 0.57
pi ~ villages + distMOUTH 4 -165.21 2.57 0.05 0.63
pi ~ NULL 2 -164.87 2.91 0.05 0.67
pi ~ NPP + distCITY 4 -164.32 3.46 0.03 0.71
93
pi ~ NPP + CVmin 4 -164.3 3.47 0.03 0.74
pi ~ NPP + CVmax 4 -164.3 3.48 0.03 0.77
pi ~ suit_past + distMOUTH 4 -164.18 3.6 0.03 0.81
pi ~ CVmax + distMOUTH 4 -163.55 4.23 0.02 0.83
pi ~ villages 3 -163.44 4.34 0.02 0.85
pi ~ CVmin + distMOUTH 4 -163.12 4.66 0.02 0.87
pi ~ distCITY + distMOUTH 4 -163.07 4.71 0.02 0.89
pi ~ CVmin 3 -162.52 5.26 0.01 0.9
pi ~ suit_pre 3 -162.21 5.57 0.01 0.91
pi ~ CVmax 3 -162.14 5.64 0.01 0.93
pi ~ suit_past 3 -162.07 5.71 0.01 0.94
pi ~ distCITY 3 -161.93 5.85 0.01 0.95
pi ~ suit_past + villages 4 -161.2 6.58 0.01 0.96
pi ~ CVmax + villages 4 -161.2 6.58 0.01 0.96
pi ~ CVmin + villages 4 -160.9 6.88 0.01 0.97
pi ~ suit_pre + villages 4 -160.63 7.15 0.01 0.97
pi ~ distCITY + villages 4 -160.4 7.38 0 0.98
pi ~ suit_past + CVmin 4 -159.23 8.55 0 0.98
pi ~ suit_pre + CVmin 4 -159.2 8.58 0 0.98
pi ~ CVmin + distCITY 4 -159.08 8.69 0 0.99
pi ~ CVmin + CVmax 4 -159.04 8.74 0 0.99
pi ~ suit_pre + CVmax 4 -158.98 8.8 0 0.99
pi ~ suit_past + CVmax 4 -158.91 8.87 0 0.99
pi ~ suit_pre + distCITY 4 -158.81 8.97 0 1
pi ~ CVmax + distCITY 4 -158.68 9.1 0 1
pi ~ suit_past + distCITY 4 -158.64 9.14 0 1
94
Table E10. All models associating haplotype diversity and local variables for P. sextuberculata.
Bolded values are considered the best models (Δ AICc < 2).
Podocnemis sextuberculata - Haplotype diversity (Hd)
K AICc Δ AICc wAICc Cum.Wt
hd ~ distMOUTH 3 -0.66 0 0.21 0.21
hd ~ villages + distMOUTH 4 -0.02 0.64 0.15 0.36
hd ~ suit_past + distMOUTH 4 1.13 1.79 0.09 0.45
hd ~ NPP 3 1.23 1.89 0.08 0.53
hd ~ NPP + villages 4 1.78 2.44 0.06 0.59
hd ~ NULL 2 2.61 3.27 0.04 0.63
hd ~ CVmin + distMOUTH 4 2.62 3.28 0.04 0.68
hd ~ distCITY + distMOUTH 4 2.74 3.4 0.04 0.71
hd ~ CVmax + distMOUTH 4 2.83 3.49 0.04 0.75
hd ~ suit_pre 3 3.52 4.18 0.03 0.78
hd ~ villages 3 3.73 4.39 0.02 0.8
hd ~ NPP + suit_past 4 3.83 4.49 0.02 0.82
hd ~ suit_pre + villages 4 3.95 4.61 0.02 0.84
hd ~ NPP + CVmax 4 4.39 5.05 0.02 0.86
hd ~ NPP + CVmin 4 4.42 5.08 0.02 0.88
hd ~ NPP + distCITY 4 4.5 5.16 0.02 0.89
hd ~ NPP + suit_pre 4 4.58 5.24 0.02 0.91
hd ~ distCITY 3 5.35 6.01 0.01 0.92
hd ~ suit_past 3 5.38 6.04 0.01 0.93
hd ~ CVmin 3 5.48 6.14 0.01 0.94
hd ~ CVmax 3 5.58 6.24 0.01 0.95
hd ~ suit_past + villages 4 5.71 6.37 0.01 0.96
hd ~ distCITY + villages 4 6.17 6.83 0.01 0.96
hd ~ suit_pre + distCITY 4 6.55 7.21 0.01 0.97
hd ~ CVmin + villages 4 6.96 7.62 0 0.97
hd ~ suit_pre + CVmax 4 6.98 7.64 0 0.98
95
hd ~ suit_pre + CVmin 4 7.01 7.67 0 0.98
hd ~ CVmax + villages 4 7.02 7.68 0 0.99
hd ~ suit_past + distCITY 4 8.59 9.25 0 0.99
hd ~ CVmin + distCITY 4 8.72 9.38 0 0.99
hd ~ suit_past + CVmin 4 8.75 9.41 0 0.99
hd ~ CVmax + distCITY 4 8.8 9.46 0 1
hd ~ CVmin + CVmax 4 8.81 9.47 0 1
hd ~ suit_past + CVmax 4 8.86 9.52 0 1
Appendix F. Correlation among predictor environmental matrices
Here we present the results from correlation (Mantel tests) between predictor environmental
matrices before and after correction by riverway distance. We applied correction by distance only
to IBR matrices: river color, slope, resistance from current suitability and resistance from
historical suitability. Number of permutations: 10,000.
