Songbird preen oil odour reflects haemosporidian parasite load

Abstract

Investigating the impact of parasitism on host phenotype is key to understanding parasite transmission ecology, host behavioural ecology and host–parasite coevolution. Previous studies have provided evidence that avian odour is one such phenotypic trait, as mosquitoes that vector the haemosporidian blood parasite Plasmodium tend to prefer birds that are already infected. Preen oil is a major source of avian odour, yet studies to date have not identified differences in preen oil odour based on the presence or absence of haemosporidian infection. Because preen oil can vary with physiological dynamics, we predicted that the composition of preen oil odours might vary according to parasite load, rather than solely by the presence or absence of infection. We used gas chromatography–mass spectrometry to characterize the composition of volatile compounds in preen oil taken from female dark-eyed juncos, Junco hyemalis carolinensis, and asked whether their composition varied with relative haemosporidian parasite load, which we assessed using quantitative PCR. We identified a subset of volatile compounds (a ‘blend’) and two specific compounds that varied with increasing parasite load. Importantly, the quantity of these compounds did not vary based on parasite presence or absence, suggesting that birds with low parasite loads might be phenotypically indistinguishable from uninfected birds. The volatile blend associated with parasite load also varied with sampling date, suggesting a possible seasonal relapse of chronic infections triggered by shifts in junco host reproductive state. Furthermore, we found a positive relationship between parasite load and a volatile blend shown in a previous study to predict reproductive success in juncos. This is the first study to demonstrate quantitative differences in avian host odour based on haemosporidian parasite load. Our findings highlight the importance of focusing on parasite load, rather than solely presence or absence, in investigating host–parasite interactions.

Individual odour is a dynamic aspect of phenotype reflecting an individual’s internal state and is therefore well suited as a focus for investigating the impact of parasitism on host phenotype. Previous work has demonstrated that parasitized hosts produce distinct odours by showing that ecologically relevant receivers—such as potential mates or social group members—can distinguish between odour cues based on host infection status. For example, heterosexual women are more likely to use negative adjectives to describe the body odour of men with active gonorrhea infections (Moshkin et al., 2012), and banded mongoose, Mungos mungo, show altered scent-marking behaviour when encountering scent marks of conspecifics with heavy loads of gut parasites (Mitchell et al., 2017). Studies also suggest that parasite vectors can distinguish the infection status of potential hosts. For example, mosquitoes that vector malaria-causing Plasmodium parasites often prefer the odours of mammalian and avian hosts that are already infected with the parasite (e.g. Cornet et al., 2013a; De Moraes et al., 2014; Robinson et al., 2018; but see Lalubin et al., 2012). Clarifying the relationship between parasitism and individual odour is therefore key to better understanding host behavioural ecology, host–parasite coevolution and parasite transmission.

Plasmodium is one of three genera of haemosporidian parasites that can cause severe sickness and even death in newly infected hosts; those that recover often suffer from long-term (i.e. chronic) infections (Palinauskas et al., 2008). In humans infected with Plasmodium, compounds such as certain aldehydes and terpenes are likely responsible for the preference that mosquitoes show for infected hosts, which may be key in developing methods to disrupt transmission of this virulent parasite from mosquitoes to humans (Robinson et al., 2018; Schaber et al., 2018). In avian hosts, the impact of Plasmodium varies widely, from population level declines, as seen in Hawaiian honeycreepers (LaPointe et al., 2012) and house sparrows, Passer domesticus, in England (Dadam et al., 2019), to economic loss through poor egg and meat production in domestic poultry (Pattaradilokrat et al., 2015), to more subtle effects, such as reduced reproductive success in great reed warblers, Acrocephalus arundinaceus (Asghar et al., 2011). Despite these clear costs of infection, identifying the source and identity of compounds reflecting Plasmodium parasitism in avian hosts has remained elusive. Choice tests have shown mosquitoes prefer the odour of live birds infected with Plasmodium (Cornet et al., 2013a, 2013b, Díez-Fernández, Martínez-de la Puente, Gangoso, Ferraguti, et al., 2020; but see Gutiérrez-López et al., 2019; Lalubin et al., 2012), yet mosquitoes fail to show the same preferences in choice tests of preen oil—a potential source of relevant odours—isolated from both infected and uninfected birds (Díez-Fernández, Martínez-de la Puente, Gangoso, López, et al., 2020). Interpreting negative results of such studies is difficult because the failure of one receiver to distinguish between host odours based on infection does not definitively show that parasitism-based differences in odour do not exist or that they would be undetectable by other types of receivers.

