High-intensity flight feather molt and comparative molt ecology of warblers of eastern North America

Rapid high-intensity molt of flight feathers occurs in many bird species, and can have several detrimental consequences, including reductions in flight capabilities, foraging performance, parental care, and plumage quality. Many migratory New World warblers (family Parulidae) are known to have intense remigial molt, and recent work has suggested that simultaneous replacement of the rectrices may be widespread in the family as well. However, the phylogenetic distribution of simultaneous rectrix molt, and high-intensity flight feather molt more generally, has not been systematically investigated in warblers. We addressed this issue by examining flight feather molt in 13 species, representing 7 different warbler genera, at Powdermill Avian Research Center in southwestern Pennsylvania, USA. All 13 species replaced their 12 rectrices simultaneously, with the onset of rectrix molt occurring in the early-middle stages of high-intensity primary molt. As expected, single-brooded early migrants molted earlier than double-brooded species whose nesting activities extend into late summer. However, our finding that late-molting species replaced their primaries more slowly and less intensively than early-molting species was unexpected, as late-molting species are widely hypothesized to be under stronger migration-related time constraints. This surprising result appears to be at least partially explained by a positive association between pace of molt and daylength; shorter late-summer days may mandate reduced daily food intake, lower molt intensity, and a slower pace of molt. In comparison to other passerines, flight feather molt in warblers of eastern North America is extraordinarily intense; at its peak, individuals are simultaneously replacing 50-67% of their 48 flight feathers (all 12 rectrices and 6-10 remiges on each wing) for 2-3 weeks or more. Because molt of this intensity is likely to present numerous challenges for flight, avoiding predators, foraging, and parental care, the period of flight feather molt for warblers constitutes a highly demanding phase of their annual cycle.

Methods
Field methods

Molt data were collected during summer and fall banding operations, 1986-2000, at Powdermill Avian Research Center (40.1637ºN, 79.2674ºW) in the Laurel Highlands of Westmoreland County, southwestern Pennsylvania, USA. Molt scores were obtained from 1289 individual captures of 13 different species representing 7 different parulid genera (Table 1). Because all 13 focal species nest regularly at Powdermill and surrounding areas of eastern Westmoreland County (Brauning 1992, Wilson et al. 2012), individuals captured during molt were likely local breeders on or near their nesting grounds (Pyle et al. 2018). In all 13 focal species, birds were captured at all stages of prebasic molt, and we found no evidence that individuals arrested or suspended molt prior to migration. All 18 remiges on the right wing (9 primaries, 6 secondaries, 3 tertials) and the 6 right rectrices were scored using the 0-5 molt scoring system of the British Trust for Ornithology: 0 = old feather, 1 = old feather missing or new pin feather, 2 = new feather emerging from sheath up to one third grown, 3 = new feather between one and two thirds grown, 4 = new feather more than two thirds grown but still sheathed at base, and 5 = fully grown new feather with no trace of sheath at its base (Ginn and Melville 1983).

Data analysis and statistical methods

For each individual we calculated a total primary molt score (0-45) and a total rectrix molt score (0-30) by summing the individual scores for each feather. To examine interspecific variation in the degree to which individuals molted all their rectrices simultaneously, we examined rectrix molt data from 582 individuals of all 13 species captured during active tail molt (total rectrix molt score > 0 and < 30). For each individual we calculated two indices of synchrony in rectrix molt: (1) the maximum difference in score between any two rectrices of any molt score (MaxDiff; 0 = completely synchronous rectrix molt, 5 = completely staggered rectrix molt), and (2) the standard deviation of molt score for the 6 scored rectrices (SD; 0 = completely synchronous, 1.87 = completely staggered). Because neither MaxDiff nor SD was normally distributed, we made between-species comparisons using Kruskal-Wallis and Dunn’s tests, nonparametric equivalents of ANOVA and post-hoc multiple comparisons. JMP Pro 12.2 (SAS Institute, Cary, North Carolina, USA) was used for all statistical calculations.

To determine the status of primary molt at the onset of rectrix molt, we examined primary feather molt scores for 126 individuals of 10 species that were captured during the initial early stages of rectrix molt (total rectrix molt score > 0 and ≤ 6). We made between-species comparisons of total primary molt score for these 10 species using ANOVA and Tukey-Kramer multiple comparison tests, as examination of normal quantile plots of residuals indicated that total primary molt scores were approximately normally distributed.

