1Male territorial aggression does not drive conformity to local vocal culture in a passerine bird
2
3This is a pre-review pre-print of the following published article:
4Parra, M., A. C. Dalisio , W. E. Jensen, and T. H. Parker. 2017. Male territorial aggression does
5not drive conformity to local vocal culture in a passerine bird. Ethology. 123: 800-810. doi:
610.1111/eth.12648
7
8Marcelina Parra1
9Anthony C. Dalisio2,3,4
10William E. Jensen3
11Timothy H. Parker1,5
12
131. Department of Biology, Whitman College, 345 Boyer Ave, Walla Walla, WA 99362, USA
142. Circumpolar Studies, Sterling College, Craftsbury Common, VT, 05827, USA
153. Department of Biological Sciences, Emporia State University, 1 Kellogg Circ, Emporia, KS
1666801, USA.
174. Current address: 204 Water St, Waterville, ME 04901, USA.
185. Corresponding author: [email protected]; +1.509.526.4777
19
20Abstract In many songbird species, young males learn songs from neighbors and then settle
21nearby, thus creating neighborhoods of conformity to local vocal culture. In some species males
22appear to postpone song learning until after dispersal, possibly to facilitate conformity to local
1 23dialects. Despite decades of study, we still lack a consensus regarding the selective pressures
24driving this delayed song learning. Two common hypothetical benefits to conformity, and thus
25delayed song learning, are rooted in male territorial interactions; males preferentially produce
26local song either to avoid detection as new arrivals (deceptive mimicry) or to be more
27effectively recognized as conspecific territory holders. The dickcissel (Spiza americana) is an
28ideal species in which to study these hypotheses. Males of this species appear to delay song
29learning until they arrive at their first adult territory, each individual sings a single song type,
30and conformity to the local song culture is high. Using playback, we contradicted both of the
31territorial hypotheses described above; male dickcissels did not respond differentially to local vs
32foreign song playback treatment. We are confident in this lack of difference because dickcissels
33clearly responded less strongly to a third treatment, neighbor song, than to the other two
34treatments, demonstrating sufficient power in our experimental design (and providing the first
35evidence of the dear-enemy effect in dickcissels). Our results raise the question of why
36dickcissels respond equally aggressively to both local and foreign songs when other bird species
37often show reduced aggression towards foreign song. If reduced aggression to foreign song is
38not ubiquitous in species that achieve conformity through delayed learning then selection from
39among-male territorial interaction seems unlikely to be a general explanation for such delayed
40learning. Reduced aggression in response to foreign songs in other species may be due to
41reduced exposure to the stimulus of foreign song or to different cost-benefit trade-offs when
42responding to songs that deviate from the local average.
43
44
2 45
46Keywords
47dear-enemy; deceptive mimicry; dickcissel, playback; recognition; song; Spiza americana
48
3 49Introduction
50
51Geographic variation in sexual signals within species is common, and is particularly conspicuous
52in the songs of many Oscine birds. These song variants, sometimes referred to as dialects
53(Baker & Cunningham 1985), can be explained at the proximate level by males learning songs
54from nearby adult conspecifics (Beecher & Brenowitz 2005). As long as individuals tend to learn
55their songs from local tutors, dispersal post-learning is limited, and new song variants emerge
56occasionally (e.g., through learning errors), song dialects should emerge (Krebs & Kroodsma
571980). Although song dialects result from cultural rather than genetic variation, this does not
58preclude a role for selection in influencing individual behaviors that promote the emergence of
59dialects. For instance, in some species individuals appear to learn song after dispersal rather
60than in their first few months of life (Payne & Payne 1997). This delay of learning until after
61dispersal promotes the development of dialects, and thus may be selected for if dialect
62conformity conveys a fitness advantage.
63
64A primary function of song in many Oscine species is male signaling of territorial ownership.
