1 ms for Journal of Avian Biology

2
3 Local survival is associated with beak morphology in crossbills feeding on a range-
4 restricted pine: due to differential survival or differential dispersal? Commented [SS1]: quizás se podría eliminar “due to”.
5
6 David Gómez-Blanco1, Simone Santoro1Santoro2, Antoni Borrás2Borrás3, Josep
7 Cabrera2Cabrera3, Juan Carlos Senar2 Senar3 & Pim Edelaar1 Edelaar2
8 1 Department of Biology, Molecular Ecology and Evolution Lab, Lund University, Ecology

9 Building, 223 62 Lund, Sweden

10 21 Department of Molecular Biology and Biochemical Engineering University Pablo de

11 Olavide, Seville, Spain

12 32 Behavioural and Evolutionary Ecology Associate Research Unit (CSIC), Natural History

13 Museum of Barcelona, Barcelona, Spain

14
15 Abstract Commented [SS2]: it has to be less than 300 words
16
17 Dozens of morphologically differentiated populations (including subspecies and species) of crossbills
18 (genus Loxia) have been described across the world. It has been suggested that this divergence is due
19 to variation in cones among the conifer species that each population specialises upon, requiring a
20 specific beak size to most efficiently separate the cone scales. If so, local survival should depend on
21 beak measures. To test this hypothesis, we undertook multievent capture-recapture modelling for
22 6,844 individuals monitored during 27 years in a relatively resident and particularly large-billed
23 Pyrenean Common crossbill (L. curvirostra) population feeding on the thick-scaled cones of the
24 range-restricted Mountain pine (Pinus uncinata). Local survival was indeed related to beak width
25 (and to a lesser extent to height), resulting in stabilizing selection around an optimum of 11.43 mm.
26 The observed population mean was close to the predicted optimum, which is further support that
27 crossbill beak morphology is adapted to the conifer they specialise upon. The encountered variation
28 in local survival might be explained by natural selection (the standard explanation), but alternatively
29 it could be explained by selective dispersal of individuals, with locally maladapted individuals
30 permanently leaving the study area. Because our data come from a single study area (Aas is often the
31 case in capture-recapture analyses), it is not possible towe could not formally decompose local
32 survival into selective mortality versus selective permanent emigration. Nonetheless, we obtained
33 several indications that selective permanent emigration also affected local survival. First, to create Commented [SS3]: played a main role?
34 the observed pattern of local survival, natural selection would have to be unusually strong. Second,
35 the observed mean beak width was a bit lower than estimated optimum beak width, suggesting
36 immigration by crossbills with smaller beaks, which may then have left the study area permanently.
37 Third, such movements predict a statistical transient effect in the data, which we indeed observed.
38 FourthThird, we found huge variation in local survival among years (15-95%),. This was not unrelated
39 to food availability but was related to crossbill density, suggesting that local survival estimates were
40 reduced in some years due to temporary arrivals and departures. In general, permanent emigration
41 is hard to study, but tThe possibility that permanent emigrationthis is phenotypically selective makes
42 itis ecologically and evolutionary very relevant, and therefore it deserves more attention. Commented [SS4]: Max. 300 palabras en J of Avian
43 Biology

44 Introduction
1
45
46 Organisms are constantly facing ecological challenges due to variation in their biotic and abiotic
47 environments, and the interaction between phenotype and environment modifies their survival
48 and/or reproductive success (Darwin 1859). It is therefore adaptive to have a better match between
49 phenotype and environment, in order to improve ecological performance. Several processes may
50 achieve this improved match, operating both at the individual and population level (Edelaar et al.
51 2008). A first process is the classical local adaptation of populations (e.g. adaptive tracking and
52 adaptive population genetic divergence), based on differences in fitness. If individual variation in
53 phenotypic match with the environment is caused by hereditary phenotypic traits, natural selection
54 causes adaptation to the environment across generations at the population level (Manly 1985, Endler
55 1986). However, in variable environments this tends to cause a lag between the required optimal
56 phenotype and the actually evolved phenotype. Therefore, a second well-known, widespread and
57 important process is matching the phenotype to the environment by adaptive phenotypic plasticity
58 (Ghalambor et al. 2007, Przybylo et al. 2000, Black 1993). A third potential way to deal with
59 environmental variation is selection of the environment, by which an individual selects an
60 environment that matches well with its phenotype, e.g. adaptive habitat choice. These processes can
61 operate simultaneously to improve fitness, but their distinction is important for our understanding of
62 the ecological processes underlying the patterns we observe in ecosystems. By default, the route to
63 increase the fit of the phenotype to the environment is population genetic adaptation and
64 differentiation through natural selection on non-plastic phenotypes. However, plasticity and
65 selection of the environment are favoured as environmental heterogeneity increases and their costs
66 are low relative to individual benefits (Scheiner 2016, Edelaar et al. 2017).
67
68 The Common Crossbill (Loxia curvirostra) is a sparrow-sized songbird specialized in feeding on seeds
69 enclosed in conifer cones. They are characterized by their thick beaks with curved and crossed bill
70 tips that are used to extract seeds from between the scales of closed cones (Benkman and Lindholm
71 1991). As is typically the case in bird taxonomy, crossbill taxa have historically been defined by
72 distinct geographic (allopatric) distributions, biometrics and plumage colouring (Cramp and Perrins
73 1994). In more recent times, morphologically, vocally and genetically distinct sympatric populations
74 (if not incipient species) of crossbills are recognised (Benkman 2016a, Parchman et al.
75 2016)(Parchman et al. 2016)(Parchman et al. 2016)(Parchman et al. 2016)(Parchman et al.
76 2016)(Parchman et al. 2016)(Parchman et al. 2016)(Parchman et al. 2016)(Parchman et al.
77 2016)(Parchman et al. 2016)(Parchman et al. 2016)(Parchman et al. 2016). This seems to be linked to
78 the availability and adaptation to different conifers. Benkman (1993) showed that conifers differ in
79 cone size and scale traits, and that there is not a single beak phenotype that allows feeding optimally
80 on all cone variation (Benkman 1993). Given that the beak is a non-plastic trait (Summers et al. 2007) Commented [SS5]: I would not say that is local survival.
81 that develops before immatures start feeding on cones, adaptive plasticity is not an option to For local survial one thinks of a parameter that is net of
recapture rates and this is not the case here (return rate =
82 increase a crossbill’s food intake on a given cone type. Benkman (2003) subsequently showed for a local survival rate x recapture rate).
83 resident crossbill population that local survivalreturn rate was correlated to bill morphology, with
This is not a useless detail. A good point of this ms, perhaps
84 highest return ratesvalues for those birds that had the highest intake rate on the local cones.
not enough emphasized, is that we are investigating
85 Therefore, by assuming return rates as proxy of local survival, divergent selection should act on bill whether selection exists on beak size/morphology by
86 traits among crossbill taxa feeding on different conifer. The use of alternative resources then would directly estimating local survival.

87 be the ultimate cause of adaptive radiation in this species complex. However, this rests on the Commented [SS6]: you meant species complex or
complex species?
88 assumption that the correlation between bill morphology and survival is general.
89 Commented [SS7]: Me confunde aquí esta frase, yo la
quitaría.

