Agonistic song rate positively correlates with male breeding success and avian malaria infection in Acrocephalus paludicola (Aquatic Warbler), a promiscuous songbird with female-only parental care

ABSTRACT

The link between male song and reproductive success has been explored at length in socially monogamous birds, but results were inconsistent and under-represented socially non-monogamous species with uniparental care. Here, we evaluated whether metrics of male song in Acrocephalus paludicola (Aquatic Warbler), a promiscuous songbird with female-only care and large repertoires, are associated with fitness- and quality-related traits. We showed that the number of 9-day-old nestlings fathered and probability of avian malaria infection increased with the rate of agonistic song, produced in male–male interactions, and that male scaled mass index (proxy for energy reserves) positively correlated with the repertoire size. The male breeding success was not explained by repertoire size and song duty cycle. None of the song variables clearly explained the male return rate or change in the tarsus length (proxy for structural body size) and scaled mass index of the fathered chicks between days 2 and 9 post-hatch. No other relationships between the song characteristics and the inbreeding coefficient, scaled mass index, wing-length and Trypanosoma infection status of the males were supported, and Plasmodium infection was unrelated to the song duty cycle and repertoire size. We conclude that in male A. paludicola the repertoire size could be a signal of early developmental stress or current body condition, and the agonistic song rate could be shaped by sexual selection and signal to females a genetic makeup that enhances survival upon infection by Plasmodium.

STRESZCZENIE

U samców ptaków zależność między śpiewem a sukcesem reprodukcyjnym badano szczegółowo u monogamistów socjalnych. Wyniki były jednak niespójne i rzadko badano gatunki niebędące monogamistami socjalnymi, u których występuje jednorodzicielska opieka nad potomstwem. W naszej pracy określiliśmy, czy u wodniczki Acrocephalus paludicola, promiskuitycznego gatunku ptaka śpiewającego o złożonej piosence, u którego potomstwem opiekuje się tylko samica, elementy śpiewu samców są powiązane z miarami ich dostosowania i jakości. Wykazaliśmy, że liczba spłodzonych 9-dniowych piskląt i prawdopodobieństwo wystąpienia ptasiej malarii wzrastały wraz z tempem produkcji piosenek odstraszających, śpiewanych w interakcjach z innymi samcami. Wskaźnik skalowanej masy ciała samców (określający ilość rezerw energetycznych) był dodatnio skorelowany z wielkością repertuaru. Sukces rozrodczy samców nie był wyjaśniony przez wielkość repertuaru ani względny czas trwania śpiewu. Żadna z badanych cech śpiewu nie była wyraźnie związana z powracalnością samców na lęgowiska, ani też zmianą długości skoku (określającego strukturalną wielkość ciała) i skalowanego wskaźnika masy ciała spłodzonych piskląt między 2. a 9. dniem od wyklucia. Nie stwierdziliśmy żadnych innych zależności między badanymi cechami śpiewu samców a ich współczynnikiem wsobności, skalowanym wskaźnikiem masy ciała, długością skrzydła ani też zakażeniem Trypanosoma. Zakażenie Plasmodium nie było powiązane z względnym czasem trwania śpiewu ani z wielkością repertuaru. Wyniki te wskazują, że u samców wodniczek wielkość repertuaru może nieść informację o poziomie stresu wczesnorozwojowego lub o ich bieżącej kondycji. Tempo śpiewu piosenek odstraszających może podlegać działaniu doboru płciowego i stanowić dla samic informację o posiadaniu przez samca genów zwiększających szanse przeżycia po zakażeniu Plasmodium.

INTRODUCTION

The evolution of the courtship song in male oscine birds has long been considered to be driven by sexual selection (Darwin 1871). It is predicted that song elaboration is adaptive and, ultimately, positively associated with fitness, through female preference and (or) effective competition with male rivals (Catchpole and Slater 2003). Song is highly variable among species, which, together with its multidimensionality, results in species-specific understanding of its complexity (Gil and Gahr 2002). However, there is a consensus that the most critical dimensions of signal elaboration are well reflected by the syllable repertoire size (Collins 2004). For example, in male Acrocephalus arundinaceus (Great Reed Warbler), the repertoire size correlates with the number of eggs laid by paired females, young produced, and offspring recruited to the population (Catchpole 1986, Hasselquist 1998). Larger repertoires are associated with faster mating or attracting more females for copulation, and siring extra-pair offspring with better post-fledging survival (Hasselquist et al. 1996). In Troglodytes aedon chilensis (House Wren), males with larger syllable diversity per song and larger song repertoires mate with females that initiate breeding earlier and lay more eggs (dos Santos et al. 2018). In Melospiza melodia melodia (Eastern Song Sparrow), the song repertoire size of males predicts their annual reproductive success (Potvin et al. 2015).

However, meta-analysis studies have indicated that the accumulated evidence is inconsistent. First, the effects of song complexity on female choice and subsequent male reproductive success identified to date are weak (Byers and Kroodsma 2009, Soma and Garamszegi 2011). Second, studies that measured them focused on male quality rather than on direct fitness components (Achorn and Rosenthal 2020). Eighteen studies of the latter type, conducted in the field, were identified by Soma and Garamszegi (2011), and we found further 2 published afterwards. Of these, only 4 used genetic methods to determine paternity (Woodgate et al. 2012, Potvin et al. 2015, dos Santos et al. 2018, Sung and Handford 2019), while genetic polyandry is known in 76% socially monogamous bird species (Brouwer and Griffith 2019). Third, the link between song and male reproductive success has been clearly understudied in non-monogamous species with female-only parental care. While in minority among birds (7%) (Cockburn 2006), they are apt models to study especially the indirect (i.e., genetic) benefits signaled by song elaboration, because song in these species can only advertise features unrelated to offspring care. Fourth, because a stronger positive association with reproductive success was found for visual plumage signals, compared to male courtship song (Soma and Garamszegi 2011), the relationship between reproductive success and song might be stronger in bird species without sexual plumage dichromatism. Finally, song is a multi-faceted signal (Gil and Gahr 2002, Rivera-Gutierrez et al. 2010) and apart from the song complexity, there might be other song features advertising the quality of the male, or increasing his attractiveness and intrasexual competitiveness, and thus affecting its reproductive success. For example, the rate of aggressive song is important for female guarding in polygynous or promiscuous species (Hasselquist and Bensch 1991). Song output was previously found to be costly in terms of energy reserves and foraging time, and to correlate with body condition, survival and pairing date, and hence it could signal male quality to females (Gottlander 1987, Radesäter et al. 1987, Saino et al. 1997, Nyström 1997, Welling et al. 1997, Garamszegi et al. 2004, Ritschard and Brumm 2012). Singing effort could also determine the speed of repertoire presentation.