65Selection could favor delayed song learning, and thus conformity to a local dialect, if this
66conformity somehow aided males as they sought to establish and maintain their claim to a local
67territory. An early hypothesis, ‘deceptive mimicry’, posited that conforming to the local dialect
68could result in reduced aggression from conspecifics, allowing the newcomer to more easily
69integrate into the neighborhood than a bird with a foreign song (Payne 1981; Rohwer 1982;
70Payne 1983). If this is the case, males should respond more aggressively to a foreign dialect
4 71than a local dialect. However, this prediction is rarely met in playback studies (e.g., Lemon
721967; McGregor 1983; Podos 2007; Osiejuk et al. 2012; Bradley et al. 2013), and so the
73deceptive mimicry mechanism is unlikely (but see e.g., Baker 1982; Briefer et al. 2010).
74Conversely, sharing song types with territorial neighbors could result in increased recognition as
75a conspecific and thus aid a newcomer in defending a territory against established males
76(discussed in Baker 1994) or facilitate longer territory tenure in established males (Beecher et
77al. 2000; but see Nelson & Poesel 2013). If this is the case, males should respond more
78aggressively to playback of an unfamiliar individual singing the local dialect than to playback of
79a foreign dialect. This prediction holds for many species of birds (e.g., Lemon 1967; McGregor
801983; Podos 2007; Osiejuk et al. 2012; Bradley et al. 2013). Of course the existence of
81heightened response to local versus foreign playback does not demonstrate that this selection
82pressure caused the evolution of local song learning and thus dialects. Conformity to local
83dialects may not be favored by selection stemming from interactions among males. It could be,
84for instance, that female choice drives dialect conformity (e.g., Searcy & Andersson 1986;
85Danner et al. 2011; Nelson & Poesel 2013). Alternatively, dialect conformity may be nothing
86more than an epiphenomenon of learning songs from conspecifics and limited post-learning
87dispersal.
88
89Dickcissels (Spiza americana) are songbirds in which individuals produce a single song type and
90conform to striking song dialects that change over the scale of a few km (Schook et al. 2008).
91They are common breeders on the Great Plains of North America (Temple 2002). Circumstantial
92evidence strongly suggests that male dickcissels delay song learning until they arrive at their
5 93first adult territory after returning from wintering in northern South America (Schook et al.
942008). One researcher at our study site banded 317 nestlings that later fledged and only one of
95the banded male nestlings was detected as a returned adult (B. Sousa, personal
96communication). This observation, combined with the fact that our site encompasses multiple
97local dialects, leads to the conclusion that a male settling on our site likely did not fledge in the
98area of the dialect he sings as an adult, or even in any of the surrounding dialect areas. We
99cannot rule out the possibility that males roam and learn multiple dialects their first summer
100and then choose to sing which ever song matches the dialect of their adult territory (Marler &
101Peters 1982; Nelson & Marler 1994). However, given subtle changes in dialects detectable on
102the scale of 100’s of meters and the high degree of conformity to local dialects (Schook et al.
1032008), this mode of learning with attrition would substantially limit dispersal options. Further,
104given the readiness with which dickcissels disperse to occupy newly available habitats (Temple
1052002), limited dispersal appears unlikely. Finally, young males tend to remain within a few
106hundred meters of their nest in the weeks after fledging, often into the period when adult
107males cease defending territories and disperse (B. Verheijen, personal communication). Thus,
108most young males are likely exposed to only a few local dialects at best before singing subsides
109and dialects dissolve in late summer.
110
111Because dickcissels appear to show delayed song learning that promotes high dialect
112conformity, we set out to test hypotheses explaining dialect conformity as an adaptation to
113influence male-male interactions. If dickcissels conform to the local dialect to reduce aggression
114from conspecifics (deceptive mimicry), we expected dickcissels would respond more
6 115aggressively to playback of a foreign dialect than to playback of song from the local dialect. If
116dickcissels conform to the local dialect to be more readily recognized as conspecific, we
117expected that dickcissels would respond more aggressively to playback of an unfamiliar
118individual’s song from the local dialect than to song from a foreign dialect.