2
 90 In Spain crossbills are irregularly distributed in areas with coniferous forests (Borrás and Senar 2003).
91 Currently, two subspecies are recognized: Loxia curvirostra curvirostra (the same subspecies as
92 occurring across most of Eurasia), and L. c. balearica, from Mallorca (Balearic Islands) and according
93 to some authors also south-eastern continental Spain. The crossbills of SE Spain have also been Commented [SS8]: would this need a reference?
94 considered to be a separate subspecies (hispana), intermediate to curvirostra of N Spain and
95 balearica of Mallorca (Cramp and Perrins 1994, Alonso et al. 2006). Loxia c. curvirostra from northern
96 Spain feeds almost exclusively on the seeds of mountain pine (Pinus uncinata) and scots pine (P.
97 sylvestris) (Clouet 2000). Loxia c. balearica feeds on Aleppo pine (P. halepensis), more abundant in SE
98 Spain and the only pine on the Balearic Islands (Alonso et al. 2006). The divergence in the bill
99 morphology of crossbill populations in Spain and a reduction in gene flow appears to be due to the
100 differential use of resources, i.e. isolation by ecology (Edelaar et al. 2012). Aleppo pine produces
101 larger and thicker cones than mountain and scots pine (Castroviejo et al. 1986), but has relatively
102 long and flexible scales. This would cause the beak of N Spanish crossbills to evolve to be relatively
103 shorter but wider than those of the birds in S Spain and Mallorca (Alonso et al. 2006, Borrás et al.
104 2008, Edelaar et al. 2012).
105
106 However, such an effect of beak morphology on adaptation has not been shown, only assumed to be
107 general based on the single result of Benkman (2003). The first objective of this study is therefore to
108 investigate if beak morphology affects local survival, and if so, to determine which kind of beak is
109 locally adaptive. For this we used an extension of the Cormack-Jolly-Seber capture-recapture model
110 (Lebreton et al. 1992) to analyse data collected over 27 years from a Pyrenean crossbill population. Commented [SS9]: In this paragraph we might emphasize
111 the novelty of using capture-recapture analyses which
provide estimate of local survival net of the observation
112 Irrespective of any relationship between beak morphology, ecological performance, and local process. I haven’t done it because I am not sure you see it
113 survival, the specialisation of and divergence among sympatric crossbill taxa is surprising, because convenient.
114 population contact and geographical overlap are normally assumed to prevent or erase divergence, Commented [SS10]: Este segundo párrafo en el que se
115 due to homogenizing gene flow (Hendry et al. 2001). Such potential for homogenizing gene flow is dice cual es el segundo objetivo me parece demasiado largo
pero me da cosa recortar (no lo he escrito yo). Cuando
116 particularly large in crossbills, due to large spatio-temporal variation in food availability. Many of the llegas al final y lees cual es el segundo objetivo te olvidas
117 conifers that crossbills feed on have a great variability in the productivity of cones between years, que habíamos llegado al punto en el que uno se tenía que
esperar el segundo objetivo.
118 with booms and busts occurring across large areas (Clouet 2000, Borrás and Senar 2003). In years of
119 low local cone production, crossbills therefore move to other areas in search of better cone crops of
120 the same type of pine (or a suitable different one). Flocks of crossbills moving to areas with other
121 pine species after local crop failures have often been observed (Clouet 2000, Borrás and Senar 2003),
122 sometimes in the form of massive invasions involving millions of birds. Indeed, the crossbills of
123 northern Eurasia may well be the most dispersive passerine on Earth, with average natal and
124 breeding dispersal distances > 2,100 km (Newton 2006, Marquiss et al. 2008). These movements of
125 individuals should increase the mixing of populations, effective gene flow, and the homogenization of
126 genetic groups. However, alternatively, these movements could bring or keep together
127 phenotypically similar individuals if, after evaluating different areas, they settle in the same area that
128 best fits their ecological needs given their phenotype, i.e. via habitat choice (a form of selection of
129 the environment). Especially matching habitat choice and its associated selective movement could
130 actually promote, instead of preventing, differentiation of populations (Edelaar et al. 2008), because
131 matching habitat choice is a form of habitat choice that depends on a direct comparison of local
132 performance (i.e. the phenotype-environment match) across different options.
133

3
134 Hence, it remains unclear if isolation by ecology in crossbills is due to natural selection (e.g. selective
135 survival) in local populations, or due to selective dispersal among populations (selection of the
136 environment). That is because even though these two processes are opposites (in the first the
137 phenotype is matched to the environment, in the second the environment is matched to the
138 phenotype), they can result in the same phenotype-related distribution of individuals.patterns. This
139 uncertainty on driving processes is also the case for any other study on isolation by ecology in which Commented [SS11]: an alternative:
140 selective dispersal and settlement cannot be excluded, or in fact for any study on local adaptation, This duality between mortality and dispersal on driving
processes…
141 but this latter possibility is often ignored. Therefore, as a second objective, we evaluate the possible
142 roles of natural selection (death) versus selection of the environment (permanent dispersal) in
143 explaining any pattern of local survival in our Pyrenean crossbill population.
144
145 Methods and materials
146
147 Study population and data collection
148 Capture of crossbills took place in the Port del Comte mountain range in the Catalan Pre-Pyrenees,
149 approximately 100 km NW of Barcelona. The only pine occurring here is Mountain pine (Pinus
150 uncinata). Mist nets were placed where birds came down to drink or feed on minerals. Captured
151 individuals were marked with individually numbered aluminium rings, and released at the capture
152 site. Sex and age were determined according to Svensson (1992). We considered three age groups.
153 Juveniles are those birds that have not yet moulted into adult plumage so their sex is difficult to be
154 visually determined (although see Edelaar and Terpstra 2004). Yearlings are birds in adult plumage
155 but with some remainder of juvenile plumage (incl. some greater coverts and flight feathers, which
156 are typically retained until the next full moult). Adults are individuals that have gone through a
157 complete moult. Biometric traits were measured by a single observer, incl. beak width (base of the
158 lower mandible, ± 0.1 mm) and beak height (height of closed mandibles at the nostrils, ± 0.1 mm).
Commented [SS12]: Al final cualquiera que lo lea se va a
159
preocupar por el exiguo número de recapturas, así que
160 The capture-recapture dataset mejor enseñar que somos conscientes de que es un número
161 The birds were ringed and recaptured between 1988 and 2014. Given that recaptures were relatively bajo.

162 scarce, we combined all the observations from May to October of the same year into a single capture Commented [e13]: It would be good to report
repeatabilities for these two beak measures (a mixed model
163 occasion. This yields higher recapture rates, which give more precise and less biased estimates of fitting individual as a random factor), especially because we
164 state transition parameters like survival rate (Hargrove and Borland 1994, O’Brien et al. 2005). suggest in the discussion that bill height is less repeatable.
165 Nonetheless, The the resulting dataset consisted of a total of 6,844 marked crossbills and only 272
166 (≈4%) recaptures. Furthermore, due to the lack of biometry for some individuals, two reduced David:
167 datasets were constructed for birds with measures of beak width or beak height (955 and 1,933 Results of the adjusted repeatability:
Rn<-R/(R+(1-R)/n0)
168 marked and measured crossbills, respectively). When a crossbill was measured multiple times, its
169 first measurement was used in the analyses. BeakW Rn= 0.299
BeakH Rn= 0.873
170
171 Multievent capture-recapture modelling Según he leído, un valor de 1 indica que la medida es
172 Multievent models were constructed and adjusted to the data using the software E-SURGE 2.1.2 perfectamente repetible y próximo a 0, las medidas son tan
distintas como si fueran de individuos distintos al azar.
173 (Choquet and Nogue 2011). The multievent framework allows one to use all the data by
174 distinguishing between what the events, i.e. the individual states as they has have been observed in Así que creo que este análisis más que apoyar la idea de
baja repetibilidad de BeakH, nos dice lo contrario.
175 the field (the events coded in the encounter histories, often incomplete or uncertain), and the
176 underlying biological states of the individuals, which must be inferred (Pradel 2005). We defined Simone: ojo, lo hemos hablado ayer con David, el tamaño
de muestra para beakW es bajísimo (12, es decir 6
177 three underlying biological states: Dead=✝, Female alive=F, Male alive=M, and four possible events: individuos que han sido capturados Y EL BEAKW MEDIDO
178 0=Non-detected, 1=Female-detected, 2=Male-detected, 3=Detected-sex-unknown. Age groups are dos veces. Yo no creo que esto es fiable...

