ArticlePDF Available

Habitat specialization predicts demographic response and vulnerability of floodplain birds in Amazonia

November 2023
Molecular Ecology 33(3)

November 2023
33(3)

DOI:10.1111/mec.17221

Authors:

Eduardo De Deus Schultz

American Museum of Natural History

Gabriela Zuquim

University of Turku

Michael Hickerson

City College of New York

Show all 6 authorsHide

The annual flooding cycle of Amazonian rivers sustains the largest floodplains on Earth, which harbour a unique bird community. Recent studies suggest that habitat specialization drove different patterns of population structure and gene flow in floodplain birds. However, the lack of a direct estimate of habitat affinity prevents a proper test of its effects on population histories. In this work, we used occurrence data, satellite images and genomic data (ultra‐conserved elements) from 24 bird species specialized on a variety of seasonally flooded environments to classify habitat affinities and test its influence on evolutionary histories of Amazonian floodplain birds. We demonstrate that birds with higher specialization in river islands and dynamic environments have gone through more recent demographic expansion and currently have less genetic diversity than floodplain generalist birds. Our results indicate that there is an intrinsic relationship between habitat affinity and environmental dynamics, influencing patterns of population structure, demographic history and genetic diversity. Within the floodplains, historical landscape changes have had more severe impacts on island specialists, making them more vulnerable to current and future anthropogenic changes, as those imposed by hydroelectric dams in the Amazon Basin.

Content uploaded by Eduardo De Deus Schultz

Content may be subject to copyright.

Molecular Ecology. 2023;00:1–17. wileyonlinelibrary.com/journal/mec

1 | INTRODUCTIO N

Amazonia is the most biodiverse region on the planet and a nexus for

discussion on how physiographic and climatic processes drive diver-

sification (Cracraf t et al., 2020; Hoorn et al., 2010). Across its basin,

several of the largest rivers on Earth along with their broad flood-

plains currently delimit the distribution of multiple upland forest

vertebrates (Cracraft, 1985; Mourthé et al., 2022). To what extent

the evolution of this continental-scale river system and past climatic

changes affecting environment distributions were responsible for

Received: 18 April 2023

Revised: 17 October 2023

Accepted: 14 November 2023

DOI: 10.1111/mec.17221

ORIGINAL ARTICLE

Habitat specialization predicts demographic response and

vulnerability of floodplain birds in Amazonia

Eduardo D. Schultz1,2 | Gregory Thom3 | Gabriela Zuquim4,5 |

Michael J. Hickerson6 | Hanna Tuomisto4 | Camila C. Ribas7

1Programa de Pós-Gra duação em Biologia

(Ecologia), Instituto Nacional de Pesquisas

da Amazôn ia, Manaus, AM, Brazil

2Department of Ornithology, American

Museum of Natural History, New York,

New York, USA

3Museum of Natural Science and

Department of Biological Sciences,

Louisiana State University, Baton Rouge,

Louisiana, USA

4Department of Biology, University of

Turku, Turku, Finland

5Department of Biology, Aarhus

University, Aarhus, Denmark

6Depar tment of Biology, City College of

New York, New York, New York, USA

7Coordenação de Biodiversidade, Instituto

Nacional de Pesquisas da Amazônia,

Manaus, AM, Brazil

Correspondence

Eduardo D. Schultz, Programa de

Pós-Graduação em Biologia (Ecologia),

Instituto Nacional de Pesquisas da

Amazônia, Manaus, AM, Brazil.

Email: edsbio@gmail.com

Funding information

Conselho Nacional de Desenvolvimento

Científ ico e Tecnológico, Grant/Award

Number: 311732/2020-8; Coordenação

de Aper feiçoamento de Pessoal de Nível

Superior; Fundação de Amparo à Pesquisa

do Estado do Amazonas, Grant/Award

Number: 002/2016 - POSGRAD 2017;

United States Agency for International

Development, Grant/Award Number:

PEER— Co Ag AID-OAA-A-11-00012

Handling Editor: Yanhua qu

Abstract

The annual flooding cycle of Amazonian rivers sustains the largest floodplains on

Earth, which harbour a unique bird community. Recent studies suggest that habitat

specialization drove different patterns of population structure and gene flow in flood-

plain birds. However, the lack of a direct estimate of habitat affinity prevents a proper

test of its effects on population histories. In this work, we used occurrence data, sat-

ellite images and genomic data (ultra-conserved elements) from 24 bird species spe-

cialized on a variety of seasonally flooded environments to classify habitat affinities

and test its influence on evolutionary histories of Amazonian floodplain birds. We

demonstrate that birds with higher specialization in river islands and dynamic envi-

ronments have gone through more recent demographic expansion and currently have

less genetic diversity than floodplain generalist birds. Our results indicate that there

is an intrinsic relationship between habitat affinity and environmental dynamics, in-

fluencing patterns of population structure, demographic history and genetic diversity.