We explored interspecific variation in the timing and duration of molt by focusing on primary molt. Primary molt in parulid warblers follows the standard passerine pattern, with loss of the first (innermost) primary invariably indicating the start of flight feather molt, with sequential loss of the remaining primaries occurring in a regular proximal-to-distal pattern (Foster 1967, Nolan 1978, Rimmer 1988). Because of the regularity and predictability of primary molt, and because the period of primary replacement nearly always encompasses molt of the rectrices, secondaries, and tertials as well (e.g., Foster 1967), our focus on the timing and duration of primary molt is appropriate.

We estimated the day of the year when the midpoint of primary molt was achieved by fitting 3-parameter logistic models to the total primary molt score data for each species, using the Nonlinear Platform of JMP Pro 12.2. We focused on estimating the midpoint (or halfway date; see Jackson 2017, Erni 2018) rather than the onset of primary molt because the estimated midpoint was much less sensitive to curve-fitting assumptions; a variety of linear and non-linear curve-fitting approaches — including Pimm regression (Pimm 1976), 3-parameter logistic models, cubic splines, and kernel smoothing — all produced similar estimates of the midpoint of primary molt. With the fitted logistic models, we then used JMP’s Inverse Prediction tool to estimate the mean ± SE day of the year that the midpoint of primary molt (primary molt score = 22.5) was achieved for each species.

We estimated the pace of primary molt for each species from total primary molt scores of individual birds captured two or more times during the same season while in primary molt (primary molt score > 0 and < 45). We favored this approach over alternatives such as Pimm regression (Pimm 1976) and Underhill-Zucchini maximum likelihood models (Underhill and Zucchini 1988, Erni et al. 2013) because it allowed us to examine the pace of molt for individual birds and also avoid problems of meeting assumptions of regression and maximum likelihood models (Newton 2009, Rohwer 2013). For each species, we fit a general linear model to total primary molt score with two fixed effects: day of the year as a continuous variable, and band identification number as a categorical variable. The coefficient (mean ± SE) for day of the year in these models represented an overall estimate of the slope of the relationship between primary molt score and day (change in molt score per day).  All models were created using the Fit Model platform of JMP Pro 12.2. With estimates of both the midpoint and pace of primary molt, we were able to estimate the mean date of onset of primary molt and mean duration of primary molt for each species. A total of 309 molt records from 137 individual bird-seasons and 12 species were included in these analyses; recapture data for Blue-winged Warblers (BWWA) were too limited to use this approach, and this species was not included in the analysis of timing and pace of primary molt. Recapture data for Canada Warbler (CAWA) were also limited, but because recaptures spanned almost the entire range of total primary molt scores (Supplemental Material Figure S1), this species was included in the analysis. Our estimates of midpoint and pace of primary molt, however, should be viewed cautiously for this species.

For all 13 species we calculated two species-level estimates of primary molt intensity: (1) Peak Intensity, which we defined as the mean number of primary feathers on the right wing being replaced by individuals with total primary molt scores > 15 and ≤ 35; our estimate of peak intensity therefore excludes individuals in the initial or terminal stages of primary molt when few primaries are being molted simultaneously. (2) Average Intensity (as described by Rohwer and Rohwer 2013), the mean number of primaries on the right wing growing simultaneously for each feather between the 2nd and 8th primary, inclusive; we used this somewhat more conservative estimate, which includes some individuals in the early stages of molt replacing only 2-3 primaries, to allow us to compare primary molt intensities of the 13 species of warblers with the values calculated for other passerines by Rohwer and Rohwer (2013).

To examine the effects of daylength on the pace of primary molt, we used the online SolarTopo calculator (van der Staay 2020) to determine daylength on particular dates for the latitude (40.1637ºN) of Powdermill Avian Research Center. For some species-level comparisons we controlled for phylogenetic relationships via phylogenetic independent contrasts, using a recent phylogeny of the Parulidae (Lovette et al. 2010) and the package ‘caper’ (version 1.01; Orme 2018) of R version 3.6.2 (R Core Team 2019).