Preen oil is a lipid-rich secretion of the uropygial gland; the uropygial gland is the main avian holocrine gland, located anterior to the tail. During grooming, preen oil is transferred to the bill and spread across feathers, providing various functions (reviewed in Moreno-Rueda, 2017). Preen oil is a metabolic product of an individual bird and of the microbial fauna of its uropygial gland (Apandi & Edwards, 1964; Whittaker, Soini, et al., 2011; Whittaker et al., 2019). The volatile fraction of preen oil is composed of many compounds that vary by species (Soini et al., 2013). Preen oil is thought to contribute strongly to individual bird odour, and previous work has shown that birds can discriminate between preen oil samples based on species and sex (Van Huynh & Rice, 2019; Whittaker et al., 2009; Whittaker, Richmond et al., 2011; Zhang et al., 2010). Given the importance of preen oil as a component of avian odour, it is thought to be the most likely source of chemical cues reflecting parasitism. In fact, adding preen oil can increase the success of CO₂ traps used to capture mosquitoes, suggesting that preen oil-based odours comprise one of the cues that mosquitoes use in host searching, alongside visual cues, heat and CO₂ (Buehlmann et al., 2020; Russell & Hunter, 2005).

However, when investigators presented preen oil isolated from infected and uninfected birds in choice tests, mosquitoes did not show a preference between the two groups, despite showing a preference for the infected group when the odour of live birds were used (Díez-Fernández, Martínez-de la Puente, Gangoso, Ferraguti, et al., 2020). These results could have several causes. Importantly, one might expect the effect of parasitism on host odour to be subtle given the variety of factors known to contribute to variation in odour composition, such as species identity, population, sex and reproductive status (Soini et al., 2007, 2013; Whittaker et al., 2010; Zhang et al., 2010). In addition, there may be a mismatch in the ecological context of the mosquito and host species sampled, perhaps attributable to a vector’s host preferences (Simpson et al., 2009; Williams et al., 2003), to the absence of multimodal cues provided by live birds, or to variation in impact of Plasmodium on host physiology depending on the host and parasite species involved (Ellis et al., 2014; Marzal et al., 2008; Palinauskas et al., 2008). Furthermore, preen oil odours may not vary substantially when compared by infection status, especially if many individuals within the infected group have low parasite loads. Although these circumstances may make it challenging to detect parasitism-based variation in host odour, such variation could have important ecological relevance because even small differences in the proportions of different compounds can influence the ways receivers perceive an odour blend (Hummel et al., 2013; Laing & Willcox, 1983).

This last possibility is supported by previous work showing mosquito feeding preferences vary with respect to Plasmodium load in both laboratory mice and wild-caught songbirds (De Moraes et al., 2014; Yan et al., 2018). Similarly, humans with higher Plasmodium loads produce larger proportions of mosquito-attracting aldehydes (Robinson et al., 2018). Finally, it is possible that haemosporidian parasites simply do not influence preen oil odour, and the observed mosquito preferences for Plasmodium-infected hosts could be attributable to odours emanating from the breath or skin (Schaber et al., 2018; Verhulst et al., 2011). To distinguish among these possibilities, it is important to determine whether patterns of haemosporidian parasitism are reflected in preen oil odour variation. Additionally, determining whether such odour differences exist clarifies whether odour-based infection information could be available to additional receivers, such as conspecific songbirds (but see Grieves et al., 2020).

Our study tested the hypothesis that haemosporidian parasite load is associated with differences in the composition of volatile preen oil compounds in female dark-eyed juncos, Junco hyemalis carolinensis. If parasitism influences host physiology in a dose-dependent manner, birds with low parasite loads should produce odours that differ little from those of uninfected birds; in this case, comparing odour composition of infected and uninfected birds may obscure patterns of variation within the infected class. Therefore, we predicted that the relative proportions of volatile compounds extracted from preen oil would vary with increasing parasite load, but not necessarily by parasite presence. Furthermore, previous studies have demonstrated the ecological relevance of specific volatile compounds in songbird preen oil, including aldehydes, which can influence mosquito attraction (Douglas et al., 2005; Robinson et al., 2018), and a subset of ketones, which predict junco reproductive success (Whittaker et al., 2013). Therefore, we also tested the hypotheses (1) that the subset of ketones negatively associated with junco reproductive success will be detected in higher proportions in juncos with heavier parasite loads and (2) that higher proportions of aldehydes will be detected in juncos with heavier parasite loads. If preen oil odour does not vary by parasite load or presence, then receivers such as mosquitoes or conspecific juncos are unlikely to access parasitism information through preen oil odour.

METHODS

Study System and Sample Collection

We collected preen oil and whole blood samples from 28 female juncos early in the breeding season (23 April–12 May 2018) near Mountain Lake Biological Station (Pembroke, VA, U.S.A.). While breeding phenology varies among individuals and between years (Kimmitt et al., 2021), females captured during this period were likely building nests or laying the first clutch of the season given known phenology of this species (Nolan et al., 2002). Furthermore, 11 of 18 females checked for brood patches showed some degree of brood patch development. With the expectation that juncos would not have been exposed to mosquitoes over the winter, we sampled individuals in early spring to maximize the likelihood that any infections we detected were in the chronic phase. Females were used for this study for logistical reasons, as we were collecting data from adult females for a separate, concurrent study investigating brood patch development. As such, results should be interpreted as specific to females, and future work could assess sex differences in the findings discussed here.

