Limits to host colonisation and speciation in a radiation of parasitic finches

Transfer experiments simulating host colonisation During January–April 2014–2017, we carried out transfer experiments within an area of about 40 km2 on and around Musumanene and Semahwa Farms (centred on 16°47′S, 26°54′E) in the Choma District of Zambia. The habitat is a mixture of miombo woodland, agricultural fields and seasonally-flooded grasslands. The experiment had three treatments: (i) pin-tailed whydah eggs transferred from common waxbill to blue waxbill nests, (ii) blue waxbill eggs transferred to blue waxbill nests, and (iii) common waxbill eggs transferred to blue waxbill nests. Additionally, we monitored survival and feeding of pin-tailed whydah nestlings in naturally-parasitised common waxbill nests. To minimise predation, eggs were taken from their natural nest and incubated in a Brinsea Octagon 20 Advance EX Incubator at 36.7°C and 60% humidity. A day before they were due to hatch, eggs were fostered to a blue waxbill nest. Occasionally (16/94 transfers), the offspring had to be fostered as a chick freshly hatched in the incubator, rather than as an egg but this was not found to influence the offspring’s subsequent survival in the novel nest environment (see Results). The modal clutch size of both common waxbill and blue waxbill nests was five. No host egg was removed when an egg/hatchling was added, because pin-tailed whydah females do not remove a host egg when they naturally parasitise a nest (Tarboton, 2011). Experimental nests were visited every two or three days, and the number of eggs and chicks in the nest was recorded. For chicks, mass and tarsus length were measured and the amount of food in the crop scored (methods below). Mass was measured on digital scales to an accuracy of 0.1 or 0.01 g depending on the model of scale. Tarsus length was measured using dial callipers to the nearest 0.1 mm. All three species used in the study are common and widespread and experience high levels of natural nest predation such that the experiments carried out for this study will have negligible effects on their populations. Data were collected under the research approval of the Department of National Parks and Wildlife in Zambia (DNPW/8/27/1). Our sample sizes were chosen to allow the between-species effects to be detected with a high degree of confidence while not including an unnecessarily high number of individuals. As such it meets the ABS/ASAB guidelines and adheres to the three R’s of replacement, reduction and refinement (Buchanan et al., 2012). Comparing survival of different species transferred to blue waxbill nests Survival analyses were carried out in the R statistical environment(R Development Core Team, 2017) using the packages Survival (Therneau, 2015) and KMsurv (Moeschberger and Yan, 2012). We monitored chick survival from the day the chick hatched in the new host nest. Chick survival was judged to end at the midpoint between the last day the chick was known to be alive and the first day the chick was known to be absent. If the nest was still active, but the transferred chick was absent, then the chick was assumed to have died. If the nest was abandoned at the point the transferred chick was absent, then the data were right-censored. Right-censoring is used when the event of interest has not occurred by the last observation (Mills, 2011). A Cox proportional hazards model was fitted to the survival data (Cox, 1972). The co-variates in the initial model were: (i) transferred chick species, (ii) presence of host nestmates, and (iii) foreign chick transferred as egg or as chick. The number of nestmates over the course of a given transfer experiment ranged from 0 to 5 (mean = 1.4). In most transfers (49 of 94), the transferred chick had no nestmates in the foster nest. We therefore modelled nestmate presence as presence/absence. Comparing the amount of food host parents fed transferred chicks of different species Crop size of the transferred chick was recorded at each nest visit. Nestling estrildid crops are transparent, allowing easy visual inspection. Crops were scored as 0 (empty), 1 (trace amounts, < c. 20 seeds, no bulge in crop), 2 (> c. 20 seeds, slight bulge) or 3 (> c. 50 seeds, large bulge). To assess whether crop sizes of chicks differed depending on the species of chick transferred, two approaches were used: First, the median crop size of the transferred chick over the first 7 days of survival in the host nest was used as the response variable; c. 80% of common waxbill and pin-tailed whydah chick mortality occurred in this period (Figure 2). A Kruskal-Wallis test was used to test whether median crop size differed between the three species. Median crop sizes were compared between species using a Dunn post-hoc test, using the dunn.test function from the R package dunn.test and with Bonferroni correction for multiple testing (Dinno, 2017). Second, ordinal mixed-effect models were implemented in the R package Ordinal (Christensen, 2015) using crop score as an ordinal response variable. In the full model, the fixed effects were chick species and the number of nestmates. Transferred chick individual was a random effect nested within the nest of origin of that transferred chick. We carried out stepwise elimination of non-significant co-variates until only significant co-variates remained. The model was initially run to include crop scores over the first 7 days of development, then re-run using crop scores over the first 4, 5, 6 and 8 days of development to test whether the findings from the first 7 days of development were robust. Ordinal mixed-effect models were used to compare crop size in pin-tailed whydahs and common waxbills occurring in their natural nests (common waxbill nests) with those experimentally transferred to blue waxbill nests. Data for pin-tailed whydahs and common waxbills in common waxbill nests are observational, unlike the experimental data from pin-tailed whydahs in blue waxbill nests. This was because high levels of nest predation meant that all common waxbill nests found at the egg stage were required as a source of eggs for transfer to blue waxbill nests. Therefore, our data on pin-tailed whydahs in common waxbill nests do not account for the effects of transferring an egg from one nest to another. However, when blue