The main proposed underlying mechanism of female preference for superior song performance is the Hamilton and Zuk (1982) hypothesis, which states that sexually selected traits advertise heritable resistance against parasites. In the light of this hypothesis, females mating with males showing better song performance secure resistance genes against infections for their progeny. In birds, special attention with respect to parasite-mediated expression of song was devoted to vector-borne haemoparasites (blood parasites), especially haemosporidians (genera Plasmodium, Haemoproteus, and Leucocytozoon), as these are among the most common parasites of birds (Valkiūnas 2005). Due to their pathogenic effects, blood parasites affect reproductive performance as well as survival of hosts (Marzal 2012). To date, however, the Hamilton and Zuk hypothesis has received mixed support, and the relationship between haemoparasite infection and male song has been rarely studied. For example, infection with Plasmodium parasites, which cause avian malaria, has been linked to reduced song output and song consistency, as well as production of simpler songs when infection occurs early in life (Spencer et al. 2005, Gilman et al. 2007), and male Acrocephalus schoenobaenus (Sedge Warbler) parasitized by at least one blood parasite of the genera Trypanosoma, Plasmodium, and Haemoproteus had smaller repertoire sizes (Buchanan et al. 1999). Conversely, Haemoproteus or Plasmodium infection was unrelated to the song repertoire size and song rate of male A. arundinaceus (Bensch et al. 2007, Asghar et al. 2015), and Plasmodium-infected Campylorhynchus rufinucha (Rufous-naped Wren) had higher song rate, but not repertoire size, compared to uninfected individuals (Meza-Montes et al. 2022). The inconsistent support for the Hamilton and Zuk hypothesis may be partly attributed to different roles of song traits with different signal designs (Garamszegi 2005). Taken together, our understanding of the relationship between blood parasite infection, song performance and female choice remains incomplete.

Acrocephalus paludicola (Aquatic Warbler) is a migratory songbird, breeding mainly in open marshland and wintering in sub-Saharan Western Africa (Le Nevé et al. 2018, Tanneberger et al. 2018). This passerine has an exceptional reproductive biology. There is no sexual plumage dichromatism and no pair-bonds. Nest-building, incubation, and offspring care are performed solely by the female. Both sexes are promiscuous, copulations are prolonged, and males have large testes, which implies high sperm competition (Schulze-Hagen et al. 1995, Garamszegi et al. 2005). Levels of multi-paternity are high, and the breeding success of males is highly asymmetric (Schulze-Hagen et al. 1993, 1995; Dyrcz et al. 2002). Males are characterized by high between-individual syllable diversity and large repertoires, with the average number of unique phrases in a male estimated at 155 (range: 18 to 300; Osiejuk and Kubacka 2023). The latter predicts a stronger correlation of song elaboration with reproductive success compared to songbird species with smaller repertoires (Robinson and Creanza 2019). Males sing courtship songs throughout the breeding season (Dyrcz and Zdunek 1993, Schmidt et al. 1999). They do not actively defend territories and their home ranges are labile (Schmidt et al. 1999, Schaefer et al. 2000). Thus, A. paludicola presents a particularly good model for testing hypotheses on the adaptive function of avian courtship song.

Here, we explored whether the repertoire size, percent time spent singing (song duty cycle), and agonistic song rate (used in aggressive male-male interactions) in A. paludicola males are associated with (1) their fitness-related parameters: seasonal breeding success, return rate and offspring growth, measured by the scaled mass index (a proxy for energy reserves) and tarsus size (a proxy for structural body size); and (2) their quality-related characteristics: scaled mass index, wing-length, inbreeding coefficient, and infection with the protozoan blood parasites Plasmodium and Trypanosoma. We predicted that the song features would be positively related to the fitness-related variables as a result of female preference, association with genes that are beneficial to offspring survival or a role in chasing away rivals and female-guarding. We further predicted that the song features would positively (scaled mass index, wing length) or negatively (inbreeding rate, infection with blood parasites) correlate with the quality-related parameters, which—if there is also a positive relationship between the song variables and the fitness-components—can be expected if the song traits convey information on the quality-traits to females. Alternatively, if song advertises male attractiveness but not the quality components to females, we expected to see a correlation between the song features and the male fitness components, but not between the song features and the male quality measures.

METHODS

Study Area and Sampling

The study was conducted between May and August of 2017-2023, in one of the core breeding sites of A. paludicola—the Biebrza Valley, Poland. This area holds ~25% (~2,700 singing males) of the global population. The habitat is permanently water-logged open fen mire, dominated by sedges Carex spp. Three study areas were used: Ławki, Szorce, and Mścichy (central positions: N 53°17ʹ11.4ʹʹ, E 22°33ʹ49.2ʹʹ; N 53°17ʹ34.8ʹʹ, E 22°37ʹ15.9ʹʹ; and N 53°25ʹ41.7ʹʹ E 22°30ʹ17.0ʹʹ, respectively), with two 10 to 20-ha study plots in each study area, totaling 70 ha. Habitat quality and male densities were comparable between Ławki and Szorce (fen mire); the study area in Mścichy was wet sedge meadow with male densities lower than in the other two study areas.

Data collected for the purpose of this study were part of a larger project (Kubacka et al. 2024a), in which, between 2017 and 2019, in total 174 adult males were caught with mist-nets. Each bird was marked with unique metal and color rings, and blood-sampled by puncturing the brachial vein with a sterile needle and collecting ~10 to 120 µL of blood with a capillary. Each blood sample was immediately stored in a vial with an o-ring seal containing 2 mL of 96% EtOH. A measurement of tarsus length (from the notch on the metatarsus to the top of the bone above the folded toes, with a caliper to the nearest 0.1 mm), wing length (maximum chord measurement, with a ruler to the nearest millimeter) and body mass (with an electronic Kern CM 150-1 N balance to the nearest 0.1 g) was taken. In 2018, between May 20 and June 8 and between June 28 and July 24, nests were located by search and observation of alarming females. The 20-day-long gap in nest search corresponded to the time between 2 breeding periods when few new nests are initiated (Dyrcz and Zdunek 1993, Dyrcz et al. 2018). The nest search was carried out every 2 to 6 days in each study plot, with the total average effort of 140 man-hours per 10 ha. The search effort was balanced between the plots. An attempt was made to find most nests within the study plots, and we assumed that this is indicated by not finding new females with potential nests (i.e., making alarming calls suggestive of a nearby nest with eggs or chicks), and the flattening of the cumulative number of nests found over time (Supplementary Material Figure 1). We found a total of 44 nests, and we missed ~6 nests of females that we had detected but not followed further. Twenty-eight nests were found at the egg stage or days 1 to 2 post-hatch, and the remaining sixteen nests were found at a later nestling stage.

We determined the age of chicks as the number of days post-hatch, with day 0 being the last day when most progeny was in the egg stage. Nests found at the egg stage were monitored every 2 days to pinpoint the hatching date (day 1; i.e., when most progeny had hatched). Chicks in nests found at the nestling stage were aged using the key by Wawrzyniak and Sohns (1977). In total, 197 chicks hatched from the detected nests. The tarsus length and body mass of chicks (n = 116) were measured as in adult males on days 2 and 9 post-hatch. On days 7 to 9 chicks (n = 124) were blood sampled through brachial venipuncture, as above. Females (n = 32) were mist-netted by nests and blood-sampled likewise. Approximately 10 to 80 and 10 to 120 μL of blood was drawn from a chick and a female, respectively. Seventy-three chicks were not blood-sampled and/or measured both on days 2 and 9 due to chick mortality and nest predation (22 chicks), finding a nest after day 9 (i.e., too late to avoid force-fledging when handling [28 chicks]) and other reasons (23 chicks). Because in A. paludicola fathers do not tend nests, the broods were monitored, and the chicks were measured and blood-sampled without knowledge of the identity of their fathers.

