Study species and study area
We studied red-winged blackbirds at Newark Road Prairie in Rock County, Wisconsin, USA (42° 32′ N, 89° 08′ W) during the breeding season of 2018. Newark Road Prairie is a 13-ha wet-mesic remnant prairie and sedge meadow habitat that supports about 35 male redwing territories38. Capture and banding were conducted under United States Geological Survey banding permit #20438 to K.Y. All males were captured in Potter traps placed on platforms and baited with non-viable sunflower seeds. Captured males were banded with USGS numbered aluminum bands and unique color combinations of plastic wraparound bands for individual identification. We did not aim to capture and band females in 2018, although a few females had been captured and individually color banded in previous years.
We used observations of territorial behavior to place an additional seed-baited trap within the breeding territory of each potential subject. We pinned open the doors of each trap, baited it daily, and then observed each baited trap to determine when the male began entering it. The male was then allowed 1–3 weeks of unrestricted access to the bait without capture, at which point the male would usually enter the trap quickly after we baited it. Playbacks were then performed as described below. In our long-term experience working at this site (K.Y.), this method allows most individual males to be reliably recaptured within 30 min. Previous work at this site on redwings’ responses to model cowbirds vs. harmless heterospecifics, coupled with their respective call playbacks, showed strongly graded aggressive responses to the former relative to the latter16.
Playbacks
The playback protocol was approved by the Beloit College IACUC (protocol #18002). We presented one of three playback types at each territory: (1) male conspecific songs (“Redwing”; highly salient), (2) female cowbird chatter calls (“Cowbird”; salient heterospecific vocalization of a brood parasite of redwing nests), or (3) dove coo (“Dove”; non-salient vocalizations of a harmless sympatric heterospecific) for broadcasts. Given that male red-winged blackbirds exhibit greater behavioral responses to female vs. male cowbird models16, the chatter call, a specific call of female cowbirds was chosen to acoustically simulate brood parasite intrusion. To address pseudo-replication39, one out of five different individuals’ song/call exemplars were assigned at random for each playback type. Audacity v 2.2.0 was used to filter playback stimuli above 2000 Hz and below 500 Hz, and to normalize mean amplitude of all stimuli. Acoustic stimuli were matched in peak amplitude and duration. Exemplars were thus similar to each other in peak amplitude with duration.
Playback types and exemplar files were randomly assigned to territorial male redwings using a stratified balanced design to keep sample sizes per type similar. For each playback type we randomly chose one of the exemplars to broadcast from an iPhone 5 or 6 (Apple Inc., Cupertino, California, USA) connected to an Ecoxgear ECOXBT speaker (Grace Digital Audio, Peterborough, Ontario, Canada) via a 30-m auxiliary cable. The speaker was placed on the trap, which was closed and unbaited during playbacks. Playbacks were broadcasted at 80–85 dB SPL at 1 m from the source (as measured by a sound pressure meter: Pyle PSPL01, Pyle Audio Inc., Brooklyn, New York, USA), which approximated natural amplitude (KY personal observations).
We used a 30-min paradigm to induce (differential) gene expression (e.g.23), but to avoid habituation in the field, each playback consisted of three 10-min segments for a total period of 30 min. The first and third segments were active sound broadcast periods and the second segment was a 10-min silent period. Each active period consisted of 10 1-min sub-segments in which a single exemplar (song, chatter, or coo) played at 0, 10, 20, 30, and 40 s, followed by 20 s of silence. As soon as the 30-min playback was completed, we removed the playback equipment and baited and set the trap; we aimed to capture each subject within 30 min of the end of the playback and to band all previously unmarked subjects as described above. We recorded the time to capture for all subjects obtained within the 30-min maximum. Each captured bird was removed from its trap within 1 minute (usually less than 30 s), processed, and released within 5 min of capture. A blood sample of about 50 μl was collected from the brachial vein of each hand-held, unanesthesized subject using a 1.27-cm 27-g needle and heparinized caraway tube (Fisher Scientific, Pittsburgh, Pennsylvania, USA). The birds remained calm, did not react to the insertion of the needle, and resumed normal behavior within minutes of release. Each blood sample was placed in 500 µl of RNAlater and then stored in a −80 °C freezer for subsequent RNA extraction and sequencing.
Prior to playback, we placed two markers each 10 m to the right and left of from the speaker to facilitate measuring proximity time spent within 10 m of the speaker. During each 10-min segment we recorded (1) number of songs (songs), (2) number of flights towards the speaker (approaches), (3) time spent within 10 m of the speaker (time in proximity), and (4) the distance (m) of the closest approach (closest approach). These behavioral variables are well known to indicate a male redwing’s aggressiveness40. We analyzed behavioral data collected during the 1st and 3rd 10-min periods, which preceded the capture timepoint by up to 30 min and, thus, is representative of the time sampled for the gene-expression patterns, too. Number of songs and number of approaches were the totals for periods 1 and 3 (active playback). Time (min) within 10 m was the total time for the active playback periods. Closest approach was the shortest distance between the subject and the speaker during the two active playback periods.