许多鸟类类群均会经历飞羽的快速高强度换羽，该过程会带来诸多不利影响，包括飞行能力下降、觅食效率降低、亲代抚育受损以及羽衣质量下滑。诸多迁徙性新大陆林莺（森莺科Parulidae）以高强度的飞羽换羽著称，近期研究亦表明，该类群中尾羽同步替换现象可能普遍存在。然而，目前学界尚未针对林莺类群，系统探究尾羽同步换羽以及更广泛意义上的高强度飞羽换羽的系统发育分布格局。

本研究于美国宾夕法尼亚州西南部的粉喙鸟类研究中心（Powdermill Avian Research Center），对涵盖7个林莺属的13个物种的飞羽换羽情况开展调查，以此解答上述问题。13个物种均同步替换全部12枚尾羽，且尾羽换羽的起始节点处于高强度初级飞羽换羽的早中期阶段。正如预期，单巢次的早期迁徙物种换羽时间早于繁殖活动延续至夏末的多巢次物种。但本研究发现，换羽较晚的物种，其初级飞羽替换速率更慢、强度更低，这一结果与预期相悖——此前学界普遍认为，换羽较晚的物种受迁徙相关的时间约束更强。这一意外结果至少可部分通过换羽速率与日照长度的正相关关系加以解释：夏末日照时长缩短，可能迫使鸟类每日摄食量下降，进而降低换羽强度、放缓换羽速率。

与其他雀形目鸟类相比，北美东部林莺的飞羽换羽强度极高；在换羽高峰期，个体需同时替换48枚飞羽中的50%~67%（含全部12枚尾羽，以及每侧翅膀的6~10枚飞羽），该状态将持续2~3周甚至更久。此类高强度换羽会对飞行、避敌、觅食以及亲代抚育造成诸多挑战，因此林莺的飞羽换羽期是其年度生活史中极具生理与行为压力的阶段。

## 研究方法
### 野外方法
换羽数据采集于1986年至2000年的夏、秋季环志工作中，地点位于美国宾夕法尼亚州西南部威斯特摩兰县劳雷尔高地的粉喙鸟类研究中心（地理坐标：40.1637°N，79.2674°W）。本研究共获取13个林莺属物种（涵盖7个属）的1289次个体捕获记录的换羽评分（详见表1）。由于本研究聚焦的13个物种均会在粉喙鸟类研究中心及威斯特摩兰县东部周边区域常规繁殖（Brauning 1992；Wilson et al. 2012），因此换羽期捕获的个体大概率为繁殖于巢区或其附近的本地繁殖个体（Pyle et al. 2018）。在全部13个目标物种中，研究均捕获了处于基础换羽各个阶段的个体，且未发现任何个体在迁徙前停止或暂停换羽的证据。本研究采用英国鸟类学信托基金会（British Trust for Ornithology）的0~5分换羽评分体系，对右侧翅膀的全部18枚飞羽（9枚初级飞羽、6枚次级飞羽、3枚三级飞羽）以及6枚右侧尾羽进行评分：0分为旧羽；1分为旧羽缺失或新生羽处于羽鞘阶段；2分为新生羽从羽鞘中长出，长度不足全长的1/3；3分为新生羽长度介于1/3~2/3之间；4分为新生羽长度超过2/3，但羽基部仍被鞘包裹；5分为新生羽完全长成，羽基部无鞘残留（Ginn & Melville 1983）。

### 数据分析与统计方法
针对每只个体，我们通过累加各枚羽毛的评分，计算得到初级飞羽总换羽评分（0~45分）与尾羽总换羽评分（0~30分）。为探究不同物种间个体同步替换全部尾羽的程度差异，我们选取13个物种中582只处于活跃尾羽换羽期的个体（尾羽总换羽评分介于0~30之间）的尾羽换羽数据开展分析。针对每只个体，我们计算了两项尾羽换羽同步性指数：（1）任意两枚尾羽的评分最大差值（MaxDiff；0分代表尾羽换羽完全同步，5分代表完全异步）；（2）6枚被评分尾羽的换羽评分标准差（SD；0分代表完全同步，1.87分代表完全异步）。由于MaxDiff与SD均不符合正态分布，我们采用Kruskal-Wallis检验与Dunn事后多重比较（对应方差分析与事后多重比较的非参数等价方法）开展物种间比较。所有统计计算均通过JMP Pro 12.2软件（美国北卡罗来纳州卡里市SAS研究所出品）完成。

为明确尾羽换羽起始时的初级飞羽换羽状态，我们选取10个物种的126只处于尾羽换羽早期阶段的个体（尾羽总换羽评分>0且≤6）的初级飞羽换羽评分开展分析。针对这10个物种，我们通过方差分析与Tukey-Kramer多重比较检验开展初级飞羽总换羽评分的物种间比较——残差正态分位数图检验结果显示，初级飞羽总换羽评分近似符合正态分布。