119
120
121Methods
122We conducted a playback experiment on male dickcissels in June and July 2014 at Konza Prairie
123Biological Station (KPBS) (39°05’ N, 96°35’ W, approximately 420 m elevation) near Manhattan,
124KS, USA.
125
126We recorded male dickcissel song onto digital media using a Marantz PMD680 recorder and a
127Sensheiser ME66 shotgun microphone. We recorded dickcissels from a distance of
128approximately 10 m for at least three song repetitions to increase the chance of recording at
129least one clear song from each bird with minimal background noise for use in playback trials.
130We edited songs using Raven 1.4 (Cornell Laboratory of Ornithology, Ithaca, NY, USA).
131
132Our primary goal was to compare responses of territorial male dickcissels between playback of
133local dialect song and highly divergent foreign dialects, but we also included a third treatment
134of playback of song from an adjacent neighbor. We added this third treatment because we had
135strong a priori expectation of a differential response to neighbor song versus non-neighbor song
136(e.g., Brooks & Falls 1975; Godard 1993; Wilson & Vehrencamp 2001; Hyman 2005; Briefer et
7 137al. 2008; Wei et al. 2011), and so this treatment could serve to establish the sufficiency of our
138protocol to detect differential responses to playback.
139
140We delineated each target male’s territory one to three days before a trial by repeatedly
141flushing him from song perches (Zimmerman & Finck 1989), recording GPS coordinates, and
142noting landmarks on a paper map. When repeatedly flushed in this manner, territorial male
143dickcissels reliably move among favored song perches on their territory and typically continue
144with singing with almost no pause. We did not band birds, as luring dickcissels into nets
145requires playback of song, so banded birds have previous experience with playback that might
146influence their response. Instead we carefully mapped each territory and returned within one
147or two days. Our (THP, WEJ) observations of within-season site fidelity of banded birds on the
148site over six field seasons demonstrated that males typically remain within the same territory
149over the course of a season. The daily apparent survival rate during the breeding season for
150territorial males on our site is 0.93 (unpublished data), meaning that that there is a 93% chance
151that a male will persist from one day to the next. Thus, we are confident that in nearly all cases,
152the bird receiving the playback treatment was the one whose territory was mapped during the
153previous three days. We scouted birds in clusters of three or four adjacent territories. The
154minimum closest distance between clusters was 205 m, and the mean closest cluster distance
155was 355 m.
156
157After birds were scouted, we determined whether each potential target bird showed a high
158degree of dialect conformity by qualitatively evaluating its song recordings in Raven. We
8 159considered males conformers if, based on careful examination of song spectrograms, a large
160majority of notes in their song closely matched, in terms of frequency, duration, and order,
161notes in the songs of its neighbors. We only used conformers in the experiment. This qualitative
162method of classifying song similarity is simple and highly repeatable in dickcissels (Schook et al.
1632008). More than 75% of birds we evaluated met this high standard of dialect conformity within
164their neighborhood. The songs that did not meet this standard were nearly all still quite similar
165and were readily identifiable as members of the broader cultural group found on our site. This
166stands in contrast to the foreign songs we recorded at other sites which differed dramatically
167from local songs.
168
169Our local song treatment was designed to mimic an unfamiliar individual singing a song that
170conformed to the local dialect. Because dickcissel dialects changes over relatively small
171geographic scales, we assumed that within an area of high conformity it was possible that any
172given male had heard the song of every other male. Thus to find an unfamiliar rendition of the
173local dialect, we planned to use songs recorded the previous decade because we assumed that
174these would likely have been sung by unfamiliar individuals. However, dialects had changed
175sufficiently that we could not find songs amongst old recordings that showed high conformity
176to modern dialects. Therefore we instead opted to create our local songs by combining the
177songs of multiple different individual males in a composite. Because we were comparing local
178songs to foreign songs and we wished to minimize any differences between these two
179treatments except song origin, we also created composite foreign song using the same method.