4
179 have been coded within the “headed format” (see Choquet and Nogue 2011) data file as: (1) juvenile,
180 (2) yearling and (3) adult (details on the probabilistic framework are given in Additional file 1). The
181 estimated probabilitiesparameters (Initial State, State Transition and Event) of multievent models
182 parameters (Initial State, State Transition and Event) are calculated simultaneously from the full
183 encounter histories by maximum likelihood (Choquet et al. 2009b).
184
185 Goodness of fit
186 Since no goodness of fit (GOF) tests are available for multievent models, we tested the fit of the
187 Cormack-Jolly-Seber model (CJS) using U-CARE 2.3.4 (Choquet et al. 2009a). The CJS is a model that
188 allows for unspecified temporal variation in the probabilities of apparent survival and of recapture
189 (Lebreton et al. 1992). The CJS model only distinguishes two biological states (alive, dead) and two
190 possible events (capture, no-capture). The GOF analysis was performed on our three groups that
191 were defined according to the age at first capture. U-CARE allows testing whether the model fits the
192 data, and investigating the potential causes of lack of fit (if any). A first reason for lack of fit is known
193 as the “transient effect”, when the apparent survival probability in the first interval after marking is
194 substantially lower than in the subsequent intervals (test component 3.SR). A transient effect can be
195 caused by the presence of individuals which use the study area sporadically and/or by a differential
196 survival of a subgroup of individuals with lower survival than others (Duriez et al. 2009). A second
197 cause of lack of fit is known as trap-dependence, when individuals captured (or recaptured) in the
198 previous occasion have a different probability of being recaptured than the other individuals (test
199 component 2.CT). When the global GOF test (which merges all the components together) is
200 significant, the extra-binomial sources of variation have to be accounted for in the structure of the
201 capture-recapture models and/or an over-dispersion coefficient (c-hat) must be used in the analyses
202 to make the model selection more conservative.
203
204 For our data, we did not find evidence of trap-dependence (Z= -0.63; P = 0.53). However, we did find
205 a transient effect (χ2 = 86.44; df = 144; Z= 3.27, P <0.001). When the analyses were performed
206 separately for the three age groups, we still found statistically significant transience effects for
207 juveniles (χ2 = 28.85; df = 22; P <0.01) and yearlings (χ2 = 8.26; df = 14; P = 0.04), but less so for
208 adults (χ2 = 8.53; df = 18; P = 0.13). In order to control for the detected transient effects, we included
209 an age effect (three age classes) on survival in the global model (i.e. the most complex model, see
210 below).
211
212 Model selection
213 We first modelled the entire dataset, independent of whether morphological measures were taken.
214 In our global model (the initial full model, before simplification) we included temporal variation in all
215 the parameters (Initial State, Survival, Capture probability, and Sex-Assignment), including in
216 interaction with sex and age for some parameters when the data allowed it (i.e. models not over-
217 parameterised). We followed a step-down model selection approach (Lebreton et al. 1992) based on
218 the Akaike Information Criterion corrected for small sample size (AICc) (Burnham and Anderson
219 2004) that consisted of sequentially determining the best structure of the parameters in this order:
220 Sex-Assignment, Capture probability, Initial State, and Survival (see Santoro et al. 2016 for an
221 analogous approach). To this aim, we ran for the first parameter (Sex-Assignment) all the nested
222 models for the entire dataset, retained the structure for this parameter given by the model with the
223 lowest AICc, and repeated the same procedure with the next parameter (Capture probability),

5
224 etcetera. As more data is involved, Using using the entire data set gives us the best demographic
225 estimates and the best assessment of which factors affect each parameter, as more data is involved.
226
227 Next, we modelled the reduced datasets with the individual covariates “beak width” and “beak
228 height”, in order to test if beak morphology influenced apparent survival. For this, the structure of
229 the simplified final model based on all data (above) was used as the global model, and the step-down
230 model selection procedure was repeated for those parameter types that were not already Commented [SS14]: me cuesta entender porque antes
231 determined to be unaffected by time, sex or age (i.e. constant). Then, after this dataset-specific estaba escrito (supongo que yo lo escribí) de esa manera
pero así me parece correcto y fácil de entender.
232 model selection, we used the simplified final model with the lowest AICc as the null model to test for
233 additive linear and/or quadratic effects of beak width and beak height on apparent survival. We
234 considered three possible models to test the effect of the individual covariate (beak width or height):
235 only a linear effect (linear directional selection), only a quadratic effect (stabilising or disruptive Commented [e15]: Could you run the model of a linear,
236 selection around the population mean), and both effects (stabilising or disruptive selection around a quadratic, and cubic effect (beak width * beak width * beak
width), to test whether the stabilising selection is
237 value different from the population mean). symmetrical around the optimum? Basically, to fit a
238 polynomial function of the third order.

Model Nº Par. Deviance QAICc ΔAICc wAICc David: Tabla con los results
42 0.34 Para BeakH, el modelo cubic no se selecciona como mejor
S(-)W 2556.44 2644.09 1.35
modelo, pero para BeakW si, al menos mejor que el modelo
S_cov123W 45 2548.54 2642.74 0.00 0.66 nulo sin ninguna cov.

Simone: El AIC del cubico realmente hay que compararlo al

Model Nº Par. Deviance QAICc ΔAICc wAICc AIC del cuadrático (que está en la table 2). Va mejor el
cuadrático para beakW y para beakH tampoco resulta
S(-)H 50 5107.7015 5210.25 0.00 0.92 relevante.
S_cov123H 53 5106.29 5212.29 4.91 0.08 ¿Lo quitamos o nos sirve para algo?
239
240
241 Finally, we calculated the standardized quadratic selection gradient (γ) to estimate the strength and Commented [SS16]: ¿No sería necesario decir cómo se
242 mode of natural selection. Negative values of γ are necessary for stabilizing selection, whereas calculó?

243 positive values of γ are necessary for disruptive selection.

244
245 Results
246
247 Model selection with the complete dataset
248 The probability of sex assignment varied over time and was, on average, higher for males (0.83,
249 95%CI: 0.74-0.89) than for females (0.52, 95%CI: 0.48-0.56). The probability of recapture was
250 independent of time and sex, and extremely low (0.05, 95% CI: 0.04-0.07). There was a slight
251 majority of males among first-captured individuals (0.53, 95% CI: 0.49-0.56). We also did not find
252 temporal variation in the probability of initial state, i.e. the probability a first-time captured
253 individual was a male (0.47, 95%CI: 0.44-0.51). For apparent survival, the best-supported model
254 included the additive effects of age, sex and time (see Table 1 and Figure 1). On average (model 11 in
255 Table 1), juveniles had the lowest survival (females, 0.30, 95% CI: 0.23-0.39; males, 0.33, 95% CI:
256 0.25-0.42), followed by adults (females, 0.45, 95% CI: 0.39-0.50; males, 0.48, 95% CI: 0.43-0.53) and
257 yearlings (females, 0.55, 95% CI: 0.41-0.67; males, 0.58, 95% CI: 0.44-0.70).
258
259 The effects of beak size on apparent survival

6
260 The model selection procedure with the two reduced datasets (individuals with beak measures
261 available) gave similar results as for the whole dataset, except for a lack of a sex effect on apparent
262 survival and on sex assignment (Supplementary material 2). Subsequent testing of the additive effect Commented [SS17]: This is not surprising as less data are
263 of beak size on apparent survival led to different results for beak width versus beak height (Table 2). available which mean less statistical power. Should we
mention it at some time?
264 For beak width, the best model included the linear and quadratic effect of the covariate (see Figure
265 2) (coefficient on logit scale for the linear term = 0.22; SE = 0.14, coefficient for the quadratic term on
266 logit scale = -0.35; SE = 0.18) closely followed by the model only including the quadratic effect (ΔAIC =
267 0.19; coefficient = -0.33; SE = 0.17). The optimal beak width with respect to local survival for
268 crossbills in the study area was estimated at 11.43 mm (see Figure 2), with the probability of
269 apparent survival decreasing away from this optimum.

