Sex allocation theory predicts that mothers should bias investment in offspring toward the sex that yields higher fitness returns; one such bias may be a skewed offspring-sex ratio. Sex allocation is well-studied in birds with cooperative breeding systems, with theory on local resource enhancement and production of helpers at the nest, but little theoretical or empirical work has focused on birds with brood parasitic breeding systems. Wood ducks (Aix sponsa) are conspecific brood parasites, and rates of parasitism appear to increase with density. Because female wood ducks show high natal philopatry and nest sites are often limiting, local resource competition (LRC) theory predicts that females should overproduce male offspring—the dispersing sex—when competition (density) is high. However, the unique features of conspecific brood parasitism generate alternative predictions from other sex allocation theories, which we develop and test here. We experimentally manipulated the nesting density of female wood ducks in four populations from 2013-2016 and analyzed the resulting sex allocation of >2000 ducklings.  In contrast to predictions we did not find overproduction of male offspring by females in high-density populations, females in better condition, or parasitic females; modest support for LRC was found in overproduction of only female parasitic offspring with higher nest box availability. The lack of evidence for sex ratio biases, as expected for LRC and some aspects of brood parasitism, could reflect conflicting selection pressures from nest competition and brood parasitism, or that mechanisms of adaptive sex ratio bias are not possible.

Field sites. We studied Wood Ducks in the Central Valley of California, USA, at four sites near the town of Davis, from 2013-2016: Russell Ranch (lat 38°32’03” N, long 121°52’05” W, USA), Putah Creek (lat 38°31’0” N, long 121°46’05” W, USA), Conaway Ranch (lat 38°38’24” N, long 121°42’0” W, USA), and Roosevelt Ranch (lat 38°49’15” N, long 121°48’39” W, USA). Sites were remnant oak woodland habitat, persisting in narrow riparian corridors (<20m wide on each bank, 2-5km long) along a stream (Russell Ranch, Putah Creek), slough (Conaway Ranch), or restored and managed wetland complex (Roosevelt Ranch) adjacent to agricultural fields.

We capitalized on an ongoing experimental manipulation of nesting female density to examine changes in offspring sex ratio. We installed a total of 72 nest boxes at Conaway Ranch (16 hectares) as a high-density treatment. We installed 16 nest boxes at Russell Ranch (8 hectares) and 6 nest boxes at Putah Creek (5 hectares) as low-density treatments in 1998-99. We installed 49 nest boxes in part of Roosevelt Ranch (277 hectares) as a low-density treatment, and 51 nest boxes in the remaining part of Roosevelt Ranch (35 hectares) as a high-density treatment. However, since females moved among treatment areas, we considered Roosevelt Ranch as a single site in the population-level analyses. Boxes were bolted 1.5-4m high on an existing tree, within 10m of water (Russell, Conaway, Putah), or attached to 3m metal poles with a sliding fixture, within 2-5m of water (Roosevelt Ranch). Each box was fitted with an antenna circling the box entrance, connected to a custom radio frequency identification (RFID) reader and 12V battery; the RFID readers recorded wood duck females carrying passive integrated transponder (PIT) tags with unique codes whenever they entered or exited the boxes during the breeding season. Logged RFID reads were used to determine the onset of incubation and to corroborate the identity of the incubating female as determined by capture on the nest. Box locations were recorded with GPS.

Field methods. At the onset of each breeding season (mid-February), boxes were checked weekly for nesting activity: “bowling” of wood shavings inside the box, which indicated that a wood duck female had rearranged them, or the presence of eggs. Once a box showed nesting activity, it was checked approximately every 2 days. New eggs were numbered at each nest check until incubation began; incubation onset was confirmed when nest checks revealed that eggs were warm and covered with a layer of down, from which we estimated the likely date of hatch (~30 days from incubation onset, Haramis 1990).