Within the floodplains, historical landscape changes have had more severe impacts on

island specialists, making them more vulnerable to current and future anthropogenic

changes, as those imposed by hydroelectric dams in the Amazon Basin.

KEY WORDS

climate change, comparative demography, dam impacts, habitat affinities, quaternary, ultra-

conserved elements

SCHULTZ et al.

the forma tion of such diversit y is much debated (Cr acraft et al., 2020;

Haffer, 1969; Musher et al., 2022; Ribas et al., 2012, 2022). Recent

studies on co-distributed bird species from upland forests have re-

covered discordant evolutionary histories, which might suggest that

a single community-wide pattern of diversification does not exist

(Musher et al., 2022; Naka & Brumfield, 2018; Silva et al., 2019;

Smith et al., 2014). However, intrinsic ecological aspects of the spe-

cies, such as physiological constraints and habitat affinities, might

affect how populations respond to ex trinsic environmental changes,

and are expected to result in idiosyncratic evolutionary patterns

(Papadopoulou & Knowles, 2016). For example, the preference for

distinct forest strata in upland birds predicts genetic differenti-

ation between populations, with canopy species having shallower

differentiation than species occupying the understory (Burney &

Brumfield, 2009; Smith et al., 2014).

In the Amazonian floodplains, a specialized biota occupies a myr-

iad of distinct environments created by the annual flood cycle of the

rivers. These environments, that cover approximately 750,000 km2

(~11%) of the area of the Amazon basin, are deeply influenced by

the geomorphology of the rivers and the geological provinces they

drain (Wittmann et al., 2022). While clear and black water rivers

drain old cratons with low amounts of sediments and suspended

matter, leading to slow-growing floodplain vegetation, white water

rivers originate on Andean slopes and carry high amounts of nutri-

ents and sediments that continually reshape their floodplains, which

sustain a more stratified and dynamic vegetation (Junk et al., 2011).

Furthermore, floodplain habitats away from the main river channel

tend to be more stable than the ones closer to the river edges and

on river islands, which are more impacted by varying rates of sed-

iment deposition and erosion. Higher substrate stability results in

floodplain forests with tall trees adapted to shorter periods of in-

undation, while dynamic environments are mostly dominated by

shorter, herbaceous vegetation that stays under water for many

months ever y year (Figure 1; Junk et al., 2 011; Tamura et al., 2019;

Wittmann et al., 2022). This environmental complexity admits an

avifauna with rather distinct habitat affinities within the floodplains,

including a remarkable diversity of river islands specialists (Remsen

& Parker, 1983; Rosenberg, 1990).

The current configuration of the Amazonian floodplains is the

result of the combined effect of high sea level and the intensifica-

tion of the South American Monsoons that enhanced sedimentation

across the basin since the Last Glacial Maximum (LGM; 21 thousand

years ago), accelerated through the Holocene (last 11.7 thousand

years; Goldberg et al., 2021; Sawakuchi et al., 2022). However, a

previous phase of floodplain formation was interrupted by a period

of reducing sea level, starting around 80 thousand years ago, when

increased river incision drastically reduced the extent of the flood-

plains (Pupim et al., 2019). This scenario point s to a dynamic history

of expansion and retraction of the floodplains during the quaternary

climatic cycles, with a potentially large and heterogeneous impact on

the distribution and persistence of populations of floodplain species.

Moreover, these fluctuations have likely affected the distribution of

specific habitats in different ways, with a stronger impact on the

more ephemeral and dynamic environments like river islands (Passos

et al., 2020; Sawakuchi et al., 2022). Evidence from different stud-

ies seem to corroborate this hypothesis showing a more tumultu-

ous evolutionary history in bird taxa associated with more dynamic

floodplain environments (Barbosa et al., 2021; Choueri et al., 2017;

Johnson et al., 2023; Luna et al., 2022; Sawakuchi et al., 2022; Thom

et al., 2020, 2022). Nonetheless, a comprehensive classification of

habitat specialization across the basin is still needed for a thorough

understanding of the effect of distinct habitat associations in the

responses of the endemic floodplain bird community to past land-

scape changes.