waxbill eggs were transferred from their own nest to another blue waxbill nest, they still showed high survival (about 76% of all chicks transferred survived to fledging), suggesting that any effects of transferring eggs between nests are insufficient to account for the differences observed in chick survival between common and blue waxbill nests. Measuring nestling diet composition Obtaining crop samples in field Nestling crops were sampled using the tube insertion method (Zann and Straw, 1983). The tube was inserted in the throat of the nestling and seeds were pushed from the translucent crop into the tube. The contents were stored in 70% ethanol. The process was repeated until about 20–30 seeds had been extracted. Chicks were sampled around the time when the primaries first erupt from pin (approximately 6–7 days of age). We sampled crops of common waxbill, blue waxbill as well crops of six other sympatric estrildid finch species: orange-winged pytilia (Pytilia afra), melba finch (Pytilia melba), Jameson’s firefinch (Lagonosticta rhodopareia), red-billed firefinch (L. senegala), African quailfinch (Ortygospiza atricollis) and bronze mannikin (Spermestes cucullatus). We sampled these other estrildid finch species in addition to the two used in the experiments to assess variation in host diet across a broader phylogenetic scale and to explore whether estrildids that hosts to Vidua have a different diet from those that are not. DNA barcoding of nestling crop contents Nestling estrildid finch crops contained almost exclusively plant seeds. DNA barcoding of samples was carried out by Jonah Ventures (Boulder, Colorado; jonahventures.com). The chloroplast trnL intron was amplified from DNA in each sample using the c and h trnL primers (Taberlet et al., 2007). The total expected amplicon length was 332bp (Jonah Ventures in litt.). A detailed protocol is described in supplementary methods. We consulted with an expert botanist based in Zambia, Mike Bingham, to validate the taxonomic identities assigned by DNA barcoding. Quantifying crop contents DNA barcoding data resolution allowed analysis at the subfamily level and not the genus level, so each OTU was assigned to one of the four subfamilies identified (see Results). For each sample, reads from OTUs mapping to the same sub-family were summed together to give a measure of the total number of reads from each subfamily, and expressed as a proportion of total reads (Craine et al., 2015; Willerslev et al., 2014). To test whether different estrildid species fed chicks different proportions of seeds from each of the four families, non-metric dimensional scaling (NMDS) was performed using the R package vegan (Oksanen et al., 2017). Comparisons of diet between species were made using the function adonis, from the R package vegan. Begging call plasticity Recording nestling begging calls Chicks were placed in an artificial nest, and given several minutes for acclimation. To stimulate begging, the chick was tapped gently with forceps on the bill. Recordings were made using an Audio-Technica ATR35S tie-clip microphone or a Sennheiser ME-66 shotgun microphone held approximately three cm away from the chick’s mouth. Recordings were made for around two minutes or until at least ten seconds of continuous begging were recorded. Analysing the effect of host environment on nestling begging calls We compared begging calls of nestling pin-tailed whydahs in natural common waxbill nests, to those transferred to blue waxbill nests. We identified four distinct call types both by listening to recordings and through visual inspection of sonograms (see Results). All four call types were detected in both pin-tailed whydahs developing in common waxbill nests, and pin-tailed whydahs transferred to blue waxbill nests. To analyse whether host environment influenced the stage in the nestling period at which each call type was made, we examined the stage in development at which chicks made each call type and compared this between pin-tailed whydahs raised in common and blue waxbill nests. We used chick tarsus length as a proxy for developmental stage, because for pin-tailed whydahs in their natural nests, the exact age in days of the chick was unknown, whereas tarsus length was available for all treatments. We examined whether within each call type, there were changes in call structure between host environments. For each call type, the following begging call parameters were extracted from each recording: minimum frequency, maximum frequency, centre frequency, peak frequency, frequency bandwidth, call duration, average entropy, and energy. These are widely used in the literature to characterise begging sounds (Anderson et al., 2009; Butchart et al., 2003; Langmore et al., 2008). For each recording, ten sequential call notes in a bout of begging were selected and call parameters extracted. Call notes were not selected if they overlapped with interfering background noises, or if they were incomplete calls. The relationship between the call types was visualised using linear discriminant function analysis with the R package MASS (Venables and Ripley, 2002). We calculated call rate by dividing the dividing the number of calls in the bout by the duration of the bout. Two approaches were used to compare the structure of each call type between pin-tailed whydahs raised in common waxbill nests, and those raised in blue waxbill nests. First, a series of linear mixed models were constructed, with each call parameter as a separate response variable. Host environment and crop size were fitted as fixed effects and individual chick identity as a random effect. Crop size was used as a proxy for chick hunger. When estrildid finch nestlings are fed, they store their seed in the crop before passing it on to the stomach. By measuring the amount of food stored in the crop, we could assess how much the chick had recently been fed in a non-invasive manner. Crops were scored as 0 (empty), 1 (trace amounts, < c. 20 seeds, and with no bulge in crop), 2 (> c. 20 seeds and with slight bulge) or 3 (> c. 50 seeds and with large bulge). We controlled for multiple testing using Bonferroni correction (Dunn, 1961). Second, we carried out a logistic regression analysis using the R package nnet (Venables and Ripley, 2002), allowing all call parameters to be considered at once.