Birds were captured using baited mist nets and released immediately following sampling. Briefly, preen oil was collected by gently probing the preen gland with a microcapillary tube while holding a bird in the ‘bander’s grip’; this technique is further described in Whittaker et al. (2010). Preen oil samples were stored frozen at −20 °C until analysis by gas chromatography–mass spectrometry (GC-MS). Whole blood samples ≤150 μl were obtained by brachial venipuncture, collected with heparinized capillary tubes, preserved in Longmire’s solution and refrigerated until DNA extraction. DNA was extracted from each sample using Maxwell RSC Whole Blood DNA Kits (Promega, Madison, WI, U.S.A.). Birds were sexed in the field by wing length and the presence of a brood patch. Because not all birds had developed a brood patch, we used PCR to confirm sex (see Appendix 1).

Parasitism Diagnostics

We used quantitative PCR (qPCR) to quantify relative amounts of parasite and host DNA from each blood sample. To quantify parasite DNA, we used primers 343F and 496R (Fallon et al., 2003), designed to amplify a conserved 154-bp region of mitochondrial rRNA genome found in Plasmodium, Haemoproteus and Leucocytozoon, the three major genera of haemosporidians. As a control for host DNA, we used primers SFSR3Fb and SFSR3Rb, which amplify a conserved, single-copy sequence of the vertebrate host genome (Asghar et al., 2012). All samples were run in triplicate 10 μl reactions, and the mean cycle threshold (‘Ct’) values were used in subsequent analyses. Each reaction included 5 μl PowerUP SYBR Green master mix (Applied Biosystems, Waltham, MA, U.S.A.), 3 μl of DNA solution (diluted to 30 ng/μl) and a final primer concentration of 0.8 μm. We used thermocycler conditions described in Arriero et al. (2018) on a QuantStudio 6 Flex thermocycler (Applied Biosystems). Standard curves were produced using five-step, twofold dilutions. The housekeeper and parasite primers showed approximately 95% and 108% efficiency, respectively.

Of the 28 birds sampled, 16 showed amplification during qPCR, with one showing unrepeatable Ct values among technical replicates (s > 0.5). We sequenced PCR products from each reaction to identify the primary genus amplified during qPCR (see Appendix 2, Supplementary material). Sequencing showed that 15 of the birds carried Plasmodium, while one bird carried Haemoproteus; see Appendix 2 and Supplementary material for sequencing data.

To calculate relative parasite load, we used a relative quantification (RQ) metric reflecting the proportion of parasite DNA relative to a control sample, which corrects for differences in primer efficiency (Pfaffl et al., 2001). We selected the sample with the lowest Ct value for the gene of interest (i.e. the highest amount of parasite DNA) as the control. The RQ is given by the formula RQ = E_parasite^{(control Ct−sample Ct)}/E_housekeeper^{(control Ct−sample Ct)}, where E is the efficiency of the primer specified. RQ values are therefore bounded by 0 (an uninfected bird) to 1 (a bird with the same amount of parasite DNA relative to host DNA as the control). Hereafter, ‘parasite load’ refers to RQ. To standardize data by parasite genus, we eliminated data from the Haemoproteus-infected bird from further analysis, producing a final sample of 27 birds. Ct values were repeatable for 14 of the 15 Plasmodium-infected birds, giving RQ values ranging from 0.49 to 0.999; RQ for the bird chosen as a reference sample was shifted to 0.999 to allow for beta regression. The volatile data from the single Plasmodium-infected bird with unrepeatable Ct data were included in subsequent statistical comparisons between infected and uninfected birds; however, these data were excluded from analyses of Plasmodium parasite load. All molecular diagnostics were conducted in the Center for Integrative Study of Animal Behavior (CISAB) core laboratory at Indiana University (Bloomington, IN, U.S.A.).

GC-MS Analysis

We used stir bar extraction followed by GC-MS to identify the volatile compounds in preen oil samples. The method has been previously described in Soini et al. (2007) and Whittaker et al. (2010). All compounds were identified by comparisons with the standard compounds from Sigma-Aldrich (St Louis, MO, U.S.A.). GC-MS peak areas in the preen oil samples were integrated by using post-run selected ion currents (SICs) and subsequently divided by the peak area of the internal standard for quantitative comparisons. Ion currents m/z 55, m/z 57, m/z 58 and m/z 60 were used for 1-alkanols, aldehydes, 2-ketones and linear carboxylic acids, respectively. Peak area of the internal standard in each sample (7-tridecanone from Sigma Aldrich, St Louis, MO, U.S.A.) was integrated by using SIC with m/z 113. All chemical analyses were conducted at Indiana University’s Institute for Pheromone Research (Bloomington, IN).