Understanding the specificities of floodplain birds’ habitat affin-

ities and their response to environmental changes involves an addi-

tional pr actical urgency considering the high anthropogenic pressure

to which Amazonian rivers are exposed. Currently, hundreds of im-

plemented, under construction and planned dams for energy gen-

eration threaten to change the sedimentation dynamics and water

flow in the entire basin (Latrubesse et al., 2017). These enterprises

affect specific environments differently, and the lack of knowledge

about differences in habitat affinities within Amazonian floodplains

prevents accurate estimates of the potential impact s of these an-

thropogenic changes in the specialized biota (Cochrane et al., 2017;

Forsberg et al., 2017; Latrubesse et al., 2021; Melo et al., 2021).

Therefore in this study, we tested whether habitat specialization

in the floodplain environments predicts population history. We ex-

pected that generalist floodplain species occupying a wide variety

of floodplain environments, or species occurring on habitats more

continuously distributed across the floodplains, to have more sta-

ble historical demographics and less spatial genetic struc ture, when

compared to species specialized on the ephemeral river edge and in-

sular environments, which are highly impacted by sedimentation dy-

namics through time. To test this hypothesis, we gathered genomic

and occurrence data for 24 bird species specialized in Amazonian

seasonally flooded environments. We used satellite images to clas-

sify habitat preferences based on the occurrence on islands and on

the dynamism of the islands with occurrence records. Then, consid-

ering our classification, we used genomic data to test if habitat use

explains patterns of demographic history and genetic diversity. We

further test the influence of population struc ture on demographic

histories and whether demographic patterns are distinct for popu-

lations occurring in different portions of the Amazon Basin. Finally,

based on the observed responses to past landscape changes we dis-

cuss the vulnerabilit y of the studied taxa to anthropogenic impacts

on the floodplain environments.

2 | MATERIALS AND METHODS

2.1 | Sampling and genomic data acquisition

We selected 24 bird species, belonging to 6 different families, that

are known specialists on Amazonian seasonally flooded environ-

ments and had tissue samples available in the largest collections of

1365294x, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/mec.17221 by Pennsylvania State University, Wiley Online Library on [29/11/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

SCHULTZ et al.

Amazonian birds in Brazil (INPA and MPEG) and USA (AMNH, ANSP,

FMNH, LSUMNS, USNM). All species are endemic of the Amazon

Basin, in some cases also including populations in the floodplains

of adjacent basins and coastal areas (Table S1, Figures S1–S5).

Together, the set of selected species occupies multiple Amazonian

seasonally flooded environments, which makes it possible to com-

pare habitat preferences and the effect these might have on the

demographic histor y of the species. The genomic approach used

to access the demographic history of the studied birds was the se-

quence capture of ultra-conserved elements (UCEs). For each spe-

cies, tissues were selected aiming to cover as broadly as possible its

distribution (Table S2, Figures S1–S5). For each sample, DNA was

extracted using Qiagen DNeasy Blood and Tissue kit (Valencia, CA)

following the manufacturer's protocol. DNA extracts were sent to

Rapid Genomics (Gainesville, FL) where UCEs were sequenced using

a probe set targeting 2321 UCEs, commonly used in avian studies

(Barbosa et al., 2021; Luna et al., 2021, 2022; Sawakuchi et al., 2022;

Thom et al., 2020, 2022).

After sequencing, data were processed from raw reads to se-

quence alignments and single-nucleotide polymorphisms (SNPs) for

downstream analyses through the Phyluce pipeline (Faircloth, 2016).

First, raw reads were cleaned from adapter contamination using

FIGURE 1 Rivers and floodplain environments in the Amazon Basin. (a) Amazon Basin map with rivers coloured by the water colour

types. (b) Early successional Várzea vegetation in the left bank of the Juruá river near Juruá – Brazil (white water). (c) Early successional

Várzea vegetation in an island in the lower Solimões river near Manacapuru – Brazil (white water). (d) Mature Várzea forest in an island in the

lower Solimões river near Coari – Brazil (white water). (e, f) Mature Igapó Forests in islands of the Mariuá archipelago in the Negro river near

the confluence with the Branco river.