Table F1. Matrix correlation among predictors of genetic differentiation of P. erythrocephala.
Upper diagonal: correlation coefficients among predictors after correction of IBR matrices (we
divided river color, slope, current and historical suitability matrices by riverway distance). Lower
diagonal: correlation among predictors before correction. Bolded values indicate significance (p
< 0.05).
IBD IBR IBB
distance river color slope current historical barrier
IBD Distance - 0.54 -0.08 0.69 0.37 0.03
IBR
river color 0.92 - 0.23 0.46 0.15 0.15
Slope 0.94 0.97 - -0.21 0.06 0.40
Current 0.99 0.91 0.93 - 0.64 -0.12
Historical 0.99 0.91 0.93 0.99 - -0.13
IBB Barrier 0.03 0.09 0.04 0.04 0.02 -
96
Table F2. Matrix correlation among predictors of genetic differentiation of P. sextuberculata.
Upper diagonal: correlation coefficients among predictors after correction of IBR matrices (we
divided river color, slope, current and historical suitability matrices by riverway distance). Lower
diagonal: correlation among predictors before correction. Bolded values indicate significance (p
< 0.05).
IBD IBR
distance river color slope current historical
IBD distance - -0.48 0.06 0.52 0.33
IBR
river
color 0.94 - -0.15 -0.56 -0.07
slope 0.84 0.77 - 0.13 0.33
current 0.96 0.88 0.81 - 0.61
historical 1 0.94 0.85 0.97 -
97
Appendix G. Genetic dissimilarity modeling (GDM) response curves for Podocnemis
erythrocephala and P. sextuberculata.
98
Figure G1. Generalized dissimilarity model-fitted I-splines (panels a–d) for connectivity variables
– for which non-zero splines coefficients could be calculated – associated to genetic
differentiation (φST) of Podocnemis erythrocephala. The variables are illustrated in descending
order of their coefficient values (maximum height reached by each curve), indicating total
amount of φST turnover associated with that variable. The shape of each function provides an
indication of how the rate of φST turnover varies along the dissimilarity environmental gradient.
Rug plots show actual dissimilarity values (LCP/riverway distance) between sampling locations.
The final two panels illustrate the relationships between (e) observed pairwise φST in the dataset
and the linear predictor of the GDM (“Predicted Ecological Distance”) and (f) observed versus
predicted φST.
99
100
Figure G2. Generalized dissimilarity model-fitted I-splines (panels a–c) for connectivity variables
– for which non-zero splines coefficients could be calculated – associated to genetic
differentiation (φST) of Podocnemis sextuberculata. The variables are illustrated in descending
order of their coefficient values (maximum height reached by each curve), indicating total
amount of φST turnover associated with that variable. The shape of each function provides an
indication of how the rate of φST turnover varies along the dissimilarity environmental gradient.
Rug plots show actual dissimilarity values (LCP/riverway distance) between sampling locations.
The final two panels illustrate the relationships between (d) observed pairwise φST in the dataset
and the linear predictor of the GDM (“Predicted Ecological Distance”) and (e) observed versus
predicted φST.
REFERENCES
Bermudez-Romero, A. L. et al. 2015. Nesting habitat of the "cupiso" Podocnemis sextuberculata
(Testudines: Podocnemididae) in Erepecu lake (Pará - Brazil). — Acta Biológica
Colombiana 20: 183-191.
Bernhard, R. 2010. Dinâmica populacional de Podocnemis erythrocephala, no rio Ayuanã,
Amazonas, Brasil. Programa Integrado de Pós-Graduação em Ecologia. INPA.
Carnaval, A. C. et al. 2014. Prediction of phylogeographic endemism in an environmentally
complex biome. — Proceedings of the Royal Society of London B: Biological Sciences
281: 20141461.
Corander, J. et al. 2006. BAPS: Bayesian analysis of population structure, Manual v. 4.1. —
Department of Mathematics, University of Helsinki.
Domisch, S. et al. 2015. Near-global freshwater-specific environmental variables for biodiversity
analyses in 1 km resolution. — Scientific data 2: 1-13.
Excoffier, L. and Lischer, H. E. 2010. Arlequin suite ver 3.5: a new series of programs to perform
population genetics analyses under Linux and Windows. — Molecular Ecology Resources
10: 564-567.
Excoffier, L. et al. 1992. Analysis of molecular variance inferred from metric distances among
DNA haplotypes: application to human mitochondrial DNA restriction data. — Genetics
131: 479-491.
101
Fachin-Terán, A. and Vogt, R. C. 2014. Alimentação de Podocnemis sextuberculata (Testudines:
Podocnemididae) na Reserva Mamirauá, Amazonas, Brasil. — Revista Colombiana de
Ciencia Animal 6: 286-309.