270
271 For beak height, the best model did not include the covariate, but it had only a moderate model AIC Commented [JC18]:
272 weight (0.45). The model including only the quadratic effect was only 1.04 AIC points removed We had really higher sample sizes for this covariate than for
bill width, so NS it is quite probable to be real...
273 (coefficient on logit scale = -0.11; SE = 0.11). Overall, models including a quadratic effect of beak
274 height obtained a cumulative model AIC weight of 0.37, and for models including the linear effect Pim: no?? Sample size is twice as large for beak width
(according to the text, line 173).
275 this was 0.26.
276 David: No, in the text the sample size was changed at some
277 Discussion point (we have double data for beak height than for beak
width)
278
279 Does beak morphology affect survival? Simone: Resumiendo, hay más datos para beak height de tal
manera que el hecho de que no salga el efecto de la
280 Our first objective was to investigate if crossbill beak morphology affects local survival, and if so, to covariable parece ser algo real como dice Juan Carles. Tal
281 determine which kind of beak is locally adaptive. Our findings suggest that beak size (especially beak vez no conviene dar tanto detalles sobre los pesos de
Akaike y más pensando que NO hemos hecho esto para la
282 width) of crossbills indeed affects their local survival in the study population, as models including
otra covariable así que parece una manera de contar los
283 effects of beak measures fitted the data on survival histories better than models without. This resultados poco consistente. ¿Qué os parece?
284 supports an earlier finding by Benkman (Benkman et al. 2003) who showed, for a resident crossbill
285 population in North America feeding on a different pine species, that survival the return rate of Commented [SS19]: I insist with that because this
286 marked individuals depended on beak morphology (especially beak height, for which our evidence is parameter is the byproduct of three processes: mortality x
dispersal x recapture.
287 weaker, but sample size is also much smaller). We estimated that the population optimum beak We don’t know which of these processes was affected by
288 width in our population is 11.43 mm, with local survival decreasing away from the optimum. The the beak morphology in Benkman’s study.
A phenotypic trait might be related to personality which in
289 crossbills utilising mountain pine, as in our study area, actually have the largest beaks among Spanish
turn might affect trapping behavior: it is not that obvious
290 crossbills (Edelaar et al. 2012). The fact that the observed average beak size is close (slightly smaller, one should rule it out…
291 see below) to the estimated locally optimal beak size (actually slightly smaller, see below), supports Commented [DGB20]: This is not true; the sample size
292 the idea that this crossbill population benefits from having such a large beak, and is in that respect was twice the one of beak width.
293 locally adapted. While we do not have data that links beak morphology to feeding performance (see Simone: Not clear to me if we are saying that our sample
294 Benkman 1993), it is likely that the need to have such large bills is related to the very thick cone size is much smaller than that of Benkman or that the
295 scales of mountain pine, requiring more force to separate them. It was not beforehand obvious that sample size for beak height is smaller than that of beak
width – the latter would not be true (as David says).
296 the crossbill population that we studied would be locally adapted, since mountain pine has a very
Commented [SS21]: I think it is important to stress out
297 restricted distribution, only occurring in the subalpine zone of the Pyrenees and western Alps. Small we are referring to LOCAL survival as indeed our hypothesis
298 ranges are typically associated with smaller population sizes, making populations more vulnerable to is that variation in this parameter is here due to variation in
299 environmental and demographic stochasticity as well as maladaptive gene flow. However, mountain dispersal rate (not in survival).

300 pine shows relatively small temporal and small-scale spatial variation in cone production (Clouet
301 2000, Borrás and Senar 2003), so this likely facilitated local adaptation and morphological
302 specialisation since resource stability within the range of a population enhances the evolutionary life
303 span of specialist populations (Björklund et al. 2013).
304

7
305 Causes of stabilizing selection
306 Our second (and more challenging) objective was to evaluate the possible roles of natural selection
307 (death) versus selection of the environment (permanent emigration) in explaining the variation in the Commented [SS22]: I have seen death and permanent
308 observed local survival as related to beak morphology. A pattern of local survival as found by us emigration used in this way before in the text. Would it not
be better saying mortality and permanent emigration?
309 (Figure 3) is generally explained interpreted by as stabilising natural selection acting through Death does not make me think of a process, it makes me
310 selective deaths of locally maladapted individuals. The alternative, that locally maladapted think of an event.
311 individuals permanently leave the study area and therefore are accounted as “locally dead”, is far
312 less often considered if not generally ignored. Given that our estimates proceed from a uni-site
313 modelling approach, it is not possible to directly discriminate between true mortality versus
314 permanent emigration (as is true for most capture-recapture studies): both processes can yield the
315 exact same pattern, since local survivors need to survive and stay present in the study area. Indeed,
316 previous studies on local survival in crossbills (Santisteban et al. 2012, Senar et al. 1993, Benkman et
317 al. 2014, Benkman 2003, Alonso and Arizaga 2012, Benkman 2017) were all done in single study Commented [SS23]:
318 areas, and virtually almost none of them (with the exception of Benkman 2017) suggested 1) Benkman made a logit regression, i.e. return rate, not
local survival. I feel we are not emphasizing enough the fact
319 permanent emigration to explain individual or annual variation in local survival rates (with the we are eliminating at least one potential confounding
320 exception of Benkman 2017). process (the observational process) and that this is novel
with respect to previous analogous studies.
321
2) Benkman 2017 * is not in the bibliography list (David
322 Despite the difficulty to distinguish between natural selection (here: selective death) and selection of please check it out in Mendeley the whole list before
323 the environment (here: selective permanent emigration), we discuss a few important indications that submitting the paper).
* Benkman, C. W. (2017). Matching habitat choice in
324 suggest to us that natural selection is not the only more likely explanation. First, when comparing our nomadic crossbills appears most pronounced when food is
325 estimate for quadratic selection (β = -0.35) with those compiled by Kingsolver and Diamond (2011) most limiting. Evolution, 71(3), 778-785.
326 (Figure 3), our estimate would be an unusually strong measure of stabilising natural selection. Said
327 otherwise, an unusual number of selective deaths would be occurring because of a phenotypic trait.
328 Therefore, it appears probable that selective dispersal of locally maladapted individuals out of the
329 study area has contributed in order to produce this high value. The same kind of observation and
330 logic led Benkman (2017) to conclude the same, in that case to explain a pattern of apparent
331 directional selection for larger bills. Our parallel observation reinforces his conclusions. Second, the Commented [SS24]: Aquí se podría mencionar que,
332 estimated optimal beak width is somewhat higher than the observed average beak width in the además, nosotros hemos excluido el posible confounding
effect en el proceso de observación.
333 population, and indeed the model which received the highest statistical support included a linear
334 effect of beak width (Table 2). That the studied population has a mean beak width which is slightly Commented [SS25]: He eliminado esa frase porque creo
335 too small for what provides the highest local survival is in line with immigration of individuals with que confundiría a unos cuantos no muy puestos en
estadística.
336 smaller beaks originating from other populations. Given that Spanish crossbills using other pines
337 (Scots pine, Black pine, Aleppo pine) which are growing relatively nearby have smaller beaks on
338 average (Senar et al. 1993, Edelaar et al. 2012), it is likely that at least a proportion of the birds we
339 captured were birds from other areas and thereby reduced mean beak size. In view of the high
340 mobility of crossbills, it is also likely that these birds subsequently would return to their original
341 habitat or move on in search of better places. Third, and in support of the previous point, we find a
342 significant transient effect in juveniles and yearlings (and a non-significant indication for it in adults).
343 Transient effects can indicate heterogeneous mortality within a group, but typically transient effects
344 are thought to be caused by heterogeneity in permanent emigration. If there is selective dispersal by
Commented [SS26]: Creo que lo hablamos en el
345 beak size within a group, then a transient effect should appear, and this is indeed what we found. congreso, este argumento es circular. Aunque se hable de
346 FourthThird, despite extremely low recapture rates as those found in this study usually increase the transient effect lo puede estar causando tanto diferencias
en mortalidad como en dispersión. Es el mismo discurso,
347 confidence interval of survival estimates, our model selectionwe still find supported a tremendous pero visto del otro lado, que aplicamos cuando decimos que
348 annual variation in survival rates, from as high as 0.95 (1998) to as low as 0.15 (2009) (Figure 1), aunque se hable de apparent survival, lo puede estar
349 which is untypical for a passerine of this size. In trying to explain this variation, we modelled if causando diferencias en mortalidad pero también en
dispersión.