Statistical Analyses

We focused our analyses on preen oil volatile proportions, which we derived from peak area data given by GC-MS. To standardize the peak area data for each bird, the observed peak area for each volatile compound detected was divided by the corresponding peak area of the internal standard. The same 26 compounds were observed for most samples, although we did not detect one or more carboxylic acids and/or 7-methyloctadecane in six samples. For compounds that were not detected in all individuals, a trace value of 1/10 of the lowest observed value was added to each observation for that compound to enable statistical analysis of the full sample (Aitchison, 1982). We then converted peak area data to proportions, such that the sum of all compound proportions for each bird was equal to 1.

To reduce dimensionality of proportion data for the compounds detected by GC-MS, we used the ‘pcaCoDa()’ function from the ‘robComparisons’ package in R to perform a robust principal components analysis (PCA; Filzmoser et al., 2009). This method applies an isometric log-ratio transformation to proportion data before running a robust PCA; data are then backtransformed to centred log-ratio space for interpretation of results. PCA has been used previously to analyse odour blends in avian preen oil (i.e. Whittaker et al., 2010, 2013, 2018). Proportion data for all 27 birds were included in the PCA. We then used separate linear models for each PC to compare principal component scores in infected (N = 15, see above) and remaining uninfected juncos (N = 12). Each PC represents negative and positive loadings of different magnitude for all 26 compounds and therefore each PC can be considered as a distinct blend. We first used linear models to test whether each odour blend varied between uninfected and infected individuals. Focusing on just the infected class, we then used univariate beta regressions fitted with the ‘mgcv’ package and a logit link to ask whether odour blends (i.e. each PC) individually explained variation in parasite load (i.e. proportional RQ) among infected birds with parasite load data (N = 14; Wood, 2017). For PCs in which loading scores were significant predictors of parasite load, we ran post hoc tests to investigate which of the compounds loading heavily onto the PC (absolute loading scores greater than 0.10) varied in proportion with parasite load; we did this by using beta regressions to determine whether parasite load predicted proportion data for each compound.

Because seasonality can influence both parasite growth and preen oil volatile composition (Applegate & Beaudoin, 1970; Soini et al., 2007), we investigated the relationship between sampling date and these two factors. Within the breeding season, volatile concentrations peak about the time females lay the first egg of the season, after which concentrations decrease (Whittaker, Soini, et al., 2011). However, it is unclear whether the relative proportions of these compounds shift accordingly. Given that birds sampled later in the breeding season were more likely to be incubating eggs or nestlings, we used linear models to determine whether ordinal date was predictive of PC scores, using data from all birds (N = 27). We also used beta regressions to investigate whether (1) ordinal sampling date predicted parasite load in infected birds only and (2) the proportion of specific compounds noted above varied with parasite load (N = 14 for both analyses).

Finally, we investigated the relationship between parasitism and specific groups of volatile compounds predicted to be relevant to junco ecology, using beta regressions as described above. A previous study identified a group of compounds (2-tridecanone, 2-tetradecanone, 2-pentadecanone) negatively associated with seasonal reproductive success in female juncos in this study population (Whittaker et al., 2013). Therefore, we asked (1) whether the total proportion of these compounds varied significantly between infected and uninfected juncos and (2) whether the total proportion could be predicted by parasite load. We also asked whether sampling date predicted the total proportion of these compounds. Finally, we also used beta regressions to ask (1) whether the total proportion of aldehyde compounds varied between infected and uninfected juncos and (2) whether parasite load predicted total aldehyde proportion in infected birds.

Ethical Note

Field capture and sampling methods were approved by the Institutional Animal Use and Care Committees of Indiana University (protocol no. 18–030). Wild juncos were sampled during the annual census of a junco population near Mountain Lake Biological Station as part of a long-term study; therefore, unbanded juncos were given a silver U.S. Fish and Wildlife Service band as well as a unique combination of colour bands to allow visual identification (Federal Banding Permit no. 20261). During capture, mist nets were monitored frequently (<30 min intervals) and birds were briefly transported to a nearby laboratory (<3.2 km) for data and sample collection. The preen oil collection technique is brief (<3 min) and does not cause apparent distress. During capture and sampling, birds were monitored for signs of stress and injury. All birds were released at the location of capture following sampling (<1.5 h).

RESULTS

Overall Odour Composition and Parasitism

The robust PCA of compound proportion data produced three principal components (PCs) with eigenvalues greater than one, which together explained 74% of the variation in proportion data (Table 1); this result is similar to a previous study of junco preen oil odour composition, which showed 73% of variation in odour composition was explained in the first three PCs (Whittaker et al., 2013). Results of linear models showed principal component scores did not vary significantly by infection status for PC 1 (F_1,25 = 0.04, R²_adj = 0, P = 0.85) or PC 3 (F_1,25 = 0.37, R²_adj = 0, P = 0.55). However, PC 2 weakly and marginally varied with infection status (F₁₂₅ = 3.76, R²_adj = 0.10, P = 0.06), with lower mean scores in infected juncos ($\overline{x}$ = −0.40) than in uninfected ones ($\overline{x}$ = 0.55). Results of beta regressions showed no significant relationship between parasite load and odour PC 1 (deviance explained = 0.2%, β = −0.16, z = −1.6, P = 0.12) or odour PC 2 (deviance explained = 0.4%, β = 0.03, z = 0.1, P = 0.92). However, higher odour PC 3 scores were associated with higher parasite loads (deviance explained = 46.1%, β = 0.40, z = 2.7, P = 0.007).