Adult females were captured in nest boxes by closing the box entrance with a wooden plug. At initial capture, females were banded with aluminum USGS numbered bands and injected with a unique PIT tag for permanent identification; mass (to the nearest 5g) and tarsus length (to the nearest mm) were measured at the first and all subsequent captures. Females were generally scheduled for capture near the end of incubation, to prevent nest abandonment associated with possible capture stress; hence size measurements (mass, tarsus) of females were standardized by reproductive stage (i.e. late-incubation) instead of calendar date. A blood sample was taken by pricking a female’s tarsal or alar vein with a 20-gauge or 30-gauge needle and collecting droplets onto a filter strip (Nobutu Blood Filter Strip, Advantec MFS, Japan).

Beginning two days before the estimated date of hatch, eggs were checked daily for evidence of tapping or external pipping of the eggshell. Upon hatching, each duckling was injected with a unique PIT tag, underneath the skin between the scapulae. Duckling blood samples were taken by pricking the tarsal vein with a 30-gauge needle and collecting droplets onto a blood filter strip.

Blood samples on filter strips were kept at ambient temperature for several hours until returned to the lab, at which point they were transferred to a freezer (0°F) for storage until extraction. DNA from samples was extracted 1-4 months after collection, using DNeasy Blood and Tissue Kit spin columns (Qiagen, USA), according to the manufacturer’s protocol, or using a plate-extraction method (Ali et al. 2016).

Genotyping and genetic sexing. We genotyped females and ducklings at 19 microsatellite loci: APH01, APH02, APH08, APH09, APH13, APH18, APH19, APH20, APH23, APH25 (Maak et al. 2000; Maak et al. 2003); APL02, APL23 (Denk et al. 2004); BCAμ5 (Buchholz et al. 1998); CM28, CM35 (Stai and Hughes 2003), SFIμ4 (Fields and Scribner 1997); SMO04, SMO07, SMO10 (Paulus and Tiedemann 2003). GTTTCTT tails were added to reverse primers to prevent split peaks. Duckling sex was determined from the genotype at 2 sex-linked loci. Primers P2/P8 (Griffiths et al. 1998) and 1237L/1272H (Kahn et al. 1998) both amplify an intron in the CHD gene on the Z and W sex chromosomes: females are heterozygous (ZW) and males are homozygous (ZZ). Molecular sex was confirmed for PIT-tagged female ducklings that returned to nest boxes as adults (n=17).

Fluorescently-labeled microsatellite and sex-linked primers were multiplexed into three 25μl reactions (Thow 2019), each consisting of 2.5μl PCR buffer [750 mM Tris-HCL pH 8.8, 200 mM (NH₄)_sSO₄, 0.1% Tween 20], 2.5 μl 25 mM MgCl₂, 2.5μl dNTPs, 0.5μl DMSO, 0.2μl Taq polymerase (Denville Choice), 6.8μl water, 7μl multiplexed primers, and 3μl template DNA. PCR reactions consisted of an initial denaturation of 5 minutes at 95°C followed by 5 minutes at 85°C; then 5 cycles of 1 min denaturation at 95°C, 30-sec annealing at 57°C, and 30-sec elongation at 72°C; 28 cycles of 45sec at 95°C, 30sec at 57°C, 30sec at 72°C; and ending with 30min final elongation at 72°C.  PCR products were visualized on an ABI 3730 sequencer and alleles were scored using STRand analysis software (www.vgl.ucdavis.edu/informatics/strand.php). Genotyping was performed by the Veterinary Genetics Laboratory at the University of California, Davis.

Maternal assignment. Genetic assignment of ducklings to females was performed using COLONY 2.0 (Jones and Wang 2010). COLONY 2.0 uses multilocus genotypes and full-pedigree likelihood methods to simultaneously infer parentage and sibships. A pairwise likelihood approach, most commonly CERVUS 3.0 (Kalinowski et al. 2007), is often used to assign parentage in wild populations, but COLONY has been found to make fewer errors in assignment compared to CERVUS for populations with female kin structure and partial sampling of mothers (Thow et al. 2022). Additionally, COLONY can identify genetically unique un-sampled parents and assign offspring to them; this feature is particularly useful for conspecific brood parasitic systems in which eggs may be laid by unsampled females (Thow et al. 2022). With either program, incorrect assignment of offspring from nesting females to other females (i.e. errors that falsely suggest brood parasitism) is rare (Thow et al. 2022).