Between May and July 2018 to 2023, systematic search of the study plots and their surroundings (up to 150 m off plot border) was conducted to re-find the males that were color-ringed. The search effort was ~35 man-hours per 10 ha. The code of the color rings was identified with an 80-mm-lens and 20 to 60 magnification scope, and the geographic position of each resighted bird was stored using a GPS receiver. In addition, the color-ringed males recaught in the study areas during mist-netting between 2018 and 2019 were included as resights. The resights and recaptures were recorded without knowledge of the analyzed traits of the males.

Song Recording and Analysis

Acrocephalus paludicola males produce relatively short, repetitive songs (Figure 1; Catchpole and Leisler 1989, 1996). Structurally, they consist of rattles and whistles. Functionally, they can be categorized into 3 types: short A-songs, consisting of a single rattle phrase; B-songs, initiated by a rattle phrase and followed by a slower, more complex whistle phrase; and C-songs, formed by more than two phrases, combining both rattles and whistles. As shown in playback experiments, A- and B-songs are produced in male–male interactions, and are used for close and aggressive (A-songs) and more distant (B-songs) communication, whereas C-songs attract females (Catchpole and Leisler 1996). A-songs were found to be produced on playback of other songs, especially A-songs of other males; in spontaneous song, B- and C-songs prevailed (Catchpole and Leisler 1989, Schmidt et al. 1999).

The song sampling and sound analysis are described elsewhere (Osiejuk and Kubacka 2023). Briefly, songs of individually marked A. paludicola males were recorded between 16:15 and 21:20 hours from 28 May to 29 July in 2018, using a Marantz PMD 661 digital recorder and a Telinga Pro 7 microphone mounted in a Telinga Universal parabola, as 48 kHz / 16-bit PCM WAV files. We obtained recordings from 45 males but discarded 5 individuals for which fewer than 10 phrases were recorded, and 1 individual that could not be identified. We obtained (mean ± SD) 703 ± 278.1 s of recording per male, with 235 ± 113.7 phrases per individual. Because A. paludicola males can present their repertoire efficiently in a few C-songs, and we used the estimated repertoire size (mean: 155, range: 18 to 300) rather than the enumerated repertoire size (mean: 99, range: 16 to 158, as determined in the same individuals; Osiejuk and Kubacka 2023), we believe that the obtained recording times suffice as a biologically relevant sample of the repertoire size. Collecting longer recordings while maintaining a fair sample of individuals is challenging in A. paludicola, given the remote and physically demanding breeding habitat and a narrow time window of peak singing intensity (Dyrcz and Zdunek 1993). The songs of each individual were manually segmented into phrases (Catchpole and Leisler 1989, 1996) in Raven Pro 1.6 (K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology 2022). Each phrase was assigned to 1 of the 3 categories: whistle, rattle, or rattle-whistle, based on visual and auditory inspection. Classification of the 3 phrase categories into discrete types (comprising the repertoire) was performed with the Koe software (Fukuzawa et al. 2020). The recordings of each male were characterized, among other, by duration (in seconds) and by unique types within these categories. To estimate the repertoire size, we used measures derived from the species diversity ecology (Magurran and McGill 2011), applying the SpadeR (Species-Richness Prediction and Diversity Estimation with R) software (Chao et al. 2016), and the Chao1 estimator. We used the Chao1 estimator and estimation, rather than enumeration of the number of unique phrases, because we earlier showed that even the longest recordings did not reveal the entire repertoire, and that out of the estimators available in SpadeR, Chao1 and ACE yielded very similar repertoire size estimates (Osiejuk and Kubacka 2023). In addition, we also categorized the recordings of each male into the A-, B-, and C-songs. We then calculated the frequency of A-song as a ratio of their number and the total singing time (i.e., time in seconds from the start of the first syllable to the end of the last syllable in a recording, excluding periods when the male was not producing any sound), multiplied by 60 to obtain the number of songs per minute of singing time. Finally, we assessed the song duty cycle, dividing the total singing time of each male by the duration of its recording(s).

We selected the above 3 song characteristics to seek relationships with the fitness- and quality-related variables because they represent different functional aspects of A. paludicola song. The repertoire size is a commonly used proxy for song complexity, which in some avian species—especially those with large repertoires—was found to predict male fitness- and quality-components (Byers and Kroodsma 2009, Soma and Garamszegi 2011, Robinson and Creanza 2019). The A-song rate is a proxy for the rate of agonistic songs produced by A. paludicola males on approach of other males (see above) and could be related to female guarding (Catchpole and Leisler 1989, Hasselquist and Bensch 1991, Schmidt et al. 1999). The song duty cycle might be associated with the energy costs of singing and thus the current condition of the male (Gottlander 1987, Saino et al. 1997, Nyström 1997, Garamszegi et al. 2004, Ritschard and Brumm 2012). The repertoire size, song duty cycle, and A-song rate were not mutually correlated (Pearson correlation coefficient r: −0.15 to 0.08, P >> 0.05 in all cases), and did not vary between the 3 study areas (ANOVA, P: 0.63 to 0.91) nor the breeding peak within the season (ANOVA, P: 0.14 to 0.90). The song analysis was performed blind to the breeding success and quality characteristics of the males, as well as to the characteristics of their offspring.

RAD-seq Library Preparation and Bioinformatic Analysis

As genetic markers, we used single nucleotide polymorphisms (SNPs), sampled with 3-enzyme restriction site-associated DNA sequencing (i.e., 3RAD; Bayona-Vásquez et al. 2019, Glenn et al. 2019). We followed the methods applied in the larger study (Kubacka et al. 2024a). Briefly, DNA was isolated with the Xpure Blood mini kit (A&A Biotechnology, Poland). Using 3 female samples with high DNA concentration, we constructed a reference genomic library following the Reduced-Representation Single-Copy Orthologs (R2SCOs) method as described by Driller et al. (2021), with modifications, using the restriction enzymes Xbal, EcoRi-HF, and Nhel. The library was sequenced with the Illumina MiSeq V3 at 600 cycles and achieved a total of 3.6 M reads per sample. We then constructed 3RAD “population” libraries (Bayona-Vásquez et al. 2019), following the protocol described above, consisting of all the samples collected in the other project (Kubacka et al. 2024a). The population libraries (9 in total, prepared in 3 batches) were sequenced with llumina NextSeq MidOutput 300 cycles and Illumina NovaSeq 6000 300 cycles with 15% of PhiX.

The raw sequencing data were processed as described elsewhere (Kubacka et al. 2024a). In short, we first removed PhiX genome reads, trimmed the adapters with the Cutadapt software (Martin 2011), and demultiplexed the sequences with FLEXBAR (Dodt et al. 2012). In 3 of the libraries, we filtered out polymerase chain reaction (PCR) duplicates using a custom Python script. We removed short fragments (< 249 base pairs [bp]) with the software PEAR (Zhang et al. 2014), and filtered the read pairs for minimum quality (Q > 30) and length (> 130 bp) using Trimmomatic (Bolger et al. 2014). We checked the presence of uncut restriction sites in the paired-read ends and filtered out fragments resulting from star activity or NheI digestion, as well as reads containing the internal restriction site of XbaI, EcoRI, or NheI, using custom Python scripts. The filtered reads were mapped against the reference locus catalog built using the R2SCOs method described by Driller et al. (2021), based on the 2 × 300 bp reads of the three reference samples. We followed the R2SCOs pipeline, including the de-replication of the preprocessed and size-selected (250 to 370 bp) sequences, requiring the minimum coverage per unique sequence equal to 3 to define a putative allele, and the clustering and definition of putative loci using the intra-specific identity threshold of 90%. We removed all the R2SCOs loci that mapped to the sex chromosomes of the Sylvia atricapilla (Eurasian Blackcap) genome (GenBank accession number GCA_009819655.1). The final reference contained 10,997 loci. For the mapping, we used the software Bowtie2 (Langmead et al. 2009) with default parameters and the flags -no-mixed and -no-discordant. We selected the range of 250 to 310 bp, which was a common range of high-coverage loci for all individuals belonging to the different libraries, and extracted it from the .sam files using SAMtools (Li et al. 2009) and an awk command.