Table 1.

Mean proportions ± SD are provided for volatile compounds detected in female dark-eyed junco preen oil (N = 27)

Volatile compound	Mean proportion (±SD)	PC 1 loading	PC 2 loading	PC 3 loading	β	z	Deviance explained	P (unadjusted)
1-Decanol	0.01 (±0.006)	0.02	0.13	0.24	1.13	1.4	15.7%	0.18
1-Undecanol	0.21 (±0.038)	−0.03	0.11	0.11	0.51	1.0	7.8%	0.33
1-Dodecanol	0.13 (±0.016)	−0.04	0.05	0.08
1-Tridecanol	0.14 (±0.011)	−0.06	0.03	0.07
1-Tetradecanol	0.13 (±0.019)	−0.07	0.02	0.06
1-Pentadecanol	0.08 (±0.014)	−0.1	0.04	0.02
1-Hexadecanol	0.19 (±0.043)	−0.12	0.08	0.01
1-Heptadecanol	0.03 (±0.008)	−0.12	0.12	0.11	1.22	1.9	22.7%	0.06
1-Octadecanol	0.01 (±0.007)	−0.05	0.09	0.55	3.63	3.8	54.9%	0.0002
2-Undecanone	0.0004 (±0.0002)	−0.1	−0.01	−0.09
2-Dodecanone	0.0006 (±0.0002)	−0.06	0	−0.15	−0.60	−0.8	5.9%	0.41
2-Tridecanone	0.0147 (±0.0049)	−0.02	0.07	0.08
2-Tetradecanone	0.0029 (±0.0011)	−0.05	0	−0.05
2-Pentadecanone	0.0208 (±0.0059)	−0.04	0.08	0.04
2-Hexadecanone	0.0008 (±0.0004)	−0.1	0.11	−0.16	0.09	0.1	0.2%	0.93
Dodecanoic acid	0.0013 (±0.0011)	0.15	−0.68	0.34	1.14	0.8	5.6%	0.40
Tetradecanoic acid	0.0045 (±0.0036)	0.03	−0.13	−0.25	−0.45	−0.5	1.9%	0.64
Hexadecanoic acid	0.0067 (±0.0061)	0.11	−0.04	−0.53	−1.45	−1.2	12.8%	0.24
Octadecanoic acid	0.0013 (±0.0012)	0.91	0.15	−0.01
2-Methyl-1-decanol	0.0003 (±0.0002)	0.09	0.05	−0.01
2-Butyl-1-octanol	0.0003 (±0.0002)	0.1	0.07	0.05
Undecanal	0.0011 (±0.0004)	−0.06	0.07	−0.05
Dodecanal	0.0004 (±0.0002)	−0.1	0.06	−0.11	−0.01	−0.01	0.1%	0.99
Tridecanal	0.0007 (±0.0003)	−0.12	0.06	−0.03
7-Methyl-octadecane	0.0003 (±0.0002)	−0.07	−0.62	−0.15	−1.67	−1.4	14.3%	0.16
2-Methyl-1-hexadecanol	0.0008 (±0.0004)	−0.09	0.1	−0.18	−1.00	−1.2	10.4%	0.23
Eigenvalue	4.33	2.10	1.40
Variance explained	41%	19%	14%
Cumulative variance	41%	60%	74%

Eigenvalues, percentage variance explained and variable loadings for a robust PCA of the relative proportion of volatile compounds detected in female dark-eyed junco preen oil (N = 27). For compounds with loadings greater than |0.10| for PC 3, results are provided for beta regressions of compound proportion on parasite load in infected birds (N = 14).

Individual Odour Compounds and Parasite Load

Mean ± SD parasite load was 0.75 ± 0.14. Of the 12 compounds with absolute PC 3 loadings greater than 0.10, parasite load was positively associated with 1-octadecanol proportions (deviance explained = 54.9%, β = 3.63, z = 3.8, P = 0.0002; Fig. 1). We found a marginally significant positive association between parasite load and 1-heptadecanol (deviance explained = 22.7%, β = 1.22, z = 1.9, P = 0.06). The remaining compounds showed no significant relationship with parasite load (all P > 0.1, Table 1). Following a Bonferroni correction for multiple comparisons, the model for 1-octadecanol remained significant (P = 0.002), but the model for 1-heptadecanol did not (P = 0.36).

Figure 1.

The proportions of volatile compounds (a) 1-octadecanol and (b) 1-heptadecanol in preen oil odour blends were regressed against parasite load (RQ) in Plasmodium-infected female dark-eyed juncos. Lines represent predicted values from beta regressions fitted with the ‘mcgv’ package in R and a logit link. Upper and lower bounds of the shaded areas represent 95% confidence intervals from the beta regressions. For uninfected birds, the mean proportion of 1-octadecanol was 0.02 ± 0 and the mean proportion of 1-heptadecanol was 0.03 ± 0.01.

