Conspicuous juvenile phenotypes are puzzling to evolutionary biologists. Why should organisms vulnerable to predation boldly broadcast their presence? We reconstructed the evolutionary history of nestling phenotypes across the estrildid finches (family Estrildidae) a radiation exhibiting unparalleled diversity in nestling ornamentation. Many are parasitised by Vidua finches whose offspring mimic host nestling phenotypes. We examined the role of brood parasitism, predation, sibling competition and signalling environment in the diversification of nestling ornamentation. We found that parasitised lineages exhibit elevated rates of nestling ornamentation evolution compared to unparasitised ones. Despite this, the extent to which nestlings were ornamented did not differ between parasitised and unparasitised lineages, contrasting with systems where coevolution proceeds at the egg stage and generates increased complexity in host traits. Species occupying denser habitats had increased ornamentation, suggesting a role for light environment in the evolution of begging displays. Nestling appearance showed strong phylogenetic signal, helping explain why successfully colonised hosts are often closely related to ancestral ones. Neither nest height nor clutch size (proxies for predation and sibling competition) predicted nestling ornamentation levels, and parasitism did not predict estrildid diversification rates. Overall, our results support a model of trait diversification in which host lead and parasites follow in the coevolutionary arms race.

Scoring estrildid nestling characters

We sourced photographs of the nestlings of as many estrildid finch species as possible and used them to score nestling appearance. Photos were sourced from G. Jelmer Huisman’s “Estrildid Finches of the World” (Huisman, 2021), Robert Payne’s monograph on estrildid mouth markings (Payne, 2005), Gerhard Hofmann’s online gallery of estrildid nestling mouth markings (www.hofmann-photography.de/index/02_estrild/index_estrild/mouthmark/) and Jamie et al.’s (2020) study on nestling estrildid finches from fieldwork in Zambia. These photos include birds reared both in captivity and the wild. Intraspecific variation in nestling appearance is extremely low in estrildid finch species (Jamie et al., 2020)and there is no evidence to suggest that birds reared in captivity have mouth markings that differ from those in the wild. Of the 138 species of estrildid finches, nestlings have been described for 109 species and we were able to source photos of 96 species.

To score nestling mouth markings, we decomposed them into following nine characters: number of medial palate spots (0, 1 or 2), number of lateral palate spots (0 or 2), number of mediolateral palate spots (0 or 2), presence of palate bar (absent or present), presence of a black ring around the mouth (absent or present), tongue markings (absent, spots, ring, broken ring, all dark, or swelling), type of swelling on upper gape (flange, arc or papillae), type of swelling on lower gape (flange, arc or papillae) and size of gape swelling (small, medium or large) (Figure 2 and Figure 3, Table S1). All nine characters are treated as categorical.

The full character matrix is included in the Supplementary Information. We decided not to use colour assessments as colour cannot be objectively measured or scored from non-standardised photographs.

To quantify the complex multi-dimensional nature of estrildid mouth markings, we used an “ornamentation index” calculated by summing across the 9 gape characteristics after standardising the above scoring scheme to yield a score from 0 to 1 for each trait (see Table S1) and a summed maximum ornamentation score of 9. For detailed comparison with the character matrix and ornamentation index used by Payne (2005), see the Supplementary Material.

Scoring nestling ecological characters

Host status

We assessed whether or not each estrildid species was a known host to a brood-parasitic indigobird or whydah (Vidua spp.) using the species accounts in Handbook of Birds of the World (Payne, 2010a; Payne, 2010b) as well as our own field experience studying these species (e.g. Sorenson et al., 2004; Jamie et al., 2020; Jamie et al., 2021). Species were scored as being either not a host (0), an occasional or secondary host that lacks a specialised parasite (1), or a primary host to a specialised parasite species (2). We ran analyses testing for associations with host status with a strict definition of host (only those classified as “2” counted as a host), and separately with a more lenient definition (both “1” and “2” counted as hosts).

Light environment

The light environment in which estrildid finches raise their young could influence the design of offspring-to-parent visual signals. Estrildid finch species primarily build domed nests with a circular side entrance, with some species also having a long entrance tunnel (Tarboton, 2011). In the absence of direct data measuring the light environment inside the nests of each estrildid species, we used the vegetation density of the species’ habitat as a proxy for the light environment of nestling signals. We used the same four-level categorical scale as Hu and Cardoso (2009), also adopted by Gomes et al. (2016). Habitats were divided into (1) open habitats (e.g., stone and semi-arid desert, gorges and rocky hills, dunes, open country, cultivated areas, savannah, parks, villages), (2) semi-closed with low vegetation (e.g., bushes, thickets, scrubs, thornscrub, reed-beds, swamps, rank vegetation, tall grass, mangroves, marshes, grassland), (3) semi-closed with high vegetation (e.g., forest edges and clearings, grassy areas with trees, acacias, bamboo thickets, woodland) and (4) closed (e.g., forest, lowland secondary-forest undergrowth). Habitat descriptions in Handbook of the Birds of the World (Payne, 2010a) were used to score each estrildid species. When species were reported to occur in habitats that fell between two categories, a 0.5 score was assigned, e.g. if habitat fell between “open” (1) and “semi-closed with low vegetation” (2), the species was scored as 1.5.