The mapped reads were then analyzed with Stacks (Catchen et al. 2011, 2013), using the reference-based pipeline. Genotypes were called and filtered with the ref_map.pl program. In the gstacks module, we set var_alpha = 0.01, assumed all the samples to form one population and retained the remaining parameters at their defaults. In the populations module, we used r = 0.8, p = 1, min-maf = 0.005 and the “write single SNP” option, and retained the remaining parameters at their defaults. Next, with VCFtools (Danecek et al. 2011), we removed 3 individuals with mean depth < 10, genotypes with depth lower than 10 reads, and SNPs with mean depth > 115 as well as loci with more than 20% missing data. We then inspected whether the loci conform to the Hardy-Weinberg (H-W) proportions using the dartR package (Gruber et al. 2018) in the R environment (R Core Team 2023). Most of the loci that were not in H-W equilibrium showed heterozygote deficiency, and there was a positive correlation between the locus F_is and locus missingness (r = 0.38), which decreased to r = 0.13 after removal of the loci that did not conform to H-W proportions (applying the false discovery rate correction and the P-value of 0.05). As this suggests the presence of null alleles, which artefactually decrease heterozygosity (Waples 2015, De Meeûs 2018), we removed the SNPs that were not in H-W equilibrium. Finally, using VCFtools, we calculated loci-pairwise r² to inspect distribution of linkage disequilibrium. Mean ± SD r² was 0.003 ± 0.012, range: 0.000 to 1. We excluded one randomly selected SNP from each pair for which r² was >0.3. The number of SNPs after filtering was 2,948.

Establishing Reproductive Success

We followed the methods applied in the larger study (Kubacka et al. 2024a). Briefly, we selected a set of informative SNPs (Andrews et al. 2018, Thrasher et al. 2018), by filtering out SNPs with minor allele frequency (MAF) below 0.3 and missing data exceeding 0.1 per SNP using VCFtools (Danecek et al. 2011), which left 333 SNPs. A high MAF increases accuracy of parentage assignments (Dussault and Boulding 2018). We then assigned paternity with the Cervus program (Kalinowski et al. 2007), running the allele frequency analysis, parentage analysis simulation and paternity analysis given known mother. We ran Cervus on the full dataset (Kubacka et al. 2024a), of which the song-recorded males were part. In the parentage analysis simulation, we assumed 100,000 offspring, 440 as the average number of candidate fathers per offspring, 0.4 as the proportion of candidate fathers sampled and 88 as the minimum number of typed loci. We selected Delta as the statistic to determine confidence, and 99% and 95% as the strict and relaxed confidence levels, respectively. In 2018 (the year of recording), 98 chicks (81%) were assigned a father, and 30 males (31%) were assigned offspring, with their breeding success not varying between the 3 study areas (linear model, P = 0.21). Among the males with song recordings, 13/39 (33%) were assigned as fathers of 49 nestlings in 2018, with both the pair (i.e., father–offspring) and the trio (i.e., father–mother–offspring) confidence of 99%, except for one case in which the mother was not sampled and trio confidence could not be determined. We did not include one assignment, in which the trio confidence was 99%, but the pair confidence did not pass 95%. This assignment was made between a male and a nest that were > 3 km apart. The breeding success among the song-recorded males ranged from 0 to 12, and had a mean of 1.26 and median of 0.

The low father assignment rate could have been caused by the core areas of home ranges of some males being far from the nearest genotyped nests. To assess this error, using QGIS v. 3.10.7 we determined distances of the males from the nests in which they were assigned young, based on their averaged coordinates of captures and subsequent resighting; and distances of the males that were not assigned offspring to the nearest genotyped nest. We used only the 2018 capture and resighting data for this purpose. A male that was assigned young resided on (average ± SD) 153.2 ± 73.1 m (range: 47 to 341) from the nest(s) with the young. The mean ± SD distance to the nearest genotyped nest for the males that were not assigned offspring was 161.8 ± 110.8 m (range: 31 to 392). In addition, based on all the records of the males in our study plots (i.e., captures and resights), we calculated the number of encounters of each male within a plot over the whole breeding season. The total number of encounters in 2018 was 188, of which 9 (5%) were recaptures and the remaining ones were resights during song-recording or search of color-marked males. Males with a high number of encounters were present in and around a study plot for most of the breeding season; and those with a low number of encounters were first captured during the second breeding attempt, or were rarely seen in the study plots. The number of encounters ranged from 2 to 9, with a mean of 4.8 and a median of 5. We used both the distance to nearest genotyped nest and number of encounters to differentiate between males that were likely to sire offspring in the study plots from those that were less likely to do so (see Statistical Analysis).

Molecular Detection of Haemoparasitic Infections

The study focused on infection status of the A. paludicola males with vector-borne blood parasites from two genera: Plasmodium, representing the order Haemosporida, and Trypanosoma, representing the order Trypanosomatida (Baker 1976, Valkiūnas 2005). Their vectors are dipteran haematophagous insects, and infection arises following a bite (haemosporidians) or either digestion or contamination of conjunctiva/skin with infective parasite stages (trypanosomes). We focused only on Plasmodium parasites among haemosporidians because previous studies on A. paludicola, sampled in the Biebrza Valley as well as in other locations, showed that they do not carry Haemoproteus infections and get infected with Leucocytozoon only occasionally (Neto et al. 2015, Kubacka et al. 2019). Although Haemoproteus parasites were outside the focus of the study, the protocol we used allowed for detection of parasites from this genus (see below).

To determine infection, only samples collected in the year of song recording (2018) were used (n = 28). The concentration of DNA was adjusted to 25 ng μL⁻¹. The samples were then tested for the presence of haemosporidian parasites using a nested PCR, which allows for the detection of parasites from genera Haemoproteus, Plasmodium, and Leucocytozoon based on the amplification of a fragment of the cytochrome b (Hellgren et al. 2004). PCR reactions that used isolates as a source of DNA contained ~50 ng of total genomic DNA. In the second run of the nested PCR, we used only primers specific for Haemoproteus and Plasmodium (HaemF and HaemR2), which amplify a 478-bp-long fragment. Thermal profiles followed Hellgren et al. (2004), and the PCR reaction mixture was prepared according to Kubacka et al. (2019), except for the number of Taq DNA polymerase units (0.625 U per 25 μL as recommended by the manufacturer, EURx, Gdańsk, Poland) and taking 1 μL of the PCR product from the first run as a source of DNA, as recommended by Hellgren et al. (2004).