Time of Sampling

We also detected a relationship between parasite load and sampling date, with higher parasite loads in birds sampled later in the breeding season (deviance explained = 47.1%, β = 0.11, z = 2.9, P = 0.004; see Fig. 2a). However, we found no relationship between sampling date and odour PC 1 (F₁₂₅ = 0.56, R²_adj = 0, P = 0.46) or PC 2 (F₁₂₅ = 0.32, R²_adj = 0, P = 0.58). Odour PC 3 increased with later sampling dates (F₁,₂₅ = 10.05, R²_adj = 0.26, P = 0.04; Fig. 2b), however sampling date did not predict 1-octadecanol (deviance explained = 16.7%, β = 0.05, z = 1.5, P = 0.12) or 1-heptadecanol proportions (deviance explained = 3.6%, β = 0.01, z = 0.7, P = 0.50).

Figure 2.

(a) Relationship between ordinal sampling date and relative parasite load (RQ). Dots represent data from 14 female dark-eyed juncos that tested positive for Plasmodium infection. (b) Relationship between ordinal sampling date and preen oil odour blend represented by principal component 3 (PC 3). Dots represent data from 27 female dark-eyed juncos. For both plots, upper and lower bounds of the shaded areas represent 95% confidence intervals from the linear model. For reference, an ordinal date of 118 corresponds to 18 April.

Ecologically Relevant Odours and Parasitism

The total proportion of 2-tridecanone, 2-tetradecanone and 2-pentadecanone did not vary between infected and uninfected juncos (deviance explained = 0.9%, β = −0.05, z = −0.5, P = 0.63). However, parasite load did predict the total proportion of these compounds in infected birds (deviance explained = 28.2%, β = 0.84, z = 2.1, P = 0.04; see Fig. 3a). Furthermore, the total proportion of these ketones increased with later sampling date (deviance explained = 29.0%, β = 0.02, z = 2.3, P = 0.02; see Fig. 3b).

Figure 3.

Proportion of preen oil odour blend associated with reproductive success (RS) and (a) parasite load and (b) sampling date. Dots represent data from 14 female dark-eyed juncos infected with Plasmodium. For both plots, upper and lower bounds of the shaded areas represent 95% confidence intervals from the linear model. For reference, an ordinal date of 118 corresponds to 18 April.

Aldehydes comprised a small percentage (0.1–0.4%) of the volatile fraction of preen oil, but infected juncos had marginally greater percentages of these compounds than uninfected juncos (deviance explained = 12.8%, β = 0.22, z = 1.9, P = 0.06). However, parasite load did not predict total aldehyde proportion within infected juncos (deviance explained = 3.1%, β = 0.42, z = 0.6, P = 0.53).

DISCUSSION

Ours is the first study to demonstrate a quantitative relationship between preen oil volatile composition and haemosporidian parasitism. We found that the proportions of a subset of volatile compounds (i.e. PC 3) varied with increasing parasite load, although the same PC 3 scores did not vary between uninfected and Plasmodium-infected birds. These results suggest that haemosporidian parasite load information is available in preen oil odour; however, further work is needed to determine whether ecologically relevant receivers, such as conspecific juncos or parasite vectors, are physiologically capable of detecting and integrating this information. These findings are particularly relevant considering recent studies of house sparrows and European blackbirds, Turdus merula, that showed no significant differences in the preen oil volatile composition of Plasmodium-infected and uninfected birds (Díez-Fernández, Martínez-de la Puente, Gangoso, López, et al., 2020; Diez-Fernandez et al., 2021). While the discrepancies between these studies and ours may be attributable to species level variation in host–parasite interactions, birds with lower parasite loads could also produce odours that are indistinct from those of uninfected birds. If so, odour composition may not be expected to vary strictly by parasite presence or absence if many birds in the infected class have low parasite loads; this is an important consideration, given that low loads are common in chronically infected wild songbirds (Asghar et al., 2012). Furthermore, these results suggest that the impacts of parasitism on host phenotype are better captured by quantitative data (i.e. parasite load) than by the presence/absence of infection alone.

Note that our qPCR methods used to assess parasite load targeted a conserved region of parasite DNA, such that our data are limited in specificity. Therefore, we cannot be certain that related haemosporidian parasites such as Haemoproteus or Leucocytozoon did not also contribute to our parasite load estimates in birds shown by sequencing to carry Plasmodium. Despite this limitation in specificity, detecting significant patterns in a modest sample size suggests that our results are not spurious and supports the possibility that haemosporidian parasite load information is available through avian preen oil odour to ecologically relevant receivers.