We used generalized linear mixed models with a negative binomial response and individual male identity as a random effect with glmmTMB in R (version 3.5.1) to analyze the responses of male redwings to the three playbacks. Each model included playback treatment, playback trail segment, and whether the male was captured as explanatory variables. A significant result was further examined with a Tukey post-hoc analysis to identify significantly different pairs of treatment. The alpha level was set at the p < 0.05.
RNA extraction and sequencing
RNA was extracted with RiboPure blood kits (Life Technologies, Carlsbad, California, USA) and treated with DNAse for purification. We assessed the quality of purified RNA on a Bioanalyzer (Agilent, Wilmington, Delaware, USA) (RIN > 7.0).
All library preparations and sequencing were performed at the University of Illinois at Urbana-Champaign Roy J. Carver Biotechnology Center, Urbana, IL, USA. A library for each sample was prepared with an Illumina TruSeq Stranded RNA sample prep kit. All libraries were pooled, quantitated by qPCR, and sequenced on one lane of an Illumina HiSeq 4000 with a HiSeq 4000 sequencing kit version 1, producing single-end 100 bp reads. Fastq files were demultiplexed with bcl2fastq v 2.17.1.14 (Illumina).
Preparation of reference genome
Lacking an annotated reference genome for the red-winged blackbird, we created a proxy reference following the pseudo-it pipeline (https://github.com/bricesarver/pseudo-it). Briefly, we extracted DNA from liver and muscle tissue of a male red-winged blackbird (cataloged at the Museum of Southwestern Biology MSB:Bird:40979) and performed paired-end (200 bp) whole-genome sequencing on one lane of HiSeq (2500) at the Duke Genome Center. After removing the Illumina adapters with Trim Galore! v 0.3.7 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), which incorporates Cutadapt v 1.7.141, we aligned the DNA reads to the closest related species publicly available, the white-throated sparrow (Zonotrichia alibicolis) genome (assembly Zonotrichia_albicollis-1.0.1), with BWA-mem42. Using GATK43 we then identified and inserted red-winged blackbird SNPs into the white-throated sparrow reference genome. To improve the proxy reference genome, we performed an additional iteration of the pseudo-it pipeline.
Gene expression
We removed Illumina adapters from RNA reads with Trim Galore! v 0.3.7. We then aligned the reads to the proxy reference genome with Hisat244 and quantified read abundance with HTSeq-count45. With DeSeq 246, we then included the playback treatments and time to capture (minutes) to analyze the gene expression patterns and also to compare candidate genes identified from the parallel expression patterns of peripheral (whole blood) in a previous RNA-sequencing study on zebra finches in response conspecific vs. heterospecific songs (n = 20 differentially expressed genes; Table 1 in23).
Next, we sought to identify networks of genes specifically responsive to Redwing song, Cowbird chatter or both Redwing and Cowbird relative to Dove coo. We performed weighted gene co-expression network analysis (WGCNA), which is used for finding clusters (modules) of highly correlated genes and determine the relationship of modules to treatments. We used the WGCNA package in R47 to identify modules of co-expressed genes in our dataset. To remove genes with low read abundance, we filtered for genes with <1 count per million in at least 10 samples. We then normalized for read-depth and extracted variance stabilizing transformed (vst) read counts from DEseq. 2 into WGCNA. To build the co-expression matrix, we chose a soft thresholding power (β) value of 12, at which at which we observed a plateau in Mean Connectivity, thus representing a scale-free topology47. We generated a signed network with minimum module size of 30 genes and merged highly correlated modules (dissimilarity threshold = 0.25). We then correlated the eigengene, which is the first principal component of a module, of these merged modules with playback treatments (redwing, cowbird, dove). Modules with p < 0.05 were considered significantly correlated with a given trait.
Finally, we tested for functional enrichment of gene ontology (GO) categories with GOrilla30. For each module, genes were ranked based on their module membership score determined in the WGCNA analysis. We preferred this rank-order based approach (as opposed to strict module assignment) as it reflects the correlation among modules, and because some genes could be assigned to multiple modules48. GOrilla performs ranked-order analyses with human gene IDs, so we identified orthologous genes from the annotated red-winged blackbird proxy-genome. Statistical significance of GO categories was determined with p-values corrected for multiple hypothesis testing (FDR < 0.05). We used REVIGO to remove redundant and overlapping GO categories, with an allowed semantic similarity measure of 0.549.
Source: Ecology - nature.com