The presence of Trypanosoma infections was confirmed with a nested PCR targeting a 326-bp-long-fragment of the 18 S rRNA gene (Sehgal et al. 2001). Thermal profiles followed Sehgal et al. (2001) and the PCR reaction mixture was prepared according to Kubacka et al. (2019), except for the number of Taq DNA polymerase units (as described above) and the volume of the PCR product from the first run used in the second run [1 μL as recommended by Sehgal et al. (2001)]. The samples were tested with each protocol twice. The ones which produced contradictory results in the 2 runs were tested for a third time, and the dominant result was treated as final. To check for contamination, a negative control (ddH₂0 instead of DNA isolate) was used, while to monitor for a possible PCR failure, a positive control (DNA isolated from a Parus major [Great Tit] infected with Plasmodium, Haemoproteus, Leucocytozoon, and Trypanosoma) was used. To check the quality of DNA isolates for samples scored as negative, we ran molecular sex identification with primers P2 and P8, which target the fragments of the chromo-helicase-DNA-binding (CHD) genes located on avian sex chromosomes (Griffiths et al. 1998). Each 10 μL PCR reaction mixture contained 1× buffer, 1.5 mM MgCl₂, 0.2 mM dNTP, 0.5 μM of each primer, and 0.5 units of Taq DNA polymerase (Eurx, Poland). Thermal profile followed Cichoń et al. (2003) except for the annealing temperature, which was set at 50 °C following Jakubas et al. (2014). Amplicons (8 μL) were run on either 2% (screening for parasites) or 3% (testing for DNA isolate quality) agarose gels stained with SimplySafe (Eurx, Poland) and visualized under UV light.

To identify haemosporidian infections to the genus level, positive amplicons were sequenced. Sequencing was performed unidirectionally with the forward primer HaemF, except for the 2 sequences that were identified as novel for the study species, and which were additionally sequenced with the reverse primer HaemR2. Amplicons were cleaned enzymatically (Exo-sap) and Sanger-sequenced by Genomed (Warsaw, Poland). Inspection of chromatograms for the presence of multiple peaks, which indicate mixed infections, sequence annotation and alignment were performed with the BioEdit software v. 7.2.3 (Hall 1999). The genus of haemosporidians was identified based on the assignment of the sequences to the lineage level using BLAST (Basic Local Alignment Search Tool) function implemented in the MalAvi database (Altschul et al. 1990, Bensch et al. 2009). Infection status was denoted as 0 = not infected and 1 = infected. Infection status did not vary by blood-sampling date (GLM with binomial distribution and log-link; P = 0.87 for Plasmodium and P = 0.32 for Trypanosoma).

Quality- and Fitness-related Metrics

In the recorded males, we measured the following variables, which we refer to as “quality-related” throughout: scaled mass index, wing length and inbreeding rate. We determined the scaled mass index using a custom script (Kubacka et al. 2024b) in the R environment v. 4.1.3 (R Core Team 2023). The scaled mass index corresponds to the relative amount of energy reserves, and is calculated using the body mass and a measure of body length (Peig and Green 2009), which was the tarsus length in our case. The scaled mass index correlates with the amount of subcutaneous fat in passerines (Peig and Green 2009, Nip et al. 2018). The body mass of the males did not change over the breeding season (general linear mixed model, GLMM, with male ID as random intercept and year and study area as fixed effects, P = 0.74). In some migratory passerines, wing length is positively associated with the speed of migration and earlier time of arrival at breeding grounds (Stolt and Fransson 1995, Risely et al. 2013, Hahn et al. 2016). In A. paludicola, male wing length was found to positively correlate with the number of young produced (Dyrcz et al. 2005). Hence, the wing length could be a signal of male quality to females. In our study, the males ringed later in the season had shorter wings compared to the males ringed earlier in the season (GLMM with male ID as random intercept and year and study area as fixed effects, P < 0.01). The inbreeding rate was estimated with the inbreeding coefficient F (a method of moments), using VCFtools (Danecek et al. 2011) (the –het option), based on all the SNPs that remained after filtering (i.e., 2,948). In the song-recorded males, the F coefficient tended to be lower as the breeding season progressed (linear model, P = 0.08). Using the R package inbreedR (Stoffel et al. 2016), we calculated an estimate of identity disequilibrium, g₂ (David et al. 2007), to assess variance in inbreeding, and hence the chances to detect a correlation between song and inbreeding. When individual inbreeding cannot be determined from a pedigree, identity disequilibrium can be used to describe the distribution of inbreeding in a population (David et al. 2007). A correlation between the trait in question and heterozygosity cannot arise when g₂ = 0; however, non-significant values of g₂ do not exclude that it is observed (Szulkin et al. 2010, Miller and Coltman 2014). We used 1,000 permutations (over SNPs) to calculate whether g₂ is greater than zero and 1,000 bootstraps (over individuals) to estimate its 95% confidence interval.

Furthermore, for the males we measured the following variables, which we refer to as “fitness-related” throughout: rate of return to breeding grounds, seasonal breeding success, and growth of nestlings that they sired. The rate of return (a proxy for survival) was quantified as the number of years during which the male was resighted or recaught in or around a study plot during 5 consecutive years following the recording year (i.e., 2019 to 2023). Thus, a male could survive 0 to 5 years. Among the 39 song-recorded males included in the analysis, the mean ± SD return rate was 0.8 ± 1.1 years (median: 0, range: 0 to 5). We used the number of 9-day-old offspring assigned to a male as a measure of his seasonal breeding success.

In chicks, we measured growth by the change in tarsus length (a proxy for the structural body size) and scaled mass index between days 2 and 9 post-hatch. This period corresponds to the beginning of the nestling period and ~5 days before fledging, which is the time of the fastest and approximately linear body mass growth in A. paludicola nestlings (Vergeichik 2007). In A. paludicola, chick age can be determined to the nearest day for nests found on days 1 to 2 post-hatch, and less accurately (±1 day) for nests found afterward (J. Kubacka, personal communication). Therefore, only the nestlings from nests found at the egg stage or 1 to 2 days post-hatch were used to relate chick growth to the song variables.

Statistical Analysis

We performed the analysis in the R environment v. 4.1.3 (R Core Team 2023). We used the information-theoretic approach, in which, unlike in hypothesis testing, an a priori set of biologically plausible candidate models is constructed and each model is assigned a relative rank corresponding to how well it explains the data (Burnham and Anderson 2002). For model ranking, we used the Akaike Information Criterion corrected for small sample size (AIC_c). We also provided quantitative measures of relative support for each model: model likelihood (relative likelihood of the model in the candidate set, given the data), Akaike weight (w_i) (probability that the given model is the best approximating model in the candidate set) and evidence ratio (describing how much more likely the best model is than the given model), as well as cumulative AIC_c (i.e., sum of AIC_c values of a given model and all the higher-ranking models; Symonds and Moussalli 2011). We constructed candidate model sets using the all-subsets option (i.e., dredging), since all the combinations of the explanatory variables in each set were biologically plausible. Each candidate set included a null model (i.e., one assuming a constant response). For selected candidate sets, we limited the maximum number of variables in a model as specified below, to allow ~10 data points per estimate and reduce the risk of over-parametrization (Harrell 2015). When calculating model estimates, to account for model uncertainty, we applied natural model-averaging (i.e., averaging estimates and their confidence intervals over all the models in a set that contained the estimated variables), because we expected small effects and aimed at evaluating effects of specific song variables (Symonds and Moussalli 2011, Grueber et al. 2011). Model selection and averaging were carried out with the package MuMIn (Bartoń 2019). In the models in candidate sets (1) and (2) (see the following paragraphs in this section) all the explanatory variables were standardized to a mean of zero and a standard deviation of one (i.e., the z-score transformation) using the scale function from the base package. In the models in candidate sets (3) and (4) both the explanatory and the response variables were z-score transformed, to allow comparison of estimates between the 2 candidate sets. In the models in candidate sets (5), (6), and (7) the numeric response and explanatory variables were standardized by 2 SD to allow comparison with categorical variables, and the categorical variables were centered, using the standardize function from the arm package (Gelman 2008, Gelman and Su 2022). For each predictor, we determined its relative importance (i.e., the sum of Akaike weights across all the models considered in a candidate set that contained this predictor; Symonds and Moussalli 2011).