我们通过聚焦初级飞羽换羽，探究换羽时间与持续时长的物种间差异。森莺科林莺的初级飞羽换羽遵循标准雀形目模式：第一枚初级飞羽（最内侧）的脱落始终代表飞羽换羽的起始，其余初级飞羽则按照从近端到远端的固定顺序依次脱落（Foster 1967；Nolan 1978；Rimmer 1988）。鉴于初级飞羽换羽具有规律性与可预测性，且初级飞羽替换阶段几乎始终涵盖尾羽、次级飞羽与三级飞羽的换羽过程（如Foster 1967所述），因此本研究聚焦初级飞羽换羽的时间与持续时长是合理的。

我们通过JMP Pro 12.2的非线性平台，为每个物种的初级飞羽总换羽评分数据拟合三参数逻辑斯蒂模型，以此估算初级飞羽换羽中点对应的年度日序。相较于换羽起始时间，我们更倾向于估算换羽中点（或称中途日期；详见Jackson 2017；Erni 2018），因为换羽中点的估算结果对曲线拟合假设的敏感度更低：包括Pimm回归（Pimm 1976）、三参数逻辑斯蒂模型、三次样条插值与核平滑在内的多种线性与非线性曲线拟合方法，得到的初级飞羽换羽中点估算结果均较为一致。基于拟合得到的逻辑斯蒂模型，我们通过JMP的逆预测工具，估算每个物种达到初级飞羽换羽中点（初级飞羽总换羽评分=22.5）时的年度日序均值±标准误。

我们通过同一繁殖季内被多次捕获（≥2次）且处于初级飞羽换羽期（换羽评分>0且<45）的个体的初级飞羽总换羽评分，估算每个物种的初级飞羽换羽速率。相较于Pimm回归（Pimm 1976）与Underhill-Zucchini极大似然模型（Underhill & Zucchini 1988；Erni et al. 2013）等其他方法，本方法的优势在于可针对个体开展换羽速率分析，同时规避了回归与极大似然模型的假设前提限制（Newton 2009；Rohwer 2013）。针对每个物种，我们构建广义线性模型，以初级飞羽总换羽评分为因变量，纳入两项固定效应：作为连续变量的年度日序，以及作为分类变量的环志标识编号。模型中年度日序的系数（均值±标准误）代表初级飞羽换羽评分与日序之间关系的斜率（即每日换羽评分的变化量）。所有模型均通过JMP Pro 12.2的拟合模型平台构建。

结合初级飞羽换羽中点与换羽速率的估算结果，我们可进一步估算每个物种的初级飞羽换羽起始平均日期与换羽平均持续时长。本部分分析共纳入12个物种的137个个体-季节记录的309条换羽记录；蓝翅黄森莺（Blue-winged Warbler, BWWA）的重捕数据过少，因此未纳入换羽时间与速率分析。加拿大森莺（Canada Warbler, CAWA）的重捕数据同样有限，但由于其重捕个体的初级飞羽总换羽评分覆盖了几乎全部范围（补充材料图S1），因此仍被纳入分析——不过对该物种的换羽中点与速率估算需谨慎解读。

针对全部13个物种，我们计算了两项物种级别的初级飞羽换羽强度估算值：（1）**峰值强度**：定义为初级飞羽总换羽评分介于16~35分的个体，其右侧翅膀同时替换的初级飞羽平均数——该估算值排除了换羽初始或末期阶段的个体，此类个体同时替换的初级飞羽数量极少；（2）**平均强度**：参照Rohwer与Rohwer（2013）的定义，即右侧翅膀第2至第8枚初级飞羽（含两端）中，每枚羽毛同时生长的个体平均数。我们采用这一相对保守的估算标准，其纳入了部分仅替换2~3枚初级飞羽的早期换羽个体，以便将13种林莺的初级飞羽换羽强度，与Rohwer与Rohwer（2013）针对其他雀形目鸟类计算的换羽强度值进行比较。

为探究日照长度对初级飞羽换羽速率的影响，我们通过在线SolarTopo计算器（van der Staay 2020），计算粉喙鸟类研究中心所在纬度（40.1637°N）特定日期的日照时长。针对部分物种间比较，我们采用系统发育独立对比法控制物种间的系统发育关系：基于最新的森莺科系统发育树（Lovette et al. 2010），并使用R版本3.6.2（R Core Team 2019）中的‘caper’包（版本1.01；Orme 2018）完成分析。