180
9 181When we created composite songs for our local and foreign song playback treatments (Fig. 1),
182we always combined songs from males whose songs had identical or nearly identical series of
183notes within phrases. Note conformity within phrases is very high among neighboring
184dickcissels, though the spacing of phrases is somewhat variable (Schook et al. 2008) (Fig. 1).
185When creating a composite song, we adopted the phrase spacing of one of the males whose
186song we were using, and replaced a subset of phrases and parts of phrases with the
187corresponding phrase portions from one or two other males. No stimulus male had its song
188used more than once in a given treatment category (neighbor, local, or foreign) to prevent
189pseudoreplication (Kroodsma 1989), though some songs used as a neighbor treatment in one
190cluster of males also contributed a phrase to a local treatment in another cluster of males. The
191parts of the local composite songs were from the same dialect as the target bird, but from sites
192on average 414 m away, and at least 100 m away (with one exception where one third of local
193song came from a non-adjacent territory ~60 m away) from the target bird’s territory. We
194expected that by using non-neighbor song, the focal birds would follow the pattern found in
195many species of treating playback of non-neighbor song more aggressively than playback of
196neighbor song, presumably because non-neighbor song is treated as a territorial invasion.
197Further, by creating composite songs, we hoped to reduce the chance that the focal male
198would recognize the song as coming from a particular known individual and thus treat that song
199differently than an unfamiliar local song. This precaution is rarely taken in playback studies
200comparing responses of local to foreign song, and this makes our method particularly robust to
201any bias that could stem from recognition of a local song as belonging to a particular familiar
202individual. Further, even without this precaution, studies usually detect greater aggression
10 203towards local song than towards foreign song (e.g., Lemon 1967; McGregor 1983; Podos 2007).
204As stated above, to control for effects of using composite local songs, we created composite
205foreign treatment song by splicing together songs of two different individuals (of the same
206foreign dialect from the same distant site) recorded previously between 2005 and 2011. The
207closest location of a foreign song recording to our study site was 22 km and the mean distance
208was 158 km. At this distance, songs show many dramatic differences (Schook et al. 2008) (Fig
2091). We recorded neighbor songs from males with territories adjacent to focal playback males.
210We included the neighbor song treatment only to assess whether males could respond
211differentially to different playback songs that were expected to induce a differential response
212(e.g., Godard 1993; Wilson & Vehrencamp 2001; Hyman 2005; Wei et al. 2011) and not because
213we wished to test hypotheses about the effects of neighbor songs per se. We did not
214manipulate neighbor song by cutting and splicing because we wanted to be sure the neighbor
215song was recognized as a song produced by a particular individual. We did, however,
216manipulate all recordings by eliminating sounds from other birds or from insects, and by
217filtering background noise below 2 kHz.
218
219To each target bird we played two of the three possible treatments, each on a different day. We
220initiated all trials between 06:00 and 09:00 CST. We did not conduct trials if it was raining or if
221wind appeared to be influencing male behavior (reduced perching and singing) or prevented us
222from making a clear audio recording of the trial. For all treatments, we placed the speaker near
223the target bird’s territorial boundary but within the estimated territory and facing the territory
224center. For birds receiving the neighbor treatment, we placed the speaker adjacent to the
11 225territory of the neighbor whose recorded song we were using for the playback. The observer
226stood at least 20 m behind the speaker to get the best view of the entire territory. We used an
227AmpliVox SW720 integrated amplifier and speaker and played each song to obtain peak volume
228of 90 dB as detected with a RadioShack Sound Level Meter at 1 m in front of the speaker. We
229tested the volume of each recording in the field, but >300 m from test territories, prior to
230playback. This produced song with high fidelity and an absence of noticeable distortion. We
231observed the target bird for one minute to confirm its presence on its territory, then played
232either a local, foreign, or neighbor song for two minutes. The song played at a rate of once
233every five seconds, the typical song rate for dickcissels (Schook et.al 2008). We continued to
234observe the target bird for two minutes after playback. Within 15 second intervals before,
235during, and after playback, we estimated the closest approach to the speaker, tallied the
236number of flights, and recorded an estimated average distance of the bird from the speaker.