To evaluate the strength of the relationship between the number of 9-day-old offspring (response variable) and male song features, we built candidate model set (1). Because the number of 9-day-old young had a strongly right-skewed distribution and about two-thirds of the males were not found to produce young in our study areas, we followed recommendations by Blasco-Moreno et al. (2019) and Campbell (2021) to select the most appropriate count model for our data, considering a Poisson (P), a negative-binomial (NB), as well as a zero-inflated model with the Poisson (ZIP) and the negative-binomial (ZINB) distribution in the count model. First, for the number of 9-day-old young, we calculated the index of dispersion (i.e., variance to mean ratio, VMR) and the index of zero-inflation (ZI), both with respect to the Poisson distribution (Puig and Valero 2006), along with their bootstrapped confidence intervals determined with the boot and boot.ci functions from the boot package (Davison and Hinkley 1997, Canty and Ripley 2024) and the “basic” option. The VMR was 4.89 (95% CIs: 2.48 to 7.48), which suggested overdispersion in the offspring number variable, and the ZI was 0.68 (95% CIs: 0.55 to 1.07), which indicated zero-inflation. We then constructed “full” P, NB, ZIP, and ZINB models with the explanatory variables in each model being the 3 song variables, distance to the nearest genotyped nest and number of encounters. None of the song variables was correlated with the number of encounters (repertoire size, Pearson correlation coefficient r = 0.12, P = 0.47; A-song rate, r = 0.18, P = 0.27; song duty cycle, r = −0.12, P = 0.48). We fitted the P models with the glm function from the stats package, and the NB models with the glm.nb function from the MASS package (Venables and Ripley 2002). We fitted the ZIP and ZINB models using the zeroinfl function from the countreg package (Zeileis et al. 2008, Zeileis and Kleiber 2022). A zero-inflated model is a mixture model that assumes excess zeros in the data and allows the zeros to be of two sources. It consists of a binomial model with logit-link (modelling the point mass zeros) and a count model with log-link (modelling a response that can take values of zero and above) (Zeileis et al. 2008, Green 2021). In our case, the binomial model (i.e., the zero model) estimated the probability of excess zeros (e.g., a male not producing 9-day-old young in our study plots because it arrived in a study plot in the second half of the season and thus sired very few or no young). The count model estimated the number of 9-day-old young produced in the study plots, allowing for no young (e.g., due to low sperm quantity or quality, post-copulatory female choice or inefficient female guarding). The zero and the count model of a zero-inflated model can have different predictors. In the zero part of each ZIP/ZINB model, we included the number of encounters and distance to the nearest genotyped nest, as we expected these variables to explain excess zeros; and in the count part we considered the number of encounters and distance to the nearest genotyped nest, as well as the three song variables. To decide on the model type (i.e., P, NB, ZIP, or ZINB), we ranked the “full” models using AIC_c, AIC, and BIC. With low sample sizes, this approach is recommended over choosing the best-model type based on score tests for overdispersion and zero-inflation (Campbell 2021). The zero-inflated models obtained the highest support, with the ZIP ranking first for AIC_c, AIC, and BIC (Supplementary Material Table 1). Therefore, we proceeded with the ZIP model structure to construct candidate set (1). We limited the maximum number of explanatory variables in a model to 2, and hence set (1) consisted of 29 models (Supplementary Material Table 2).

To evaluate the relationship between survival of a male over the 5-year-period (response variable) and his song features, we built candidate set (2), which used GLMs with Poisson error structure and log link, fitted with the glm function from the stats package. The explanatory variables were the 3 song features and hence candidate set (2) consisted of 8 models (Supplementary Material Table 3).

To explore how well the song variables predict chick growth-related metrics, we constructed candidate model sets (3) and (4), with the response variable being change in chick tarsus length and change in chick scaled mass index between days 2 and 9 post-hatch, respectively. For these candidate sets, we used GLMMs built using the lmer function from the lme4 package (Bates et al. 2015) and fitted by REML, with mother ID and father ID as random intercepts. Models in candidate sets (3) and (4) included a combination of the repertoire size, A-song rate, song duty cycle and brood size on day 9 as explanatory variables. We allowed not more than 2 explanatory variables in a model. Hence, each of the candidate sets (3) and (4) consisted of 11 models (Supplementary Material Tables 4 and 5).

To explore the relationship between the song features and the individual quality measures: scaled mass index, wing-length, inbreeding coefficient, and infection by the blood parasites Plasmodium sp. and Trypanosoma sp., we constructed candidate model sets (5), (6), and (7), with the response being repertoire size, song duty cycle, and A-song rate, respectively. We included only the males that were ringed and blood-sampled in 2018 (n = 27). These candidate sets consisted of linear models fitted with the lm function from the stats package. We limited the number of predictors in each model to 2, and hence each set comprised 16 models (Supplementary Material Table 6).

We used the packages AICcmodavg (Mazerolle 2023) and emmeans (Lenth 2023) for prediction, and ggplot2 (Wickham 2016) to visualize results.

RESULTS

In candidate set (1), the model assuming that male breeding success varied with the number of encounters (in the zero model) and A-song rate (in the count model) obtained very high support (w_i = 0.91, Supplementary Material Table 2). Accordingly, the relative importance of these 2 variables was very high (Table 1A). This indicates that the number of encounters is a strong predictor of the chances to detect a male producing 9-day-old chicks in a study plot, and that the A-song rate is the strongest predictor, out of the 3 song variables, of the number of 9-day-old young produced after accounting for excess zeros. Males observed more frequently in our study plots were more likely to produce at least one 9-day-old chick in the plots (Table 1A, Figure 2A). With one standard deviation increase in A-song rate, the number of young increased by ~50% (Table 1A, Figure 2B).

In candidate set (2), no model received higher support than the null model, and the strength of evidence was spread over several models (Supplementary Material Table 3). The A-song rate obtained the highest relative importance of 0.43; however, the back-transformed confidence intervals of all the song variables spanned one (Table 1B).