Fachín-Terán, A. et al. 2004. Patterns of Use and Hunting of Turtles in the Mamirauá Sustainable
Development Reserve, Amazonas, Brazil. — In: Silvius, K. M. et al. (eds), People in
Nature: Wildlife Conservation in South and Central America. Columbia University Press,
pp. 362-377.
Fachín-Terán, A. et al. 2006. Seasonal movements of Podocnemis sextuberculata (Testudines:
Podocnemididae) in the Mamiraua sustainable development reserve, Amazonas, Brazil.
— Journal Information 5: 18-24.
Fagundes, C. K. et al. 2015. Testing the efficiency of protected areas in the Amazon for
conserving freshwater turtles. — Diversity and Distributions 22: 123-135.
Fuchs, J. et al. 2013. Extending ecological niche models to the past 120 000 years corroborates
the lack of strong phylogeographic structure in the Crested Drongo (Dicrurus forficatus
forficatus) on Madagascar. — Biological Journal of the Linnean Society 108: 658-676.
Hijmans, R. et al. 2005. WorldClim, version 1.3. — University of California, Berkeley
Hijmans, R. J. et al. 2015. dismo: Species distribution modeling. R package version 1.0-12.
Hijmans, R. J. and Van Etten, J. 2014. raster: Geographic data analysis and modeling. R package
version 2.2-31.
IBGE 2015. BC250 - Base Cartográfica Contínua do Brasil.
Mittermeier, R. A. et al. 2015. Podocnemis erythrocephala (Spix 1824) – Red-headed Amazon
River Turtle, Irapuca. — In: Rhodin, A. et al. (eds), Conservation Biology of Freshwater
Turtles and Tortoises: A Compilation Project of the IUCN/SSC Tortoise and Freshwater
Turtle Specialist Group. Chelonian Research Monographs, pp. 087.1-10.
NASA 2016. Moderate Resolution Imaging Spectroradiometer (MODIS), NASA.
Pantoja-Lima, J. et al. 2014. Chain of commercialization of Podocnemis spp. turtles (Testudines:
Podocnemididae) in the Purus River, Amazon basin, Brazil: current status and
perspectives. — Journal of Ethnobiology and Ethnomedicine 10: 8.
102
Pantoja-Lima, J. et al. 2009. Seleção de locais de desova e sobrevivência de ninhos de quelônios
Podocnemis no baixo Rio Purus, Amazonas, Brasil. — Revista Colombiana de Ciencia
Animal 1: 37-59.
Pearse, D. E. et al. 2006. Estimating population structure under nonequilibrium conditions in a
conservation context: continent-wide population genetics of the giant Amazon river turtle,
Podocnemis expansa (Chelonia; Podocnemididae). — Molecular Ecology 15: 985-1006.
QGIS Development Team 2014. QGIS geographic information system. — Open Source
Geospatial Foundation Project.
Rangel, T. F. et al. 2010. SAM: a comprehensive application for spatial analysis in
macroecology. — Ecography 33: 46-50.
Rueda-Almonacid, J. V. et al. 2007. Las tortugas y los crocodilianos de los paises andinos del
trópico. — In: Internacional, C. (ed), Serie de guias tropicales de campo n. 6.
Salzburger, W. et al. 2011. The performance of phylogenetic algorithms in estimating haplotype
genealogies with migration. — Molecular Ecology 20: 1952-1963.
Santos, R. C. et al. 2016. Testing the effects of barriers on the genetic connectivity in Podocnemis
erythrocephala (Red-headed Amazon River Turtle): implications for management and
conservation. — Chelonian Conservation and Biology 15: 12-22.
Singarayer, J. S. and Valdes, P. J. 2010. High-latitude climate sensitivity to ice-sheet forcing over
the last 120kyr. — Quaternary Science Reviews 29: 43-55.
Sioli, H. 1984. The Amazon: limnology and landscape ecology of a mighty tropical river and its
basin. — Springer Science & Business Media.
Smith, N. J. 1979. Aquatic turtles of Amazonia: an endangered resource. — Biological
Conservation 16: 165-176.
Stamatakis, A. 2006. RAxML-VI-HPC: maximum likelihood-based phylogenetic analyses with
thousands of taxa and mixed models. — Bioinformatics 22: 2688-2690.
van Etten, J. 2012. gdistance: Distances and routes on geographical grids. R package version 1.1–
4.
Venticinque, E. et al. 2016. An explicit GIS-based river basin framework for aquatic ecosystem
conservation in the Amazon. — Earth System Science Data 8: 651.
103
Wang, I. J. et al. 2013. Quantifying the roles of ecology and geography in spatial genetic
divergence. — Ecology Letters 16: 175-182.
Wang, I. J. et al. 2009. Landscape genetics and least‐cost path analysis reveal unexpected
dispersal routes in the California tiger salamander (Ambystoma californiense). —
Molecular Ecology 18: 1365-1374.
Zeller, K. A. et al. 2012. Estimating landscape resistance to movement: a review. — Landscape
Ecology 27: 777-797.