Interestingly, a common set of volatile compounds was detected among all sampled individuals, regardless of infection status or parasite load. Thus, variation in volatile composition was attributable to quantitative differences among a common set of compounds, rather than to the presence or absence of unique compounds. A common set of preen oil compounds was also detected in a study investigating preen oil wax ester composition in song sparrows, Melospiza melodia, before and after experimental exposure to Plasmodium. Preen oil wax composition shifted after parasite exposure, even in hosts that did not develop active infections; thus, the physiological response to Plasmodium, rather than parasite growth, was likely responsible for the change (Grieves et al., 2018). Given that a common set of volatile compounds tend to be detected in a species’ preen oil, regardless of infection status, it seems unlikely that haemosporidian parasitism yields unique volatile compounds in preen oil odour; rather, growth likely influences odour by perturbing existing physiological pathways within or on the host, such as metabolism, immune function, endocrine responses and/or microbial dynamics of the preen gland. For example, both mice and humans produce distinct odours in response to innate immune system activation by lipopolysaccharide injection, as do zebra finches, Taeniopygia guttata, given corticosterone implants (Gervasi et al., 2016; Kimball et al., 2014; Olsson et al., 2014). Additional work is therefore needed to determine whether the observed relationship between haemosporidian parasite load and preen oil odour is unique or whether the odour profiles seen in heavily infected juncos represent a general state of disease associated with elevated corticosterone levels (Love et al., 2016).

Olfactory perception is influenced not only by the identity of compounds present, but also by the relative proportions of individual compounds (Livermore & Laing, 1998). Furthermore, some relationships may be convex or concave, as opposed to linear. For example, the aldehyde nonanal is attractive to mosquitoes at relatively small proportions, yet it reduces mosquito activity when it comprises a larger proportion of a volatile mixture (Logan et al., 2008). Within the PC 3 odour blend, larger proportions of 1-octadecanol and 1-heptadecanol were present in the preen oil of juncos with higher parasite loads. Both are common preen oil volatiles across a diverse range of avian taxa, including New World warblers, blackbirds, woodpeckers, manakins, cardinals, American sparrows, mimic thrushes, parrots and ducks (Soini et al., 2013; Stenhagen & Odham, 1971; Whittaker et al., 2019). Both compounds are produced by aerobic bacteria endogenous to the preen gland. Specifically, Kocuria produces both compounds, while 1-heptadecanol is also produced by Paenibacillus and Pseudomonas. All three bacterial genera have been detected in the uropygial glands of many avian species (Whittaker et al., 2019). While additional studies are needed to determine whether 1-octadecanol and 1-heptadecanol vary with haemosporidian parasite load in other avian species, it is promising that these compounds (and the bacteria that produce them) have been detected among such a wide variety of avian taxa.

1-Octadecanol is one of several linear alkanols that increase in abundance in both male and female junco preen oil during the reproductive season, although 1-heptadecanol does not vary with reproductive condition in females (Soni et al., 2007). In budgerigars, Melopsitacus undulatus, larger proportions of 1-octadecanol are detected in male preen oil, and an alcohol blend containing larger proportions of 1-octadecanol is preferred by females in choice tests (Zhang et al., 2010). This suggests songbirds are likely capable of detecting differences in 1-octadecanol proportion, at least as one component of an odour blend. Similarly, 1-octadecanol is present in high proportions in the femoral secretions of the Algerian sand racer, Psammodromus algirus. In these lizards, larger proportions of 1-octadecanol are detected in males with lower haemogregarine blood parasite loads and higher immune responses (as assessed by cutaneous swelling in response to phytohemagglutinin injection), and females preferred such males in choice tests (Martín et al., 2007). It is unclear why the opposite pattern was observed in our study, with higher proportions of 1-octadecanol detected in the preen oil of female juncos with higher parasite loads. These different results might reflect distinct impacts of haemogregarine parasites on lizard host physiology when compared to haemosporidians in avian hosts, or indicate differences in physiological responses to parasitism based on host sex, as our study investigated female odour only.

In juncos, females can discriminate species and sex information in preen oil odour (Whittaker, Richmond et al., 2011). Past work has shown that a specific blend of volatile compounds (i.e. a blend of ketones) is associated with reproductive success in both males and females (Whittaker et al., 2013). Together, these studies indicate a high level of olfactory sensitivity in juncos, underlining the possibility that parasite load information detected in this study may be available to conspecifics (but see Grieves et al., 2020). Interestingly, our study showed larger ketone blend proportions later in the breeding season. In female juncos, this ketone blend is negatively associated with reproductive success, and thus our results suggest a broad shift in reproductive state may have occurred during the sampling period. Furthermore, this ketone blend, as well as odour PC 3 scores, were positively associated with parasite load; both blends also increased as the breeding season progressed. Therefore, it is possible that hormonal shifts accompanying this shift in reproductive state may have triggered an increase in parasite growth, as in the ‘spring relapse’ phenomenon (Applegate & Beaudoin, 1970). Further work is needed to determine whether parasite growth directly influences the production of ketone or PC 3 odour blends, or whether the relationships are correlative only; for example, controlling for breeding stage and sampling date in a larger sample of female juncos would help distinguish between these alternatives. If higher parasite loads are responsible for larger proportions of the ketone blend found in female preen oil odour, this is a potential mechanism for parasitism to reduce female junco reproductive success, assuming male juncos evaluate female ketone blend proportion during mate selection.