Separate assignments were conducted for the ducklings hatched in each population in each year. All females captured in nest boxes or by RFID in the current and previous year(s) were included as candidate mothers, excluding only females that were known to have died (i.e. were depredated on the nest or recorded shot through BBL hunter band return records). No adult males were genotyped in this study, so males were not included as candidate fathers. We specified an outbreeding model, with no known sib-ships, no excluded mothers, no excluded sib-ships, and no sib-ship scaling or size prior. We allowed a polygamous mating system for males and females. We conservatively set the probability that a mother is included in the female candidates to 0.7, to allow for high numbers of un-captured, un-sampled exclusive parasite females; setting this parameter lower than the true probability may reduce COLONY’s reported confidence in an individual assignment, but does not change the identity of the assigned parent(s) (Thow et al. 2022). We selected the longest processing run option, using the full likelihood approach, with four replicates to reduce sampling bias (Wang 2016; Thow et al. 2022). We accepted all maternity assignments made regardless of probability, since errors are not associated with low probabilities (Thow et al. 2022), using the BestCluster output.

Parasitic ducklings. Parasitic ducklings were identified by comparison between the genetically assigned mother and the incubating female. The identity of the incubating female for each nest was known from her capture on the nest, or by RFID reads indicating incubation (i.e. full days spent on the nest until the eggs hatched). Ducklings that were genetically assigned to the incubating female were categorized as “nest” ducklings. Ducklings that were genetically assigned to a female other than the one that incubated them, including to unsampled females inferred by COLONY, were categorized as “parasitic” ducklings.

Female alternative reproductive tactics. Female ARTs were determined by the hatching location of her ducklings. Females that incubated a nest and were assigned maternity only to ducklings that hatched from that nest were categorized as “Nest” females. Females that incubated a nest and were assigned maternity to ducklings that hatched from that nest and ducklings that hatched from a different nest were categorized as “Nesting Parasite” females; because sex allocation may change across the laying sequence (Cassey et al. 2006; Bowers et al. 2014), sex allocation for nest ducklings and parasitic ducklings were considered separately for Nesting Parasite females, which laid both. Females that did not incubate a nest but were assigned maternity to ducklings were categorized as “Parasite” females.

Statistical analyses. Following other studies of offspring sex ratio in ducks, in this study, we analyzed secondary sex ratios only, defined as the ratio of male to female ducklings at hatch (Blums & Mednis 1996). A few eggs in some successful clutches did not hatch due to embryonic mortality during incubation, or because their development was not synchronous with the rest of the clutch.

To test predictions from multiple sex allocation hypotheses (Table 1), we analyzed offspring sex allocation at the population and the individual level. At the population level, we used generalized linear models (GLMs) to predict the Z-score sex ratio (Thogerson et al. 2013) produced for each site year: 2013-2015 at Putah Creek, 2013-2016 at Russell Ranch, 2014-2016 at Conaway Ranch, and 2015-2016 at Roosevelt Ranch (n=12). Data from Conaway Ranch 2013 were excluded because excessive predation of hens and clutches that year precluded complete genotyping of offspring, and therefore the number and sex of offspring produced. Putah Creek was not monitored in 2016, and complete sampling of offspring was not initiated at Roosevelt Ranch until 2015. For each population predictor variable (e.g. female density), we specified separate models to predict the sex ratio of three categories of offspring: all ducklings produced, nest ducklings (i.e. those incubated by the genetically-assigned mother), and parasitic ducklings (i.e. those incubated by a female other than the genetically-assigned mother). We examined these three categories separately to allow patterns of sex allocation to vary by duckling type. 