8
350 variation in annual survival rate depended on an index of cone production (scored as high, average or
351 low; local data available from 1991 onwards), on an index of local crossbill density (number of birds Commented [SS27]: He eliminado esa parte porque, tal
352 caught by us), and their interaction. Crossbills are non-territorial scramble competitors, and food como discutimos en Mieres, las estimas de apparent
survival no son independientes del numero de individuos
353 availability generally is the main regulator of crossbill density and reproduction (refs.). In contrast to capturados.
354 this expectation, our obtained estimates of local survival did not appear to only depend on the
Commented [SS28]: ¿Alguien encuentra una referencia
355 number of birds caughtshow any relationship with the index of cone production. This suggests that para esto?
356 competition over food does not play a role, since neither cone production by itself nor its interaction Commented [e29]: This should be tested, but it looks that
357 with density explained local survival. Instead, this result suggests to us that significant numbers of way.
358 birds may have temporarily immigrated into our study area, were caught and marked, and
In fact, we could include cone crops and number of birds
359 subsequently left the study area permanently (independent of cone production). As an effect, local directly in our main models, to improve the estimates. But I
360 survival estimates are decreased. This explanation is in line with the detection of transient effects think their effects are already fully captured by the year
effects?
361 (point 3), and with the suggested immigration of birds with smaller beaks (point 2). We also explored
362 whether such influxes into our study population could be due to immigration from northern Eurasia, By the way, the density/immigration effect may not explain
the very high survival in some years, which also seems
363 since it is well-known that when the availability of resources is limited, instead of facing death by unrelated to cone crops. Any ideas?
364 starvation, crossbills undertake huge irruptions towards the southwest (Cramp and Perrins 1994,
365 Davis 1964, Summers et al. 1996, Herremans 1988, Edelaar and Terpstra 2004). This can cause Spain Juan Carlos, in which part of the year have cone crops been
estimated? And were they counts of green, closed cones or
366 to receive crossbills coming from further north (Senar and Borrás 1985). However, in our study area brown cones?I assumed here they were green cones
367 years with high numbers of birds caught do not coincide at all (not shown) with years in which counted in summer t, which mature and become available
in autumn and winter, and provide food until the next
368 invasions appear to be occurring in central Europe (based on Marquiss et al. 2012, and the websites spring t+1., and therefore affect the survival of the birds
369 Trektellen.org, Waarneming.nl, Euro Bird Portal, and Vogelwarte.ch). This suggests that our influxes considered in year t. Is that correct?
370 are coming from closerbynearby, possibly Spain itself (which supports point 2). Commented [SS30]: Yo eliminaría esta parte, véase
371 comentario mio de abajo también.
372 In view of these indications, and in view of the biology of crossbills (high mobility, ease of estimating
373 individual local adaptedness via food intake rate, ease of recognising different pine species from a
374 distance), we feel that it is highly likely that morphologically maladapted crossbills (either immigrants
375 or local offspring) were caught and marked by us, but subsequently left the study area permanently Commented [SS31]: Yo lo eliminaba, añade más
376 because of low food intake rates. How large this effect of selection of the environment on local elementos sobre los cuales pensar en la cabeza de quien
está leyendo.
377 survival is relative to natural selection cannot be established, but it could conceivably be the main
378 contributor to the observed pattern of local survival. We also feel that our study system is not unique
379 in this respect: selective immigration or emigration associated to phenotype and ecological
380 performance could occur in many organisms, as long as they have at least some control over their
381 dispersal events. Commented [SS32]: También lo eliminaba, el paper es un
382 poco largo y esta frase no me parece indispensable
sobretodo porque el mismo concepto se dice (más sintetico)
383 Concluding remarks unas líneas más abajo “…we think habitat choice has been a
384 Throughout their global distribution, common crossbills show a wide variation in their beak major contributor to the diversification in this species
complex, and perhaps many others”.
385 morphology, resulting in the description of dozens of subspecies. For allopatric, resident populations
386 that are users of a single conifer, natural selection is the most likely explanation for this phenotypic
387 divergence. However, because of their resource-driven mobility, the remaining populations could be
388 at least part of the time in contact with each other, which would be expected to lead to
389 homogenization and prevent divergence. Under those conditions, the tendency of birds to settle in
390 or depart from a particular area as a function of their local ecological performance (food intake)
391 could lead to a maintenance of spatially and ecologically segregated populations with larger or
392 smaller beaks (Edelaar et al. 2008, Holt and Barfield 2008, Armsworth and Roughgarden 2008,
393 Edelaar and Bolnick 2012, Bolnick and Otto 2013, Berner and Thibert-Plante 2015, Benkman 2016b).
394 Therefore, we think habitat choice has been a major contributor to the diversification in this species

9
395 complex, and perhaps many others. For further proof, monitoring across ecologically heterogeneous
396 areas is necessary to tease apart the process of selective survival from that of selective dispersal.
397
398
600 1.00 Commented [SS33]: yo eliminaría esta gráfica (era solo