SCHULTZ et al.

illumiprocessor function. Then, for each species a single sample with

high amounts of reads were assembled into contigs using SPAdes

(Bankevich et al., 2012), mapped against the probes using the func-

tion phyluce_assembly_match_contigs_to_probes and recovered con-

tigs were used as reference to all other samples. After wards, each

samples' reads were mapped against the reference alignment and

phased with phyluce_snp_bwa_multiple_align and phyluce_snp_phase_

uces respec tively (Andermann et al., 2018). Finally, sequences were

aligned into separate UCEs sequences files using mafft (Katoh &

Standley, 2013) and using phyluce_align_seqcap_align and trimmed to

improve alignment quality using phyluce_align_get_gblocks_trimmed_

alignments_from_untrimmed. We excluded samples with low cover-

age, usually less than 1500 UCEs, and loci that were not present

in all samples. This approach was used to have for each species a

dataset with more loci and shared by all samples. A SNP dataset was

obtained from the UCEs dataset by randomly selecting one SNP per

UCE, without including missing data, using snps_from_uce_align-

ments.py (h t t p s : / / g i t h u b . c o m / t o b i a s h o f m a n n 8 8 / U C E - d a t a - m a n a g

e m e n t / b l o b / m a s t e r / s n p s _ f r o m _ u c e _ a l i g n m e n t s . p y ).

2.2 | Population structure and definitions of

taxonomic units

To avoid population structure mistakenly affecting demographic

inferences, the SNPs datasets were used to estimate the best-

fit number of populations and define the taxonomic units using

sNMF function from LEA package in R (Frichot et al., 2014; R Core

Tea m, 2020). For each species, we estimated the cross-entropy of K

values from 1 to 10 populations, considering four different values

for the alpha parameter (1, 10, 100, 1000) with 100 replicates each,

and selected the K value with lowest cross-entropy across all runs.

This K value was used to define the working taxa for the main de-

mographic analyses and habitat categorization (Figures S1–S 5a ). To

visually access the spatial distribution of sNMF results, the ancestral

coefficient estimated for each sample was plotted in the map at each

sample's locality, for species with lowest cross-entropy value K = 1,

the second lowest value of K = 2 was plotted (Figures S1b–S5b).

As a complement to sNMF results, the estimated effective mi-

gration surfaces (EEMS) were used to look for areas of higher or

lower gene flow than expected under isolation by distance, as these

also reflect population structuring (Petkova et al., 2015). For each

species, Euclidean genetic distances (Rogers' distance) between

samples were estimated from the SNPs data with ADEGENET'S dist.

genpop function (Jombart & Ahmed, 2011). The described distribu-

tion of the species according to IUCN (https:// www. iucnr edlist. org/ )

was used as the distribution polygon, which was further edited when

necessary to include samples outside the polygon. Depending on the

size and complexity of the species' distribution polygon, values from

500 to 700 demes were assigned. The number of demes defines the

density of the population grid within the given distribution and, from

preliminary tests, we found these values to be suitable for the stud-

ied species. For each species, we present a single run of 10 × 106

generations, excluding the first 5 × 106 as burn-in (Figures S1–S 5c ).

The convergence of the runs was visually assessed in the posterior

probability trace plots.

For working taxa, as defined in the first sNMF analysis described

earlier that still included two geographically distinct clusters in sNMF

separated by areas of higher resistance to gene flow in agreement

with EEMS results (Figures S1–S5), we performed a second round of

demographic analyses treating these two clusters as distinct units.

The taxa resulting from this second approach are identified on the

maps in Figures S1–S5, and referred hereafter as populations, to

differentiate from the taxa defined in the first approach. We used

this approach to test if our demographic results were influenced by

population structure and if more geographically restricted lineages

would reveal distinct demographic patterns that could reflect local

history. Populations with few samples (fewer than nine individuals)

were excluded, as preliminary tests showed that the demographic

analyses with few samples recovered inconsistent results.

Working taxa and populations are identified in Figures S1–S5 by

its subspecies name when available, or by geographical identifiers

representing their distributions (W – West; E – East; Mad – Madeira;

Sol – Solimões; Ama – Amazonas; Neg – Negro).

2.3 | Habitat affinities

To test for differences in how species use the Amazonian flooding

environments, we gathered meta-data from the specimens in the

collections of INPA, LSUMNS and MPEG, and screened photo, vid-

eos and sound recordings available at the Macaulay Library (www.

macau layli brary. org) and Xeno-canto (www. xeno- canto. org) sub-

mitted until December 2020. Since available occurrence data for

Amazonian birds are not sufficiently accurate to infer fine-scale

habitat association, we relied on the pat tern of island specialization

to identif y differences in habitat affinities based on how each taxon

uses fluvial islands. Because islands have finite limits and are often

identified by the obser vers, we were able to calculate how often

each taxon was registered on islands or in the floodplains at the mar-

gins. Additionally, to test if different taxa prefer islands with distinct

dynamism, we used satellite images to assess how dynamic these

islands were based on changes in the amount of exposed water sur-

face between 1984 and 2018. This interval is not informative about

changes on an evolutionar y time scale, but depicts the characteris-

tics of the environments in which the taxa occur.