At the individual level, females were often observed in multiple years (n=284 observations of 175 females). Therefore, we used generalized linear mixed models (GLMMs) to fit our data, including female identity as a random effect with a varying intercept. We used the bglmer function (from package blme, Dorie &Dorie 2015) to obtain non-zero random effects for individual females. Models were specified with a binomial error structure and a logit link, with separate models predicting the proportion of males produced by a female each year in three categories: proportion of her total offspring that year, of her nest offspring that year, and of her parasitic offspring that year. Effects are reported as beta-coefficients plus standard error; positive beta-coefficients represent an increase in the proportion of males produced.

First, we tested predictions from the local resource competition and local resource enhancement hypotheses, that females bias the sex ratio of offspring according to local resource availability. Female density is a common proxy for resource availability, and we defined it in two ways. Females per hectare were the number of females that bred in a given site in a given year, divided by the site area. The nearest neighbor distance was the shortest Euclidian distance in meters (determined from GPS locations) from the nest box used by the focal female and another occupied box. The average nearest neighbor distance was the average of the distances to the two closest occupied boxes for individual females; the average of this value was taken for all individual females to determine the population nearest neighbor distance. The distribution of average nearest-neighbor distances was right-skewed for individual females across all populations, so this variable was log-transformed in individual models; log transformation was not necessary for population averages. A positive effect of females per hectare on the proportion of males produced, or a negative effect of nearest neighbor distance on the proportion of males produced, would support the local resource competition hypothesis that wood duck females produce more sons at high density. We used nest site availability as another measure of resource competition at the population level, defined as the average number of nest boxes available per breeding female. A negative effect of nest site availability on the proportion of males produced would support the local resource competition hypothesis that females produce more sons when nest sites are relatively scarce.

Additionally, since LRC predictions are particularly applicable to competition with relatives (West et al. 2005), we analyzed sex ratios of ducklings produced in populations that varied in average relatedness among females, and by females that differed in their average relatedness to their two closest neighbors. We used ML-Relate (Kalinowski et al. 2006) to estimate pairwise relatedness, r, between adult females present in the population each year. A positive effect of relatedness on the proportion of males produced would support the local resource competition hypothesis that females produce more sons when competition with relatives is high.

Next, we tested predictions from the Trivers-Willard hypothesis that females bias the sex ratio of offspring according to their structural size or body condition. For these predictions, we fit separate binomial GLMMs of total, nest, and parasite ducklings; size or body condition were tested as fixed effects, and female identity was included in all models as a random effect. We used tarsus length as our estimate of female structural size (Jaatinen et al. 2013). We defined body condition as the residual from a linear model of mass on tarsus length, using the mass at the first capture of the year. To determine if variation in female body condition could have contributed to site-specific variation in offspring sex ratio, we fit an ANOVA of condition residuals by site.

Last, we tested the effect of the three alternative reproductive tactics of female wood ducks (Nesting, Parasite, and Nesting Parasite) on offspring sex allocation by fitting binomial GLMMs including ART as a categorical fixed effect. We compared a) the sex ratio of total ducklings produced by all three ARTs, b) the sex ratio of nest ducklings produced by Nest vs. Nesting Parasite females, and c) the sex ratio of parasitic ducklings produced by Parasite vs. Nesting Parasite females. A higher (male-biased) sex ratio produced by Parasite females or a lower (female-biased) sex ratio produced by Nesting Parasite females would support Silk’s (1983) expansion of the local resource competition hypothesis that females bias sex ratio according to their competitive ability. To determine if Nesting Parasite females produced distinct sex ratios according to each of their dual tactics, we fit a binomial GLMM of proportion male ducklings restricting the dataset to Nesting Parasite females only (n=66 observations of 61 females), with duckling type (“nest” vs. “parasite”) as a fixed effect. A higher (male-biased) sex ratio for parasitic vs nest offspring would highlight laying-order effects in females laying supernumerary eggs. For all of the models in this section, we included the site as a fixed effect to control for the effect of the experimental density treatment.