In candidate set (3), the models assuming that the change in chick tarsus length varied with A-song rate and song duty cycle (w_i = 0.18); and with only the brood size (w_i = 0.16) obtained higher support than the null model (w_i = 0.12). However, the strength of evidence was spread over a number of models (Supplementary Material Table 4). The A-song rate and brood size had the highest relative importance (0.41 to 0.42), with positive and negative effect on chick tarsus growth, respectively. However, their confidence intervals spanned zero and hence these effects were very weak. The confidence intervals of the other variables also spanned zero (Table 2A). In candidate set (4) the model assuming that chick scaled mass index varied with the A-song rate and song duty obtained the highest support (w_i = 0.28), but it was only slightly better than the null model (w_i = 0.20; Supplementary Material Table 5). The A-song rate had the highest relative importance (0.49) and a negative effect on chick scaled mass index. However, the confidence interval of both the A-song rate and the other predictors spanned zero (Table 2B).

The inbreeding coefficient F had a mean ± SD of 0.013 ± 0.041 and its range was −0.08 to 0.10. The estimated g₂ was low (0.0008, 95% CIs: −0.0001 to 0.002) but significantly different from zero (P = 0.001). We did not detect any Haemoproteus infections in the males. The prevalence of infection was 36% (14/39) for Plasmodium and 23% (9/39) for Trypanosoma. In candidate set (5), the model assuming that repertoire size varies with the scaled mass index obtained the highest support (w_i = 0.32; Supplementary Material Table 6A). The scaled mass index was an important predictor of the repertoire size (relative importance: 0.72), and with one standard deviation increase in scaled mass index the repertoire size increased by 0.42 standard deviation (Table 3, Figure 3A). In candidate set (6), no model ranked better than the null model (w_i = 0.24; Supplementary Material Table 6B). The scaled mass index obtained the highest relative importance of the examined predictors with a negative effect on the song duty cycle, however, the confidence intervals of all the variables spanned zero (Table 3). In candidate set (7), the model assuming that the A-song rate varies with Plasmodium infection and scaled mass index ranked first (w_i = 0.43) and was 12 times more probable than the null model (Supplementary Material Table 6C). The relative importance of Plasmodium infection was high (0.90; Table 3, Supplementary Material Table 6C). Males infected with Plasmodium sang on average 0.58 SD more A-songs per minute than uninfected males (Table 3, Figure 3B). The scaled mass index ranked second in terms of importance (0.46), with a negative effect but the confidence interval spanning zero.

DISCUSSION

We showed that the frequency of agonistic songs (A-songs) is positively associated with the seasonal breeding success in A. paludicola males and is its most important predictor of the three song parameters that we studied. The A-song rate is also positively related to avian malaria infection, and the estimated repertoire size—a proxy for song complexity—is positively correlated with male body condition. In contrast, the repertoire size and the song duty cycle—a proxy for singing effort—are unlikely to predict the seasonal breeding success in A. paludicola males. Likewise, we failed to find conclusive evidence that the return rate of A. paludicola males and the quality of their offspring, or the remaining male quality traits are associated with the studied song features.

Given the high sperm competition inferred in A. paludicola (Schulze-Hagen et al. 1995), males are expected to compete for copulations and to guard females after copulation, by preventing them from copulating with other males and deterring other males. It was hypothesized that the former is performed through frequent and extraordinarily long copulations (averaging 24 min in captivity), with several cloacal contacts (Schulze-Hagen et al. 1995). Deterring rivals is believed to be achieved with A-songs, which are produced in aggressive male–male interactions (Catchpole and Leisler 1989, Schmidt et al. 1999). Hence, in A. paludicola males, a higher frequency of A-songs could translate to higher efficiency in chasing away rivals, more mating opportunities and thus higher effectiveness of post-copulation female-guarding. Alternatively, increased aggressive song rate could be associated with older age in male A. paludicola. In passerine birds, song characteristics are known to differ between age classes (Kipper and Kiefer 2010) and older males have a higher reproductive success (Forslund and Pärt 1995). A. paludicola males of moderate age (i.e., past the first calendar but not very old, since both fecundity and song deteriorate due to senescence; Berg et al. 2020) could both have increased A-song rate and be more efficient at siring young, for example thanks to higher sperm production (Laskemoen et al. 2008) or more experience (Geslin et al. 2004), and thus improved female-guarding. Finally, female A. paludicola could prefer older (i.e., of moderate age) individuals and use the A-song rate as a cue for age during mate choice (Freeman-Gallant and Taff 2018). In sum, both causal and correlative mechanisms could explain the positive relationship between the agonistic song rate of A. paludicola males and their seasonal breeding success, and the presented hypotheses remain to be tested in the future.

Because we obtained a low offspring assignment rate to males (33%), we might have not determined their absolute breeding success. This is not likely to result from errors in the paternity assignment analysis, as we used a high number of informative SNP markers, and the parentage assignment results were of high confidence. Neither was the low offspring-to-father assignment rate caused by a male being too far from the nearest genotyped nest, since the distance to the nearest genotyped nest was not a good predictor of the breeding success, and hence males had similar chances to produce young irrespective of whether they were far or close to a nest that we genotyped. Skewed and highly variable reproductive success was found earlier in A. paludicola (Dyrcz et al. 2002, 2005) and in another female-only care non-monogamous songbird, Ammospiza caudacuta (Saltmarsh Sparrow), in which low offspring-to-father assignment rate was also observed (17 to 42%; Hill et al. 2010, Walsh et al. 2018). This suggests a natural pattern in our study, which could partly stem from imperfect detection of young. Although we put relatively high effort into finding most nests in the study plots, we could have missed some nests because of incomplete female detection and the break in nest-search between the two breeding periods. We were also unable to blood-sample all the chicks in the detected nests due to chick mortality, nest predation, too late nest finding and other reasons. In addition, some males might have arrived to our study area for the second breeding attempt only, which has generally been observed in the Biebrza Valley (Grzywaczewski et al. 2014, Kubacka et al. 2014), or they could have sired young outside the study plots. These last two factors are consistent with the probability of a male to produce young within the plots being well explained by its number of encounters in the study plots. The number of encounters thus likely explained variation in male breeding success due to a male arriving late in the study area or visiting it only occasionally. However, and importantly, the number of encounters did not correlate with any of the song characteristics that we studied. We therefore think that the error in breeding success measurement is unlikely to introduce a bias in our study, because it will have comparable effects on the offspring number of all the males, regardless of their song traits.

In Acrocephalus warblers, song complexity was shown to be important for female choice (Hasselquist et al. 1996, Buchanan and Catchpole 2000, Bell et al. 2004, Marshall et al. 2007). Females prefer males with larger repertoires as social mates (e.g., A. schoenobaenus) or for extra-pair mating (e.g., A. arundinaceus), and males with a larger repertoire produce more and fitter offspring (Hasselquist et al. 1996). While the large repertoire size of A. paludicola (Osiejuk and Kubacka 2023) justifies an expectation that it should correlate with fitness-related parameters (Robinson and Creanza 2019), we failed to find such a relationship, which does not support the hypothesis that the evolution of song complexity could be driven by female choice and remain under sexual selection in this species. Also, the effort which male birds, including A. paludicola, put into singing was suggested to be important for their fitness (Hoi-Leitner et al. 1995, Dyrcz et al. 2011), and hence the song duty cycle could indicate male investment. The time a male spends singing could additionally affect his chances for a copulation and success of female-guarding. However, our study does not indicate that singing effort is related to the studied fitness-traits in A. paludicola. Although in some songbirds it was found that repeatability of the repertoire size is greater than expected by chance (Naguib et al. 2019, Salazar et al. 2021), future studies on A. paludicola should attempt to establish variation in song traits more precisely (e.g., by recording males repeatedly within a season, and to cover more breeding seasons).