To calculate the proportions of registers on islands, ever y occur-

rence record was checked and, based on locality and observation

details, the occurrence was classified into one of the following two

categories: island or rivers margin. Records that could not be classi-

fied were excluded. To reduce bias from double counting the same

individual registered by different observers, distinct registers of the

same species on the same locality and date were counted only once

(Figures S1–S 5d, Table S3).

For the classification of island dynamism preference, the oc-

currences in islands were filtered, and only records on islands that

SCHULTZ et al.

could be identified were considered (Table S4). The identification of

islands was made by combining the geolocation, locality information

and obser vation details that indicated the specific island where the

variables to represent the dynamism of the island: amount of new

land, amount of lost land and degree of stability. The values for each

variable were derived from the water occurrence change intensity

dataset w hich measured the v ariation in water occ urrence on the sur-

face over multiple years using images from Landsat 5, 7 and 8 (Pekel

et al., 2016). This dataset measures both the direction of change

(increase, decrease or no change in water surface) and it s intensity.

Data were downloaded from https:// globa l- surfa ce- water. appsp ot.

com/ download and included information from 1984 to 2018. New

land was calculated based on the number of pixels with decrease in

water occurrence, weighted by the intensity of the change; lost land

was calculated by the number of pixels with increase in water occur-

rence, weighted by the intensity of the change; stability was calcu-

lated counting the pixels with invariant land through the observed

time. Constant water pixels were excluded. For each island, the val-

ues of the three variables were added and divided by the total to

obtain a proportional estimate. To extract the values, polygons were

drawn around each studied island in Qgis 3.14 (QGIS.org, 2020) and

exported as a shapefile. In R, the values within the polygon were

extracted from the raster with extract function from raster package

(Hijmans, 2020). The water colour classific ation of the river in which

each island was situated was obtained from a previous classification

of main Amazonian rivers (Venticinque et al., 2016) using the func-

tion ‘join attribute by location’ in Qgis.

The mean stability, gained and lost land for each taxon were cal-

culated considering all islands in which that taxon was registered. For

each of the three variables, the values extracted from each island were

added an d divided by the tot al of islands the ta xon were registere d. This

way, for each taxon we ended up with a proportional estimate of sta-

bility, gained and lost land in the islands they were registered that add

up to 1 (Tables 1 and S5). Following long-described evidences of bird

specialization in different successional stages of seasonally flooded en-

vironments (Remsen & Parker, 1983) and on islands (Rosenberg, 1990),

we tested if mean stability – representing the amount of dynamism –

and proportion of registers on islands could identify ecological groups,

which could be further used to test if birds with similar habitat affini-

ties shared similar demographic histories.

2.4 | Demographic inferences and genetic diversity

Demographic inferences were estimated in DILS (Fraïsse et al., 2021).

Using an approximate Bayesian computation framework, DILS allows

the estimation of demographic models considering the potential ef-

fects of linked selection. The analysis calculates the posterior proba-

bilities of models of demographic expansion, contraction or stability,

and estimates parameters like the time of demographic expansion

and met rics of genetic divers ity. DILS runs were perfor med using sin-

gle population analysis option, without outgroups. First, the 29 taxa

identified in the population structure analysis were used as evolu-

ti ona ry u nit s. To keep mo st lo ci an d con sider in g the data wer e al rea dy

filtered in Phyluce, max_N_tolerated was set to 0.3, Lmin to 100 and

nMin to the number of sequences per file. A general mutation rate

of 2.5 × 10−9 substitutions/site/generation with a 1-year generation

time was used to calibrate the analyses. Despite the uncertainty in-

volving substitution rates, the rate used here matches the rate pro-

posed for the genome of birds by Nater et al. (2015), the mean rate

of 2.3 × 10−9 across multiple bird lineages obtained by Nadachowska-

Brzyska et al. (2015) and previous studies of Amazonian floodplain

birds, making results comparable (Barbosa et al., 2021; Sawakuchi

et al., 2022; Thom et al., 2018, 2020). The values of the priors for

minimum and maximum time of demographic change were set to 100

and 5 × 105 generations, respectively, and limits of population sizes

were set at the default values of 100 for the minimum and 106 for

the maximum, encompassing with a margin previously published es-

timates of these parameters for Amazonian floodplain birds (Barbosa

et al., 2021; Luna et al., 2021, 2022; Sawakuchi et al., 2022; Thom

et al., 2020). To check for the consistency of the analyses, for each

taxon three runs were performed with the same priors and the run

with the highest sum of goodness-of-fit p values was chosen. The

parameter values used for downstream analyses were obtained fol-

lowing the optimized posterior of the neural network method.