Sibling competition

Sibling competition is difficult to measure directly, and therefore several studies have used clutch size as a proxy (Kilner and Davies, 1998; Aviles et al., 2008; Soler and Aviles, 2010; Morales et al., 2019). We scored clutch size as the midpoint of the ranges reported in Handbook of Birds of the World (Payne, 2010a). Percentage of extra-pair young and rates of brood reduction (common in estrildids;  Hauber and Kilner (2007); Payne et al. (2001); Schuetz (2005)) are other proxies for sibling competition (e.g. Kilner, 1999) but these data are available for few estrildid species. It is important to acknowledge that, among Estrildidae, clutch size variation across estrildid finches is low (mean = 4.48±0.75) (based on Handbook of Birds of the World data), giving little power to detect a correlation with ornamentation and so a negative result should be interpreted in this context.

Predation pressure

We used nest height as a proxy for predation pressure, scored as the midpoint of the ranges reported in the Handbook of Birds of the World (Payne 2010). Ground nests generally experience higher predation levels than elevated nests (Martin, 1988; Martin, 1993). Therefore, the “predator warning hypothesis” predicts that transitions to ground nesting should be correlated with increases in nestling ornamentation. However, species experiencing higher predation rates could also be under selection to be less conspicuous, to reduce detection (e.g. by evolving nestling begging calls that are harder to locate (Briskie et al., 1999)). Therefore, the “predator avoidance hypothesis” predicts that transitions to ground nesting should be correlated with decreases in nestling ornamentation.

Evolutionary relationships

We carried out phylogenetically controlled analyses using the phylogeny of Gomes et al. (2016), which was based on mitochondrial DNA sequences and expanded on previously published estrildid phylogenies (Sorenson et al., 2003; Sorenson et al., 2004). It includes data from 254 estrildid finch samples representing 137 of the 138 estrildid species recognised by Clements et al. (2023), plus 33 outgroup samples. The one estrildid species not included in the phylogeny is New Hannover Mannikin (Lonchura nigerrima) the nestling of which species remains undescribed. Gomes et al. (2016) calibrated the phylogeny using the estimate of 15.69 million years for the divergence between Estrildidae and Viduidae (Gibb et al., 2015) which corresponds to 11.71 million years for the common ancestor of estrildid finches. For further details on mtDNA data and tree reconstruction, see Gomes et al. (2016).

Reconstructing ancestral states

We reconstructed ancestral states for continuous traits using the fastAnc function from the R package phytools under a Brownian Motion model of evolution (Revell, 2012). Ancestral states for categorical traits were reconstructed with marginal state reconstruction using the ancr function from phytools. The fitMk function from phytools was used to fit models in which transition rates between character states were set as all equal, all different or symmetric (for categorical traits with >2 levels) and the appropriate model chosen through AIC comparison.

Calculating phylogenetic signal

For continuous traits, we quantified phylogenetic signal using Blomberg’s K parameter (Blomberg et al., 2003) and Pagel’s λ (Pagel, 1999). We calculated K using the phylosig function (method=’K’) from the R package phytools (Revell, 2012). For estrildid mouth marking traits, λ was calculated using the phylosig function (method=”lambda”) from the R package phytools (Revell, 2012). For binary traits, we used the D statistic of Fritz and Purvis (2010), to estimate phylogenetic signal. We calculated D using the function phylo.d from the R package caper (Orme et al., 2013). Stand-alone methods for estimating phylogenetic signal in categorical traits with more than two levels do not exist in standard packages, and so phylogenetic signal was not estimated for traits of this type. Instead, we treated ordered categorical variables (gape swelling size, number of mediolateral spots) as continuous to calculate K and λ values.

Phylogenetic generalised least squares regression

We tested ecological hypotheses in a single model using the pgls function from the package “caper” (Ormeet al., 2013) in which we modelled ornamentation index as a function of the species’ host status, clutch size, nesting habitat and nest height. The analysis included the 96 estrildid species with complete information on ornamentation index and the four ecological factors.

Analysing rates of evolutionary change

We tested for an association between parasitism and rates of mouth marking evolution using Ornstein-Uhlenbeck models in the R package “OUwie” (Beaulieu and O'Meara, 2016), with ornamentation level as the response variable. Whereas Brownian Motion models assume a random walk over character space, Ornstein-Uhlenbeck models allow for evolution to move around some central trait value (q) at a certain rate (s) and with a certain “pull” (a) back towards that central value. When there is no pull (a=0), the OU model simplifies to a Brownian Motion model. We compared the likelihoods of models in which each of the three parameters (q, s, a) can either have different values for parasitised versus unparasitised lineages or are constrained to have the same value for both states. By comparing the likelihood of these models, we tested whether parasitism is associated with different rates of mouth marking evolution.

Analysing diversification rates

We investigated the association between parasitism and speciation rates in estrildid lineages using the R package “diversitree” (FitzJohn, 2012). We compared the log-likelihood of a model in which speciation rates can differ depending on a species’ host status, with one in which speciation rates are constrained to be equal across character states.