Because A. paludicola males do not participate in nest building, incubation or offspring care, if their song is a signal of male quality, females that use this signal as a cue can be expected to obtain only indirect genetic benefits from the males (Leisler and Schulze-Hagen 2011), although see (Dyrcz et al. 2011), which could arise either from increased attractiveness of progeny and (or) its better viability/quality. However, we did not find compelling evidence for a relationship between any of the three song features and chick tarsus or scaled mass index change between days 2 and 9 post-hatch. This does not support the prediction that the song metrics that we used signify enhanced genetic male quality to females. Still, our results might have been affected by the moderate sample size of males recorded and low number of nests found to be fathered by the males, as well as only one breeding season of study.

In male birds, song metrics are negatively correlated with the inbreeding rate and positively associated with genetic diversity (Seddon et al. 2004, Reid et al. 2005, Araya-Ajoy et al. 2009, de Boer et al. 2016). They are also positively related to the body condition (Kipper et al. 2006, Pfaff et al. 2007, Wang et al. 2019, Lyra et al. 2022) and survival (Rytkonen et al. 1997, Rivera-Gutierrez et al. 2010, Berg et al. 2020). In contrast, we did not find any of the studied song variables to vary with wing-length, inbreeding rate and long-term survival of the males, and the song duty cycle and A-song rate did not vary by the scaled mass index. Our estimate of g₂ was above zero but low, indicating low variance in inbreeding, which, however, does not rule out finding a correlation with individual traits (Szulkin et al. 2010, Miller and Coltman 2014). Heterozygosity-fitness correlations are typically weak (Chapman et al. 2009), which is in line with the low support for such a relationship in our study. Thus, in A. paludicola, the song duty cycle and A-song rate are not likely to be used by females to assess male quality, as measured with the above metrics, nor to be affected by them. However, we found evidence that song repertoire increased with scaled mass index. Some previous studies have demonstrated that the repertoire size in birds is consequent on conditions during early developmental stages (MacDougall-Shackleton and Spencer 2012). Therefore, the repertoire size in A. paludicola males could signal to females how well they tolerated conditions during development, which—if this has a heritable component—could affect the fitness of the females’ offspring. Alternatively, lower current body condition could decrease the frequency of syllables that are the costliest to produce (Vallet and Kreutzer 1995, Suthers et al. 2012). If this causes the repertoire to be presented at a slower speed, within a short time window lower body condition could correlate with a lower repertoire size.

In our study, we failed to find strong evidence for an effect of Trypanosoma infection on song, and Plasmodium infection was not related to song duty cycle and repertoire size, but contrary to our expectations, we found that infection by Plasmodium was positively related to the frequency of A-songs. This association could be driven by age, assuming that this factor also affects the agonistic song rate, as already suggested above. This could take place if the probability of acquiring an infection by A. paludicolas increases with age, as has been observed in many avian systems (Marzal et al. 2016; Freeman-Gallant and Taff 2017), and once parasitized with Plasmodium, birds only rarely clear the infection (Rooyen et al. 2013). Furthermore, if only males with a certain genetic makeup can withstand the infection, females choosing parasitized males may select for enhanced survival of their progeny. Clearly, the mechanism underlying the positive association between Plasmodium infection and agonistic song in A. paludicola requires further research. It is also worth noting that song performance may be better explained by infection intensity (i.e., parasite load) rather than infection status (infected vs. uninfected), which to date has been the most commonly used infection metric (Buchanan et al. 1999, Redpath et al. 2000, Lopez-Serna et al. 2021). The relationship between song traits and parasite load also remains to be studied.

Conclusions

In a highly promiscuous songbird without paternal care, A. paludicola, we found support that it is the agonistic song rate, rather than song elaboration and singing effort, which could have evolved through sexual selection. We also did not find firm evidence that song complexity, agonistic song rate or singing effort could convey information on male genetic quality in this species, as manifested by chick growth. Finally, our findings suggest that the agonistic song rate could signal to females a male’s genetic makeup that enhances his survival upon infection with Plasmodium, and that the repertoire size might be consequent on the past developmental stress or current condition of the male. We must stress, however, that our results await reproducing using greater sample sizes (more breeding seasons, males, and songs sampled). Also, the associations that we observed call for identifying the underlying mechanisms, such as relating the agonistic song rate to amount of female-guarding, sperm quality, male age, and nest predation; and differentiating between the hypotheses linking Plasmodium infection to agonistic song rate.

Supplementary material

Supplementary material is available at Ornithology online.

Acknowledgments

We highly value the field assistance of Grzegorz Kiljan, Piotr Guzik, Marta Celej, Julien Foucher, Beata Głębocka, Aneta Gołębiewska, Maciej Kamiński, Agnieszka Kuczyńska, Romuald Mikusek, Felix Närmann, Edyta Podmokła, Aneta Rybińska, Paulina Siejka, Michał Walesiak, and Laura Zani. Song recording was laboriously carried out by Pedro Costa. We thank the Gugny Field Station of the Institute of Biology, University of Białystok, Poland for providing accommodation and workspace during fieldwork. Susan Mbedi, Ismael Reyes, and Katarzyna Dudek readily supported the preparation and sequencing of RAD-seq libraries. Wiesław Babik offered lab space and equipment for molecular analyses.

Funding statement

The lab work and bioinformatic analysis was partially funded by the German Federal Ministry of Education and Research (BMBF, Förderkennzeichen 033W034A). The sequencing was performed in cooperation with the Competence Center for Genome Analysis at the Christian-Albrechts-University in Kiel and at the Berlin Center for Genomics in Biodiversity Research. The study was funded by the Polish National Science Centre’s grant no. 2016/20/S/NZ8/00434 awarded to Justyna Kubacka.

Ethics statement

The study was performed in accordance with the requirements of the Polish Law regarding animal welfare and conservation. Birds were caught under permits from the Polish Ministry of Environment (DOP-WPN.286.38.2016.AN, DOP-WPN.286.40.2016.AN) and the Regional Directorate of Environmental Protection in Białystok (WPN.6401.244.2016.MN). Ringing was conducted by licensed ringers. Blood samples were collected under a permit from the I Local Ethical Committee in Warsaw (185/2016).

Conflict of interest statement

Authors declare no financial or non-financial interests that are directly or indirectly related to the work.

Author contributions

J.K. and T.O. conceived the idea and study design. J.K., M.H-R., S.S., and A.D. performed the study. J.K., T.O., A.D., L.A., and M.H-R. wrote and substantially edited the paper. J.K. (field methods), T.O. (song methods), C.M. (laboratory and bioinformatic methods), and L.A. (bioinformatic methods) developed or designed methods. T.O. (song analysis), C.M. (bioinformatic analysis), L.A. (bioinformatic analysis), M.H-R. (bioinformatic analysis), A.D. (blood parasite identification), and J.K. (STACKS and statistical analysis) analyzed the data. T.O., A.D., and C.M. contributed substantial materials, resources, or funding.

Data availability

Datasets and R code are available in the Dryad repository under Kubacka et al. (2024b). Sequences are available at https://www.ncbi.nlm.nih.gov/sra/PRJNA1155693 (BioProject accession number, PRJNA1155693).

LITERATURE CITED