To further explore if population structure influenced demo-

graphic results, additional demographic inferences were performed

using the populations within taxa detected in our secondary popula-

tion structure analysis. This procedure was per formed only for taxa

with cross-entropy values of K = 2 close to K = 1 and with the two

putative populations having distinct geographic distribution accord-

ing to sNMF and EEMS results. The same approach with the same

set of priors was used in DILS as when considering the taxon as a

single unit.

For taxa with cross-entropy values of K = 2 close to K = 1 and with

the two putative populations having distinct geographic distribution

according to sNMF and EEMS results, additional demographic infer-

ences were performed for these populations separately (Table S1).

This approach was used firstly to explore if population structure in-

fluenced demographic results and was compared with the result s

considering them a single taxon (Figure S6a). Moreover, as distinct

portions of the Amazon Basin could have gone through different

geological and environmental changes, there could be geographi-

cally distinct trends in population history along the vast Amazonian

floodplains. Therefore, we also compared the demographic signal of

populations with relatively overlapping distributions (Figure S6b).

Populations with widespread or idiosyncratic distributions were not

included in this geographic comparison. Demographic analyses for

these populations were performed in DILS with the same approach

and set of priors described earlier.

To access the genetic diversity of studied taxa, we used the av-

erage Tajima's pi (π) and Waterson's theta (θ) calculated in DILS for

each taxon. Values of π and θ describe the proportion of polymor-

phic sites and are indicatives of genetic diversity, with lower values

suggesting lower genetic diversity. The diversity measures were

SCHULTZ et al.

compared with the estimates of habitat affinities to test if there is a

relationship between genetic diversity and distinct habitat affinities.

2.5 | Hypotheses, statistical analyses and data

presentation

Our aim was to test if differences in habitat affinity influenced evo-

lutionary histories of co-distributed taxa. For that, we used linear

regressions and analyses of variance (ANOVAs) to test if our metric

of habitat use – represented by the propor tion of registers on islands

and the mean stabilit y of those islands – and our ecological groups

(see Results) explain the variation in demographic history (time of

demographic change) and population genetic parameters (π and θ). A

null hypothesis of no statistical correlation between habitat use and

the demographic and genetic variables means that the differences

between species are random or require an alternative explanation.

Alternatively, statistical correlation between variables indicates an

influence of habitat affinity in shaping demographic histories and/or

genetic diversity. Linear regressions between continuous variables

were used to test if more specialized birds had more pronounced

demographic responses, independent of categories. In turn, we used

the ANOVAs to test if our ecological categorization predicts similar

demographic histories for birds that occupy similar environments.

TABLE 1 Sampling, island use, genetic diversity and demographic variables for each taxon.

Tax on

Genomic

samples UCEs SNPs Habitat Islands Margins

Island

proportion

Mean

stability

Mean lost

land

Mean gained

land Average πAverage θ

Expansion

posterior

probability

HPD 2.5%

expansion time

Median

expansion time

HPD 97.5%

expansion time

Attila bolivianus 22 2242 1686 Generalists 27 106 20.3 0.6131 0.1153 0. 2717 0.0009 0.00164 0.9997 84,673.09 95, 417.19 1 07,1 26. 67

Capito aurovirens 13 2203 1177 Generalists 12 90.6 9. 4 0.6958 0 .1381 0.1660 0.0007 0.0009 0.9820 16 ,764.87 19,382. 65 22,412 .84

Conirostrum bicolor minus 14 2174 1596 Dynamic Specialists 63 12 84 0.4941 0.1074 0.3985 0.00 09 0.0013 0.9996 45,294.48 51, 65 0. 27 58,573.94

Cranioleuca gutturata 20 2256 1650 Generalists 16 80 21 .1 0.8490 0.0181 0.13 30 0.0013 0.00309 1.0000 150,903.81 161,455.87 173,331.51

Cranioleuca vulpecula 18 2192 1054 Dynamic Specialists 111 397. 37 0.2305 0.1498 0.6197 0.0004 0.00 069 0.9901 15 ,769.68 18,612. 62 21 ,93 9.0 3

Dendroplex kienerii 19 2237 1 519 Generalists 56 54 50.91 0.76 49 0.1209 0.1142 0.0008 0.00113 0.9980 44,841.92 55,043.63 66341.72