Aldehydes are another group of volatile compounds predicted to have ecological relevance for songbirds. Specifically, octanal and decanal repel mosquitoes at high concentrations, and such compounds are thought to serve as invertebrate repellants for birds such as the crested auklet, Aethia cristatella (Aldrich, 1988; Douglas, 2013; Douglas et al., 2005; Logan et al., 2008). Other aldehydes, such as nonanal, are known to attract mosquitoes that carry Plasmodium (Syed & Leal, 2009). Therefore, variation in aldehyde proportions could impact both the attractiveness and repellence of an individual bird’s odour blend to invertebrate vectors. Octanal, decanal and nonanal were all detected in this junco population in previous work (Soini et al., 2007). However, undecanal, dodecanal and tridecanal were the only aldehydes detected in our current study. The basis for this discrepancy is unclear; however, some variation in volatile composition could be attributable to differences in avian diet or photoperiod, as the birds in the 2007 study were kept in captivity. We also found higher total aldehyde proportions in the preen oil of infected versus uninfected juncos. However, there was no relationship between total aldehyde proportion and parasite load, and our data suggest individual aldehyde proportions also did not vary by parasite load (see Table 1). This indicates haemosporidian parasite load does not influence total aldehyde proportion, at least in female juncos; however, it is unclear why total aldehyde proportion varied between infected and uninfected juncos. The major metabolic sources of aldehydes are known to be due to oxidation of mobilized fatty acids. Additionally, some aldehydes can also be produced through peroxidation of membrane lipids (reviewed in Esterbauer et al., 1991), globally known as ‘oxidative stress’ and relating to a complex set of pathophysiological conditions. This could, in turn, include infection and inflammation. Another explanation that is not mutually exclusive is that juncos with larger total aldehyde proportions were more likely to be targeted by infected vectors; however, further information is needed about the influence of these specific aldehydes on vector feeding activity.

Having demonstrated that haemosporidian parasite load is reflected in the composition of volatiles extracted from avian preen oil, further studies exploring the ecological relevance of these compounds are warranted. Choice tests using either mosquitoes or conspecific songbirds as receivers could be conducted along with manipulations of the proportions of candidate compounds identified in this study (i.e. those loading heavily onto PC 3, including 1-octadecanol and 1-heptadecanol). However, it is important to remember that the odour profile of a live bird will vary from that of preen oil expressed directly from its uropygial gland, as the compounds are expected to degrade on feathers after preening (Leclaire et al., 2011). Therefore, 1-octadecanol and 1-heptadecanol might also serve as precursors to additional compounds that have ecological relevance.

Altogether, these results underscore a pressing need for studies exploring the influence of host endocrinology on the biosynthesis of preen oil. With this information, ecologists can further explore which environmental and endogenous factors are ultimately responsible for variation in odour as well as the downstream interspecific and intraspecific interactions that might be impacted. Such information will be broadly relevant to behavioural and disease ecologists alike.

Appendix 1

Sexing PCR

Birds were sexed in the field by wing length and the presence of a brood patch. However, brood patch data were not available for all birds, and not all phenotypically female individuals had developed a brood patch; therefore, we used PCR with primers P2 and P8 to confirm sex assignments (Griffiths et al., 1998). We ran 10 μi of PCR reactions with primers at 5 μM, dNTPs at 0.2 mM, magnesium chloride at 2 mM, 2 μl of GoTaq Green Flexi Buffer, 0.2 μl of GoTaq DNA polymerase and 0.5 μl of neat DNA. We ran products on 2% agarose gels stained with GelRed at 90 V for 2 h and visually confirmed that all 27 juncos were female.

Parasite DNA Sequencing

To identify the primary parasite genus being amplified by qPCR, we ran 20 μl PCR reactions for each infected bird in duplicate using primers 343F and 496 R. Each reaction included Promega PCR Master Mix (PR-M7501), as well as primers and template DNA in the same proportions as described for qPCR. The same thermocycler conditions specified for qPCR were used for PCR on a Nexus gradient thermocycler (Eppendorf). PCR products were separated by electrophoresis on 2% agarose gels stained with GelRed (Biotium, Fremont, CA, U.S.A.) at 90 volts for 30 min. Bands were excised and purified using a QIAquick gel extraction kit (Qiagen, Valencia, CA, U.S.A.). Purified PCR products were submitted for Sanger sequencing with primer 343F. We manually trimmed nonspecific calls from the ends of each sequence and analysed each sequence with a Blastn search (NCBI, Bethesda, MD, U.S.A.).

Data Availability

Raw data for this study are available on Zenodo (doi: 10.5281/zenodo.6456224) https://zenodo.org/record/6456224#.YnFNa5LMJ68