SURVIVAL
CAPTURES

0.90 provisional) pero lo que es mas importante es que no

500 0.80 correría el modelo CR que viene comentado en el siguiente
comentario (e21). Tal como hablamos en Mieres, los
400 0.70
argumentos para hacerlo no son muy fuertes y tenemos
0.60 cierta circularidad (e.g. no. de individuos capturados tiene
300 0.50 que ver con las estimas de supervivencia).
0.40
200 0.30
100 0.20
0.10
0 0.00
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Captures Surv YEAR
399
400 Number of crossbills caught and survival per capture session. Commented [e34]: Parece que hay una correlación
401 negativa entre supervivencia y número de individuos
capturados.
402 Call:
403 lm(formula = data$survival ~ data$capture * data$product) Esto apoya la idea que la supervivencia esta “diluida” por un
404 gran número de transeúntes. Seria bien hacer un sencillo
405 Residuals: modelo, con supervivencia estimado como variable
406 Min 1Q Median 3Q Max dependiente, y los números capturados y el índice de
407 -0.28517 -0.11651 0.00384 0.09855 0.48084 producción de piñas anual (ver tabla en versión anterior) y
408 su interacción como variables explicativas. Si la
409 Coefficients: supervivencia depende de competición por comida, yo
410 Estimate Std. Error t value Pr(>|t|) esperaría un efecto de todos los variables, incluso la
411 (Intercept) 5.756e-01 3.576e-01 1.609 0.126 interacción. Si
412 data$capture -3.364e-04 7.721e-04 -0.436 0.669 solo hay un efecto del número capturado, seria más
413 data$productC 3.231e-01 4.082e-01 0.791 0.440 probable un efecto de tener mucho transeúnte.
414 data$productN -9.733e-02 3.858e-01 -0.252 0.804
415 data$capture:data$productC -1.315e-03 1.168e-03 -1.125 0.276 Luego hace falta hacer un plot entre supervivencia y
416 data$capture:data$productN 9.321e-05 9.112e-04 0.102 0.920 número de individuos capturados, para ver la correlación.
417
418 Residual standard error: 0.221 on 17 degrees of freedom David: Resultados en el texto
419 Multiple R-squared: 0.2658, Adjusted R-squared: 0.04982
420 F-statistic: 1.231 on 5 and 17 DF, p-value: 0.3377 A simple vista, parece que no existe una relación
421 significativa entre la supervivencia y el número de
individuos capturados y índice de producción de piñas anual
422 Acknowledgments (R2 muy baja)
423 We thank ... (the people who collected the data, anybody that gave advice on the analyses?), and .... Commented [DGB35]: “the people who collected the
424 (project codes and organizations) for funding. This work was supported by funds from the Ministry of data”, en este caso ya están incluidos como autores del ms.
425 Economy and Competitivity, Spanish Research Council (CGL-2016-79568-C3-3-P). “anybody that gave advice on the analyses”, ¿alguien
426 externo a los autores ha proporcionado consejos sobre la
metodología?
427 References
428 Commented [DGB36]: Cuando esté finalizado el ms
revisare todo de nuevo, porque ha habido mucho vaivén de
429 Alonso, D., and J. Arizaga. 2012. The impact of vagrants on apparent survival estimation in a referencias.
430 population of Common Crossbills (Loxia curvirostra). Journal of Ornithology 154:209–217.
431 Alonso, D., J. Arizaga, R. Miranda, and M. Á. Hernández. 2006. Morphological diversification of
432 Common Crossbill Loxia curvirostra populations within Iberia and the Balearics. Ardea 94:99–
433 107.

10
434 Armsworth, P. R., and J. E. Roughgarden. 2008. The structure of clines with fitness-dependent
435 dispersal. The American Naturalist 172:648–657.
436 Benkman, C. W. 2003. Divergent selection drives the adaptive radiation of crossbills. Evolution
437 57:1176–1181.
438 Benkman, C. W. 2016a. The Natural History of the South Hills Crossbill in Relation to Its Impending
439 Extinction. The American Naturalist 188:589–601.
440 Benkman, C. W. 2016b. Matching habitat choice in nomadic crossbills appears most pronounced
441 when food is most limiting. Evolution 71:778–785.
442 Benkman, C. W., J. S. Colquitt, W. R. Gould, T. Fetz, P. C. Keenan, L. Santisteban, C. Patrick, L.
443 Santisteban, and B. J. O. Y. S. Colquitt. 2014. Can selection by an ectoparasite drive a population
444 of Red Crossbills from its adaptive peak? Evolution 59:2025–2032.
445 Benkman, C. W., and A. K. Lindholm. 1991. The advantages and evolution of a morphological novelty.
446 Nature 349:519–520.
447 Benkman, C. W., T. L. Parchman, A. Favis, and A. M. Siepielski. 2003. Reciprocal Selection Causes a
448 Coevolutionary Arms Race between Crossbills and Lodgepole Pine. The American Naturalist
449 162:182–194.
450 Berner, D., and X. Thibert-Plante. 2015. How mechanisms of habitat preference evolve and promote
451 divergence with gene flow. Journal of Evolutionary Biology 28:1641–1655.
452 Björklund, M., D. Alonso, and P. Edelaar. 2013. The genetic structure of crossbills suggests rapid
453 diversification with little niche conservatism. Biological Journal of the Linnean Society 109:908–
454 922.
455 Black, A. R. 1993. Predator induced phenotypic plasticity in Daphnia pulex: Life history and
456 morphological responses to Notonecta and Chaoborus. Limnology and Oceanography 38:986–
457 996.
458 Bolnick, D. I., and S. P. Otto. 2013. The magnitude of local adaptation under genotype-dependent
459 dispersal. Ecology and Evolution 3:4722–4735.
460 Borrás, A., J. Cabrera, and J. C. Senar. 2008. Local divergence between Mediterranean crossbills
461 occurring in two different species of pine. Ardeola 55:169–177.
462 Borrás, A., and J. C. Senar. 2003. Piquituerto común Loxia curvirostra. Pages 588–589 in R. Martí and
463 J. C. del Moral, editors. Atlas de las aves reproductoras de España. Dirección General de
464 Conservación de la Naturaleza-Sociedad Española de Ornitología, Madrid.
465 Burnham, K., and D. Anderson. 2004. Model selection and multimodel inference. Page (K. P. Burnham
466 and D. R. Anderson, Eds.). Springer New York, New York, NY.
467 Castroviejo, S., M. Laínz, G. López Gonzalez, P. Montserrat, F. Muñoz Garmendia, J. Pauci, and L.
468 Villar. 1986. Flora Ibérica. Plantas vasculares de la Península Ibérica e Islas Baleares. Volume I.
469 Real Jardín Botánico-C.S.I.C., Madrid.
470 Choquet, R., J. D. Lebreton, O. Gimenez, A. M. Reboulet, and R. Pradel. 2009a. U-CARE: Utilities for
471 performing goodness of fit tests and manipulating CApture-REcapture data. Ecography
472 32:1071–1074.
473 Choquet, R., and E. Nogue. 2011. E-SURGE 1.8 User’s Manual. English ed. CEFE, Montpellier, France.
474 Choquet, R., L. Rouan, and R. Pradel. 2009b. Program E-Surge: A software application for fitting
475 multievent models. Pages 845–865 Modeling Demographic Processes In Marked Populations.