Furnarius minor 10 2194 1108 Dynamic Specialists 75 26 74.26 0.2791 0.1905 0.5305 0.0006 0.00079 0.9885 2 7,9 06. 28 32,213.99 36,746 .61

Knipolegus orenocensis sclateri 21 2239 1498 Dynamic Specialists 55 493.22 0. 2589 0.2678 0.4733 0.0009 0.00135 0.9878 24,088.68 28,454.01 33,368.78

Knipolegus poecilocercus 31 2145 1670 Generalists 32 51 38.55 0.8813 0.0331 0.0857 0.001 0.00195 0.9996 62,813.48 69,104 .33 76 ,437.4 0

Mazaria propinqua 22 2267 1450 Dynamic Specialists 82 20 80.39 0.2893 0.1668 0.5439 0.0006 0.00098 0.9990 38,286.81 43,892.87 50,207.36

Myiopagis flavivertex 21 2242 2130 Generalists 20 86 18.87 0.8192 0.0686 0.1122 0.0023 0.00432 1.0000 202,537.55 218,799.58 235,126.93

Myrmoborus lugubris lugubris 12 2256 2011 Generalists 313 18.75 0.810 0 0.1297 0.0603 0.0007 0.00088 0.9073 28 929.81 3 9,0 28 .4 8 49, 070.44

Myrmoborus lugubris berlepschi 28 2256 2011 Stable Specialists 82 297. 62 0 .610 0 0.1056 0.2844 0.00 05 0.00072 0.9823 12,642.98 15,395.36 18,661 .85

Myrmoborus lugubris femininus 13 2256 2011 Stable Specialists 9 2 81.82 0.7405 0.1435 0.1161 0.0008 0.00102 0.9958 63,114.72 80,610.59 99,674. 28

Myrmochanes hemileucus 22 2219 1063 Dynamic Specialists 129 894.16 0.3942 0.1627 0.4431 0.0004 0.00079 0.9823 18,791.46 21,009.60 23,442 .55

Myrmotherula assimilis Ama 16 2253 1985 Stable Specialists 51 11 82. 26 0.7576 0.0527 0.1898 0.001 0.00129 0.9996 72,867.78 88,796.22 107,444.59

Myrmotherula assimilis Mad 82253 1985 Generalists 214 12.5 0.8792 0.0867 0.0341 0.0011 0.0 0155 1.0000 118,448.27 138,385.30 160,544.54

Myrmotherula assimilis Sol 10 2253 1985 Generalists 18 21 4 6.15 0.6450 0.14 06 0. 2144 0.0011 0.00139 1.0000 90,453.46 111 ,468.95 136,10 9.3 9

Myrmotherula klagesi 21 2191 1314 Stable Specialists 48 10 82. 76 0.8338 0.0533 0.1129 0.0011 0.0022 1.0000 80,723.79 89, 13 0.1 1 97,986. 44

Myrmotherula multostriata 18 2237 1705 Generalists 21 105 16 .67 0.8934 0.0494 0.0572 0.0015 0.00277 1.0000 146,650.63 1 59,873 .6 0 173,603.83

Nasica longirostris 38 2227 2058 Generalists 33 207 13.75 0.7249 0.114 3 0.160 9 0.0012 0.00253 0.9992 123,446.58 135,090.56 147, 45 2. 35

Ochthornis littoralis 24 2081 14 47 Generalists 21 91 18.75 0.7318 0.1482 0.120 0 0.0007 0.00132 1.0000 56,203.12 67,703.40 80,140.40

Serpophaga hypoleuca hypoleuca 18 2233 1378 Dynamic Specialists 38 15 71.7 0.2742 0.2153 0.5105 0.0009 0.00165 1.0000 60,199.81 66 ,12 9.6 0 72,674 .39

Stigmatura napensis napensis 29 2231 1256 Dynamic Specialists 59 13 81.94 0 .3861 0.1848 0.4291 0.0007 0.00115 0.9651 22,812 .19 25,759.18 29, 099.85

Synallaxis albigularis 25 2233 1660 Generalists 60 69 46.51 0.5009 0.1242 0.3750 0.0009 0.00194 1.0000 92,530.14 9 9,3 51. 95 107,080.50

Thamnophilus cryptoleucus 23 2244 2117 Dynamic Specialists 114 39 7.4 4 0.4825 0.1512 0.3663 0.0007 0.00124 1.0000 70,125.48 81 ,141.58 92,995.31