模拟宿主定殖的转移实验
2014年至2017年的1-4月期间，我们在赞比亚乔马区Musumanene农场与Semahwa农场及其周边约40平方公里的区域内（中心坐标为16°47′S, 26°54′E）开展了系列转移实验。该区域生境由密奥姆博林地（miombo woodland）、农田与季节性泛滥草原混合构成。本实验设置3组处理：(i) 将针尾维达雀（pin-tailed whydah）卵从普通梅花雀（common waxbill）巢转移至蓝梅花雀（blue waxbill）巢；(ii) 将蓝梅花雀卵转移至蓝梅花雀巢；(iii) 将普通梅花雀卵转移至蓝梅花雀巢。此外，我们还对自然寄生在普通梅花雀巢中的针尾维达雀雏鸟的存活与取食情况进行了监测。

为降低捕食风险，实验人员从天然巢中取出卵，使用Brinsea Octagon 20 Advance EX孵化器在36.7℃、相对湿度60%的条件下进行孵化。在雏鸟即将出雏的前一日，将卵寄养至蓝梅花雀巢中。少数情况下（94次转移中共计16次），需将刚在孵化器中孵出的雏鸟而非卵进行寄养，但后续分析显示该操作并未影响雏鸟在新巢环境中的存活（详见结果部分）。普通梅花雀与蓝梅花雀的典型窝卵数均为5枚。由于针尾维达雀雌鸟自然寄生巢时不会移除宿主卵（Tarboton, 2011），因此在添加卵或雏鸟时不会移除宿主巢内原有卵。实验巢每2至3日巡查一次，记录巢内卵与雏鸟的数量。对于雏鸟，需测量其体重与跗跖长度，并对嗉囊内的食物量进行评分（测量方法详见下文）。体重使用电子天平测量，精度可达0.1g或0.01g，依天平型号而定；跗跖长度使用游标卡尺测量至最近0.1mm。

本研究涉及的三个物种均为常见广布种，自然巢捕食压力较高，因此本实验对其种群的影响可忽略不计。数据采集工作获得了赞比亚国家公园与野生动物部（Department of National Parks and Wildlife, DNPW）的研究许可（许可编号：DNPW/8/27/1）。样本量的设置旨在以高置信度检测物种间的效应差异，同时避免纳入过多不必要的个体，符合ABS/ASAB指南，并遵循动物实验3R原则（替代、减少、优化，Buchanan et al., 2012）。