11
476 Springer US, Boston, MA.
477 Clouet, M. 2000. The breeding biology of the Common Crossbill Loxia curvirostra in the Central
478 Pyrenees. Bird Study 47:186–194.
479 Cramp, S., and C. M. Perrins. 1994. Handbook of the birds of Europe, Middle East and North Africa.
480 Volume VIII. Oxford University Press, Oxford.
481 Darwin, C. 1859. On the origin of species by means of natural selection, or the preservation of
482 favoured races in the struggle for life. John Murray, London.
483 Davis, P. 1964. Crossbills in Britain and Ireland in 1963. Brittish Birds 57:477–501.
484 Duriez, O., S. A. Sæther, B. J. Ens, R. Choquet, R. Pradel, R. H. D. Lambeck, and M. Klaassen. 2009.
485 Estimating survival and movements using both live and dead recoveries: A case study of
486 oystercatchers confronted with habitat change. Journal of Applied Ecology 46:144–153.
487 Edelaar, P., D. Alonso, S. Lagerveld, J. C. Senar, and M. Björklund. 2012. Population differentiation
488 and restricted gene flow in Spanish crossbills: Not isolation-by-distance but isolation-by-
489 ecology. Journal of Evolutionary Biology 25:417–430.
490 Edelaar, P., and D. I. Bolnick. 2012. Non-random gene flow: an underappreciated force in evolution
491 and ecology. Trends in Ecology & Evolution 27:659–665.
492 Edelaar, P., R. Jovani, and I. Gomez-Mestre. 2017. Should I change or should I go? Phenotypic
493 plasticity and matching habitat choice in the adaptation to environmental heterogeneity. The
494 American Naturalist 190.
495 Edelaar, P., A. M. Siepielski, and J. Clobert. 2008. Matching habitat choice causes directed gene flow:
496 a neglected dimension in evolution and ecology. Evolution 62:2462–2472.
497 Edelaar, P., and K. Terpstra. 2004. Is the nominate subspecies or the common crossbill Loxia c.
498 curvirostra polytypic? I.morphological differences among years at a single site. Ardea 92:93–
499 102.
500 Endler, J. A. 1986. Natural selection in the wild. Princeton University Press.
501 Genovart, M., R. Pradel, and D. Oro. 2012. Exploiting uncertain ecological fieldwork data with multi-
502 event capture-recapture modelling: An example with bird sex assignment. Journal of Animal
503 Ecology 81:970–977.
504 Hargrove, J. W., and C. H. Borland. 1994. Pooled population parameter estimates from mark-
505 recapture data. Biometrics 50:1129–1141.
506 Hendry, A. P., T. Day, and E. B. Taylor. 2001. Population mixing and the adaptive divergence of
507 quantitative traits in discrete populations: a theoretical framework for empirical tests. Evolution
508 55:459–466.
509 Herremans, M. 1988. Measurements and moult of irruptive Common Crossbills (Loxia curvirostra) in
510 Central Belgium. Gerfaut 78:243–260.
511 Holt, R. D., and M. Barfield. 2008. Habitat selection and niche conservatism. Israel Journal of Ecology
512 & Evolution 54:295–309.
513 Kingsolver, J. G., and S. E. Diamond. 2011. Phenotypic selection in natural populations: What limits
514 directional selection? The American Naturalist 177:346–357.
515 Lebreton, J., K. P. Burnham, J. Clobert, and D. R. Anderson. 1992. Modeling survival and testing
516 biological hypotheses using marked animals: A unified approach with case studies. Ecological

12
517 Monographs 62:67–118.
518 Manly, B. F. J. 1985. The statistics of natural selection on animal populations. Page The statistics of
519 natural selection on animal populations. Springer Netherlands, Dordrecht.
520 Marquiss, M., K. A. Hobson, and I. Newton. 2008. Stable isotope evidence for different regional
521 source areas of common crossbill Loxia curvirostra irruptions into Britain. Journal of Avian
522 Biology 39:30–34.
523 Newton, I. 2006. Movement patterns of Common Crossbills Loxia curvirostra in Europe. Ibis 148:782–
524 788.
525 O’Brien, S., B. Robert, and H. Tiandry. 2005. Consequences of violating the recapture duration
526 assumption of mark-recapture models: A test using simulated and empirical data from an
527 endangered tortoise population. Journal of Applied Ecology 42:1096–1104.
528 Parchman, T. L., C. A. Buerkle, V. Soria-Carrasco, and C. W. Benkman. 2016. Genome divergence and
529 diversification within a geographic mosaic of coevolution. Molecular Ecology 25:5705–5718.
530 Pradel, R. 2005. Multievent: An extension of multistate capture-recapture models to uncertain
531 states. Biometrics 61:442–447.
532 Pradel, R. 2009. The stakes of capture–recapture models with state uncertainty. Pages 781–795
533 Modeling Demographic Processes in Marked Populations. Springer US, Boston, MA.
534 Pradel, R., L. Maurin-Bernier, O. Gimenez, M. Genovart, R. Choquet, and D. Oro. 2008. Estimation of
535 sex-specific survival with uncertainty in sex assessment. Canadian Journal of Statistics 36:29–42.
536 Przybylo, R., B. C. Sheldon, and J. Merilä. 2000. Climatic effects on breeding and morphology:
537 Evidence for phenotypic plasticity. Journal of Animal Ecology 69:395–403.
538 Santisteban, L., C. W. Benkman, T. Fetz, and J. W. Smith. 2012. Survival and population size of a
539 resident bird species are declining as temperature increases. Journal of Animal Ecology 81:352–
540 363.
541 Santoro, S., A. J. Green, and J. Figuerola. 2016. Immigration enhances fast growth of a newly
542 established source population. Ecology 97:1048–57.
543 Scheiner, S. M. 2016. Habitat choice and temporal variation alter the balance between adaptation by
544 genetic differentiation, a jack-of-all-trades strategy, and phenotypic plasticity. The American
545 Naturalist 187:633– 646.
546 Senar, J. C., and A. Borrás. 1985. Crossbill irruptions. British Trust for Ornithology News 141:6.
547 Senar, J. C., A. Borras, T. Cabrera, and J. Cabrera. 1993. Testing for the relationship between
548 coniferous crop stability and common crossbill residence. Journal of Field Ornithology 64:464–
549 469.
550 Summers, R. W., R. J. G. Dawson, and R. E. Phillips. 2007. Assortative mating and patterns of
551 inheritance indicate that the three crossbill taxa in Scotland are species. Journal of Avian
552 Biology 38:153–162.
553 Summers, R. W., D. C. Jardine, M. Marquiss, and R. Proctor. 1996. The biometrics of invading
554 Common Crossbills Loxia curvirostra in Britain during 1990–1991. Ringing & Migration 17:1–10.
555 Svensson, L. 1992. Identification guide to european passerines. Fourth edition. British Trust for
556 Ornithology.
557

13
558 Tables and figures (in text appearance order):

Initial
No. Sex Assignment Capture Survival np Deviance QAICc ΔAICc wAICc
State
Modeling sex assignment probability
1 sex . t sex . t t a . sex + t 164 17185.35 17521.13 9.82 0.00
2 sex + t sex . t t a . sex + t 138 17229.81 17511.31 0.00 0.00
3 t sex . t t a . sex + t 137 17247.60 17527.02 15.71 0.00
4 sex sex . t t a . sex + t 112 17845.46 18073.07 561.76 0.00
5 c sex . t t a . sex + t 111 18004.79 18230.34 719.03 0.00
Modeling capture probability
2 sex + t sex . t t a . sex + t 138 17229.81 17511.31 39.04 0.00
6 sex + t sex + t t a . sex + t 113 17261.36 17491.04 18.78 0.00
7 sex + t t t a . sex + t 112 17262.10 17489.72 17.45 0.00
8 sex + t sex t a . sex + t 88 17295.70 17473.93 1.66 0.05
9 sex + t c t a . sex + t 87 17296.09 17472.27 0.00 0.12
Modeling Initial state probability
9 sex + t c t a . sex + t 87 17296.09 17472.27 0.15 0.12
10 sex + t c c a . sex + t 61 17349.05 17472.12 0.00 0.12
Modeling survival probability
10 sex + t c c a . sex + t 61 17349.05 17472.12 2.37 0.12
11 sex + t c c a + sex + t 59 17350.75 17469.75 0.00 0.41
12 sex + t c c a+t 58 17353.45 17470.42 0.66 0.29
13 sex + t c c a . sex 36 17440.00 17512.37 42.62 0.00
14 sex + t c c a + sex 34 17444.00 17512.34 42.59 0.00
15 sex + t c c a 33 17445.16 17511.48 41.72 0.00
Table 1. Model selection with the complete dataset. Abbreviations: np, number of parameters;
QAICc, Quasi-Akaike Information Criterion corrected for over-dispersion; ΔAICc, AICc differences
between models; wAICc, Akaike weight (support of the current model with respect to the
candidate set of models); t, time effect; a, age effect, c, constant effect. The best model selected
for each model selection are in bold characters.
559

14
 1

0.9

0.8

0.7
Apparent survival rate

0.6

0.5

0.4

0.3

0.2
Adult
Yearling
0.1
Juvenile

0
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
Year
Figure 1. Apparent survival during the years of study. Given are the estimates for three age classes (from top to bottom: yearling, adult, juvenile; see
legend). For graphical clarity estimates are represented without the (minor) differences between males and females.