Thamnophilus nigrocinereus cinereoniger 82249 2117 Stable Specialists 67 26 72.04 0.8382 0.0205 0.1412 0.0009 0.00095 0.9292 11 , 30 7.36 14,716.4 6 18,313.63

Thamnophilus nigrocinereus nigrocinereus 10 2249 2117 Stable Specialists 43 786 0.9248 0.0518 0.0234 0.0011 0.0014 0.9998 69,593 .2 2 91 , 697.2 0 116,710.01

Xiphorhynchus obsoletus 60 2191 1416 Generalists 80 2 61 23.46 0. 8317 0.0634 0.10 49 0.0006 0.00228 0.9990 69,9 86 .17 74,455.70 79,443.79

Note: For each taxon are depicted number of genomic samples; recovered UCEs; recovered SNPs; habitat affinity category; number of registers on

islands; number of regis ters on margins; propor tion of registers on islands; mean stability; mean lost land; mean gained land on islands it was

registered at; average π; average theta; average Tajima's D; posterior probability of population expansion model estimated in DILS; time of

demographic expansion in years, including lower and upper values of 95% confidence interval and the median.

SCHULTZ et al.

All statistical analyses on the correlation between variables

were performed in R and plotted with ggplot2 (Wickham, 2016).

The correlation coefficients between continuous variables pre-

sented in sc atter plots were estimated using stat_cor function using

Pearson method from the ggpubr package (Kassambara, 2020).

Confidence intervals shown in grey shades around the mean were

plotted with geom_smooth function from ggplot2. The ANOVA s be-

tween variables and habitat affinities categories were per formed

with stat_compare_means function from ggpubr package using

anova method.

The ternary plot with the mean values of the stability, lost land

and gained land in the islands each taxon occur (Figure 2b) was made

with ggtern (Hamilton & Ferry, 2018). The mean stability calculated

for each taxon was used in the ordination of the taxa in Figure 2c.

The ternary plots showing the proportion of the three variables

in each island the taxa were registered (Figure 3) were made with

ggtern. Islands were classified and coloured according to the rivers

where they occur. Habitat affinity categories were defined accord-

ing to the islands in which taxa were registered.

To test for phylogenetic independence of the results, we calcu-

lated the phylogenetic independent contrasts (PIC) of each variable

following the method described by Felsenstein (1985) using the fun c-

tion pic from ape package in R (Paradis & Schliep, 2019; Figure S7).

The phylogeny used as input for PIC calculation was built based on

TABLE 1 Sampling, island use, genetic diversity and demographic variables for each taxon.

Tax on

Genomic

samples UCEs SNPs Habitat Islands Margins

Island

proportion

Mean

stability

Mean lost

land

Mean gained

land Average πAverage θ

Expansion

posterior

probability

HPD 2.5%

expansion time

Median

expansion time

HPD 97.5%

expansion time

Attila bolivianus 22 2242 1686 Generalists 27 106 20.3 0.6131 0.1153 0. 2717 0.0009 0.00164 0.9997 84,673.09 95, 417.19 1 07,1 26. 67

Capito aurovirens 13 2203 1177 Generalists 12 90.6 9. 4 0.6958 0 .1381 0.1660 0.0007 0.0009 0.9820 16 ,764.87 19,382.65 22,412 .84

Conirostrum bicolor minus 14 2174 1596 Dynamic Specialists 63 12 84 0.4941 0.1074 0.3985 0.00 09 0.0013 0.9996 45,294.48 51, 65 0. 27 58,573.94

Cranioleuca gutturata 20 2256 1650 Generalists 16 80 21 .1 0.8490 0.0181 0.13 30 0.0013 0.00309 1.0000 150,903.81 161,455.87 173,331.51

Cranioleuca vulpecula 18 2192 1054 Dynamic Specialists 111 397. 37 0.2305 0.1498 0.6197 0.0004 0.00 069 0.9901 15 ,769.68 18,612. 62 21 ,93 9.0 3

Dendroplex kienerii 19 2237 1 519 Generalists 56 54 50.91 0.76 49 0.1209 0.1142 0.0008 0.00113 0.9980 44,841.92 55,043.63 66 341.72

Furnarius minor 10 2194 1108 Dynamic Specialists 75 26 74.26 0.2791 0.1905 0.5305 0.0006 0.00 079 0.9885 2 7,9 06 . 28 32,213.99 36,746.61

Knipolegus orenocensis sclateri 21 2239 1498 Dynamic Specialists 55 493.22 0. 2589 0.2678 0.4733 0.0009 0.00135 0.9878 24,088.68 28,454.01 33,36 8.78