237We determined number of songs by listening to an audio recording of the trial. We placed
238marking flags at four to eight points 10 m and 20 m away from the speaker to help estimate
239distance of the target bird from the speaker. We do not expect that playback has long-term
240detrimental impacts on dickcissel males. In our previous work that used playback to attract
241males to nets for capture, playback often lasted much longer than the two minutes imposed in
242our current study, and banded birds in those prior studies rarely abandoned territories and
243often returned to the site in later years. All song recording of focal birds, song classification and
244processing, and field trials were implemented by one researcher (M. Parra), and so it was not
245possible to blind the observer to the treatment category of the individual trial.
246
12 247We anticipated high variance among males in strength of response that would be unrelated to
248playback treatment, so each male received two treatments to allow us to partition variance
249resulting from differences among males in their tendency to respond to playback. We
250systematically distributed treatments across dates and thus dates did not differ among the
251three treatments (ANOVA: F2,61 = 0.09, p = 0.91). We also randomized treatment order amongst
252individual males and thus order (first vs. second) did not differ amongst the three treatments
2 253(GLM with binomial error: Wald χ 2= 0.09, p = 0.95). Because part of the among-male variance
254could come from positions of the speaker and the observer, we marked the position of the
255speaker and the observer with flags to be consistent for the second trial, which took place 1 to
2562 days after the first trial.
257
258We used generalized linear mixed models (SPSS version 23) to analyze responses to our
259playback experiment as measured by the number of songs, number of flights, and mean
260distance from the speaker. Because of the anticipated importance of partitioning variance
261among territories, we excluded the few trials in which the corresponding second trial on the
262given territory was never completed and we included individual territory identity as a random
263effect in all analyses. For the count data (# songs, # flights), we assumed a negative binomial
264distribution. However, the average distance to speaker was skewed right, so we performed a
265square root transformation which produced a distribution that did not differ substantially from
266normal (during playback: Kolmogorov-Smirnov statistic = 0.101, df = 64, p = 0.117; after
267playback: Kolmogorov-Smirnov statistic = 0.119, df = 64, p = 0.024), and thus is reasonable to
268analyze assuming normal error (McDonald 2014). We report 2-tailed p-values for all tests. In
13 269our first round of analyses, we assessed whether territorial male dickcissels appeared to
270respond to the song playback by comparing behavior among the three experimental time
271periods: before, during, and after the playback. For this comparison, we used a standardized
272per-two-minute rate. This meant doubling the one minute pre-playback counts. We limited our
273pre-playback counts to one minute to minimize the chance that we would lose track of the
274target individual before it was time to begin the trial. For both the count variables, values
275differed significantly between the during-playback time period and both before and after, but
276the before and after periods were not significantly distinguishable from each other (see
277Results). Thus we elected to compare responses to the different playback treatments (foreign,
278local, neighbor) with both count variables during, but not after, playback. Average distance to
279the speaker was significantly shorter during and after playback than before playback, but did
280not differ between the periods during and after playback (see Results). Thus we also chose to
281compare among the experimental treatments based on average distance to speaker, and we
282chose to use data from both during and after the playback. In the analyses of response to the
283three experimental treatments, we also included trial number as a fixed effect to account for
284any possible habituation effects in each male’s second trial. The analysis of the response of
285song rate to experimental treatment during playback was unsuccessful because the final
286Hessian matrix was not positive definite and so did not produce reliable results. In an attempt
287to create a successful statistical model assessing song response, we combined the during- and
288after-playback data and included trial identity nested within experimental treatment as a
289random effect and during versus after playback as a fixed effect. However, this model failed for