比较转移至蓝梅花雀巢的不同物种的存活情况
生存分析在R统计环境（R Development Core Team, 2017）中开展，使用了Survival包（Therneau, 2015）与KMsurv包（Moeschberger and Yan, 2012）。监测从雏鸟在新宿主巢出雏当日开始，雏鸟的存活时间判定为最后一次确认存活日期与首次确认失踪日期的中点。若巢仍处于活跃状态但转移的雏鸟失踪，则假定该雏鸟死亡；若雏鸟失踪时巢已被弃用，则该数据为右删失数据（右删失指在最后一次观测时仍未发生目标事件，Mills, 2011）。本研究拟合了Cox比例风险模型（Cox, 1972），初始模型包含3个协变量：(i) 转移雏鸟的物种；(ii) 宿主巢内同伴的存在情况；(iii) 外来雏鸟以卵还是雏鸟的形式进行寄养。单次转移实验中，巢内同伴的数量范围为0至5（均值为1.4）；在多数转移实验中（94次转移中的49次），转移雏鸟无巢内同伴，因此我们将巢同伴存在情况建模为存在/缺失二分类变量。

比较宿主亲鸟饲喂不同物种转移雏鸟的食物量
每次巢巡查时，均记录转移雏鸟的嗉囊大小。梅花雀科（Estrildidae）雏鸟的嗉囊呈半透明状态，便于目视检查。嗉囊评分标准如下：0分（空）、1分（微量食物，<约20粒种子，嗉囊无隆起）、2分（>约20粒种子，嗉囊轻微隆起）、3分（>约50粒种子，嗉囊明显隆起）。本研究采用两种方法分析嗉囊食物量与物种的关系：第一种，以雏鸟在宿主巢存活前7天的中位嗉囊大小作为响应变量——约80%的普通梅花雀与针尾维达雀雏鸟死亡均发生于此阶段（图2）。使用Kruskal-Wallis检验比较三个物种间的中位嗉囊大小差异，随后采用Dunn事后检验结合Bonferroni校正进行物种间的多重比较（使用R包dunn.test中的dunn.test函数，Dinno, 2017）。第二种，使用R包Ordinal（Christensen, 2015）构建序数混合效应模型，以嗉囊评分作为序数响应变量；全模型的固定效应为雏鸟物种与巢同伴数量，转移雏鸟个体作为随机效应，嵌套于该雏鸟的原巢。通过逐步剔除不显著的协变量，直至仅保留显著协变量。模型首先使用雏鸟发育前7天的嗉囊评分运行，随后分别使用前4、5、6、8天的嗉囊评分重新运行，以检验前7天分析结果的稳健性。我们还使用序数混合效应模型，比较了针尾维达雀与普通梅花雀在天然巢（普通梅花雀巢）与实验转移至蓝梅花雀巢中的嗉囊大小。普通梅花雀巢中的针尾维达雀与普通梅花雀数据为观测数据，而非实验数据，原因是高巢捕食压力导致所有发现的卵期普通梅花雀巢均需作为卵源，用于转移至蓝梅花雀巢，因此针尾维达雀在普通梅花雀巢中的数据未考虑巢间卵转移的效应。但当蓝梅花雀卵从自身巢转移至另一蓝梅花雀巢时，仍表现出较高的存活率（约76%的转移雏鸟存活至离巢），表明巢间卵转移的效应不足以解释普通梅花雀与蓝梅花雀巢间雏鸟存活的差异。

雏鸟日粮组成分析
1. 野外获取嗉囊样本
使用插管法采集雏鸟嗉囊内容物（Zann and Straw, 1983）：将插管插入雏鸟喉部，将半透明嗉囊中的种子推入管内，采集到的内容物储存在70%乙醇中，重复操作直至提取约20至30粒种子。采样时间约在初级飞羽刚从羽鞘长出时（雏鸟约6至7日龄）。采样对象包括普通梅花雀、蓝梅花雀，以及另外6种同域分布的梅花雀科雀类：橙翅斑腹雀（Pytilia afra）、梅利巴雀（Pytilia melba）、詹姆森火雀（Lagonosticta rhodopareia）、红嘴火雀（L. senegala）、非洲鹌鹑雀（Ortygospiza atricollis）和铜色文鸟（Spermestes cucullatus）。除实验所用的两个物种外，额外采样这些梅花雀科物种，旨在更广泛的系统发育尺度上评估宿主饮食变异，并探究作为维达雀属（Vidua）宿主的梅花雀与非宿主的饮食是否存在差异。