15
(a) Beak width
No. Sex Assignment Capture Initial State Survival np Deviance QAICc ΔAIC wAIC
1 t c c a+t 42 2556.43 2644.09 2.76 0.11
Modeling beak width effect
5 t c c (a + t) + ß 43 2554.55 2644.38 3.04 0.09
6 t c c (a + t) + ß + ß2 44 2549.32 2641.33 0.00 0.42
7 t c c (a + t) + ß2 43 2551.69 2641.53 0.19 0.38

(b) Beak height

No. Sex Assignment Capture Initial State Survival np Deviance QAICc ΔAIC wAIC
1 t c c a+t 50 5107.70 5210.25 0.00 0.45
Modeling beak height effect
2 t c c (a + t) + ß 51 5107.70 5212.35 2.10 0.16
3 t c c (a + t) + ß + ß2 52 5106.50 5213.26 3.01 0.10
4 t c c (a + t) + ß2 51 5106.63 5211.28 1.04 0.27
Table 2. Model selection with the effect of (a) beak width and (b) beak height on apparent survival.
Abbreviations: np, number of parameters; QAICc, Quasi-Akaike Information Criterion corrected for
over-dispersion; ΔAICc, AICc differences between models; wAICc, Akaike weight (support of the
current model with respect to the candidate set of models); t, time effect; a, age effect; c, constant
effect; ß, linear effect of the covariate; ß2, quadratic effect of the covariate. The best model
selected for each model selection are in bold characters.

16
 0.70 Commented [SS37]: cuando estemos a punto de
enviarlo, puedo preparar una figura más bonita que esta.
0.65

0.60

0.55

0.50

0.45
Apparent survival rate

0.40

0.35

0.30

0.25

0.20

0.15 Adults

0.10 Yearlings
Juveniles
0.05

0.00
9.6 9.8 10 10.2 10.4 10.6 10.8 11 11.2 11.4 11.6 11.8 12 12.2 12.4 12.6 12.8
Beak Width (mm)
Figure 2. Model-averaged annual apparent survival rates of crossbills in relation to their beak width. Given are the estimates for three age
classes (from top to bottom: yearling, adult, juvenile; see legend). The 95% CI are not shown for clarity (at optimal beak width; juveniles: 0.32-
0.43, yearlings: 0.57-0.68, adults: 0.47-0.59).

17
Figure 3. Probability densities for survival quadratic selection gradients. Highest probability
density values of quadratic selection marked in grey. The quadratic selection coefficient
obtained in the study is highlighted (-0.35). Figure adapted from Kingsolver and Diamond
2011.
18
Commented [e38]: I would move this to the main text.
JCS: I fully agree. We need a section in the paper (last part
RESULTS) about selection gradients. We can not provide the
value of selection for our Crossbills in DISCUSION without
providing the value and computation in RESULTS
David: I moved the figure to the main text (now Figure 3)
and I mentioned in methodology:
"Finally, we calculated the standardized quadratic selection
gradient (γ) to estimate the strength and mode of natural
selection. Negative values of γ are necessary for stabilizing
selection, which are positive values of γ are necessary for
disruptive selection."
and in results:
"For beak width, the best model included the linear and
quadratic effect of the covariate (see Figure 2) (coefficient
on logit scale for the linear term = 0.22, SE = 0.14,
coefficient for the quadratic term on logit scale = -0.35; SE
= 0.18) "
Additional file:

Additional file 1. Multievent probabilistic framework of the study

Multievent models combine information from events with the underlying states to estimate
probabilities of several parameters. A multievent model accounts for three parameter types: Initial
State, State Transition and Event probabilities. Here we report the parameterization of the models
used for the present study. The first parameter is Initial State, relative to the first capture of
individual, which expresses the probability of being in a specific state. In our case, Initial State (IS) can
be considered as a proxy of the sex ratio of the population (see Pradel (2009) and Genovart et al.
(2012) for a discussion on this) as it represents the probability that a first-captured individual is a
male when first captured. Since an individual must be alive at the first capture, the Initial State does
not contemplate the “dead” state:

F M
1–π π

where π is the probability, at a certain occasion t, that a first detected crossbill is male or female. The
second parameter, Transition (T), represents the transition probability between states that is the
probability of survival:

F M ✝
F φF 0 1 - φF
M 0 φM 1 - φM
✝ 0 0 1

The third state, Event, is related to the probabilities of being recaptured according to the event-
mediated information on states. We decomposed Event into two steps. The first step is Capture (C),
which represents the probability of recapture (δ):

non-detected F detected M detected

F 1 - δF δF 0
M 1 - δM 0 δM
✝ 1 0 0

where δF and δM refer respectively to the probability a female and a male to be recaptured.
Conditional on being captured, the second step is Sex Assessment (SA), which represents the
probability that an individual has been sexed when detected (Pradel et al. 2008).

3= Detected-sex-
0= non-detected 1= F-detected 2= M-detected
unknown
Non-detected 1 0 0 0
F detected 0 γF 0 1 - γF
M detected 0 0 γM 1 - γM

19
Additional file 2: Model selection with the reduced datasets

(a) Beak width

No. Sex Assignment Capture Initial State Survival np Deviance QAICc ΔAIC wAIC
Modeling sex assignment probability
1 sex + t c c a + sex + t 45 2553.91 2648.11 1.19 0.05
2 t c c a + sex + t 44 2554.90 2646.92 0.00 0.10
3 sex c c a + sex + t 26 2651.04 2704.44 57.52 0.00
4 c c c a + sex + t 25 2651.82 2703.12 56.20 0.00
Modeling survival probability
2 t c c a + sex + t 44 2554.90 2646.92 2.83 0.10
5 t c c a+t 42 2556.43 2644.09 0.00 0.85
6 t c c a + sex 26 2651.73 2705.12 61.03 0.00
7 t c c a 25 2652.09 2703.38 59.29 0.00

(b) Beak height

No. Sex Assignment Capture Initial State Survival np Deviance QAICc ΔAIC wAIC
Modeling sex assignment probability
1 sex + t c c a + sex + t 53 5107.05 5215.91 1.47 0.10
2 t c c a + sex + t 52 5107.69 5214.45 0.00 0.18
3 sex c c a + sex + t 30 5287.63 5348.55 134.11 0.00
4 c c c a + sex + t 29 5287.88 5346.74 132.29 0.00
Modeling survival probability
2 t c c a + sex + t 52 5107.69 5214.45 4.20 0.18
5 t c c a+t 50 5107.70 5210.25 0.00 0.73
6 t c c a + sex 30 5204.25 5265.18 54.93 0.00
7 t c c a 29 5204.27 5263.13 52.88 0.00
Supplementary material 2. Model selection with the reduced datasets containing only the
individuals for which we disposed of measures of (a) beak width and (b) beak height
Abbreviations: np, number of parameters; QAICc, Quasi-Akaike Information Criterion corrected
for over-dispersion; ΔAICc, AICc differences between models; t, time effect; a, age effect, c,
constant effect. The best model selected for each model selection are in bold characters.

20