2. 嗉囊内容物DNA条形码分析
雏鸟嗉囊内容物几乎仅包含植物种子。样本的DNA条形码分析由Jonah Ventures（科罗拉多州博尔德；jonahventures.com）完成。使用c和h trnL引物扩增每个样本DNA中的叶绿体trnL内含子（Taberlet et al., 2007），预期扩增子总长度为332bp（Jonah Ventures, 私人通信）。详细实验方案见补充方法。我们咨询了赞比亚的植物学专家Mike Bingham，以验证DNA条形码鉴定的分类学身份。

3. 嗉囊内容物定量分析
DNA条形码数据的分辨率仅可至亚科水平，无法达到属水平，因此每个操作分类单元（Operational Taxonomic Unit, OTU）被分配至已鉴定的四个亚科之一（详见结果部分）。对每个样本，将映射至同一亚科的OTU读数总和作为该亚科的总读数，并表示为总读数的比例（Craine et al., 2015; Willerslev et al., 2014）。为检验不同梅花雀物种饲喂雏鸟的各亚科种子比例是否存在差异，使用R包vegan（Oksanen et al., 2017）进行非度量多维尺度分析（NMDS），并使用vegan包中的adonis函数进行物种间饮食组成的比较。

乞食鸣唱可塑性
1. 记录雏鸟乞食鸣唱
将雏鸟置于人工巢中，给予数分钟适应时间。为刺激雏鸟发出乞食鸣唱，用镊子轻触雏鸟的喙部。使用Audio-Technica ATR35S领夹式麦克风或Sennheiser ME-66枪式麦克风，距离雏鸟口部约3cm进行录音，录音时长约2分钟或直至记录到至少10秒的连续乞食鸣唱。

2. 分析宿主环境对雏鸟乞食鸣唱的影响
比较天然普通梅花雀巢中的针尾维达雀雏鸟与转移至蓝梅花雀巢中的针尾维达雀雏鸟的乞食鸣唱。通过聆听录音与声谱图目视检查，共鉴定出4种鸣唱类型（详见结果部分），且两种宿主环境下的针尾维达雀均能发出这4种鸣唱类型。为分析宿主环境是否影响每种鸣唱类型出现的雏鸟发育阶段，我们检查了雏鸟发出每种鸣唱类型时的发育阶段，并比较了普通梅花雀巢与蓝梅花雀巢中抚育的针尾维达雀的该指标。我们使用跗跖长度作为发育阶段的替代指标，因为天然巢中的针尾维达雀雏鸟确切日龄未知，但所有处理组均有跗跖长度数据。我们还检查了每种鸣唱类型在不同宿主环境下的鸣唱结构是否存在差异。针对每种鸣唱类型，从每个录音中提取以下乞食鸣唱参数：最低频率、最高频率、中心频率、峰值频率、频率带宽、鸣唱时长、平均熵与能量——这些参数在相关文献中被广泛用于表征乞食声音（Anderson et al., 2009; Butchart et al., 2003; Langmore et al., 2008）。对每个录音，选取乞食鸣唱序列中连续的10个鸣唱音节，提取相关参数，避开与背景噪音重叠或不完整的鸣唱音节。使用R包MASS（Venables and Ripley, 2002）中的线性判别函数分析可视化鸣唱类型间的关系。通过将鸣唱次数除以鸣唱总时长，计算鸣唱率。本研究采用两种方法比较普通梅花雀巢与蓝梅花雀巢抚育的针尾维达雀的每种鸣唱类型结构：第一种，构建一系列线性混合模型，将每个鸣唱参数作为单独的响应变量，宿主环境与嗉囊大小作为固定效应，雏鸟个体身份作为随机效应。嗉囊大小作为雏鸟饥饿程度的替代指标——梅花雀科雏鸟取食后会将种子储存在嗉囊中再送入胃内，通过测量嗉囊内储存的食物量，可无创评估雏鸟近期的取食情况。嗉囊评分标准同前。使用Bonferroni校正控制多重检验（Dunn, 1961）。第二种，使用R包nnet（Venables and Ripley, 2002）进行逻辑回归分析，同时纳入所有鸣唱参数进行分析。