290the same reason and so we were unable to assess response to the experimental treatments as
14 291measured by song. We wanted to base our inferences on strengths of effects (Nakagawa &
292Cuthill 2007), and we wished to avoid the elevated risk of type I error that occurs when
293assessing a hypothesis with multiple comparisons. Thus, for all pairwise-comparisons of means,
294we calculated Hedge’s d, a standardized mean difference commonly used to assess strength of
295effect in meta-analysis. A common set of benchmarks is that Hedge’s d of approximately 0.2 is a
296weak effect, 0.5 is a moderate effect, and 0.8 is a strong effect (Rosenberg et al. 2013). We then
297calculated average Hedge’s d values and meta-analytical 95% confidence limits (controlling for
298non-independence among measures) for each set of comparisons testing the same general
299hypothesis (Nakagawa & Santos 2012). Our calculation of confidence limits accounted for non-
300independence by accounting for the correlations between pairs of different measures of
301dickcissel response. For instance, the greater the correlation between number of flights and
302distance to speaker across trials, the less independent information these two response
303variables contribute to the calculation of the confidence limits. In cases in which no correlation
304existed because the measures were based on different sets of trials (for instance, when
305comparing number of flights in response to neighbor song vs. number of flights in response to
306local song) we made the conservative choice to assign the highest correlation value in our
307correlation matrix thus reducing the degree to which those two measures contributed
308independent information to the calculation of our confidence limits. We calculated average
309Hedge’s d and associated confidence limits for (1) comparisons of before-playback behaviors to
310during-playback behaviors to assess the overall hypothesis that male dickcissels did respond to
311playback, (2) for comparisons of response to first versus second trials, (3) for comparisons of
312responses to playback of foreign versus local song, and (4) for comparisons of playback of
15 313neighbor song versus the other two treatments. When calculating averages we assumed that
314more flights and a lower average distance to the speaker indicated greater aggression. For
315songs, we assumed that a reduced rate during the trial and an elevated rate after the trial
316indicated greater aggression. We made this assumption based on pilot data from 2007 in which
317we observed dickcissel males reduce their song rate in response to playback and then elevate it
318relative to pre-playback levels as soon as the playback ceased. We observed a similar pattern in
319our 2014 data, though the post-playback song rate was not significantly greater than the pre-
320playback rate (see Results). Our averages of Hedge’s d provide robust tests of these hypotheses
321without elevated type I error from multiple comparisons.
322
323
324Results
325
326Territorial male dickcissels responded strongly to song playback treatments in their territory by
327approaching the speaker and remaining near it for the duration of the trial (Table 1, 2). Males
328also made more frequent flights and decreased their song rate during the two minutes of
329playback (Table 2). The absolute value of the effect size averaged across these response
330variables was moderate to large and convincing with very small confidence limits (Fig 2). During
331the two minutes after the playback ceased, males remained in close proximity to the speaker,
332but returned to approximately pre-playback rates of flight and song (Table 2).
333
16 334Male dickcissels responded less strongly, based on approach distance and number of flights,
335during the second playback trial relative to the first (Table 3, 4). The averaged effect size was
336moderate with small confidence limits that do not approach zero (Fig 2).
337
338Territorial male dickcissels responded almost identically to the local vs. foreign song playback
339treatment according to the three response variables assessed (Table 5). The average effect size
340was very close to zero and confidence intervals encompassed zero (Fig 2).
341
342However, focal males responded more strongly to the local and foreign treatments than to the
343neighbor treatment as measured by the number of flights and the mean distance to the
344speaker (Table 6). The averaged effect size was moderate to large with confident limits well
345clear of zero (Fig 2).
346
347
348Discussion
349
350Our results contradict the applicability of both the deceptive mimicry hypothesis and the
351recognition hypothesis in the dickcissel. The lack of differential response to local and foreign
352playback treatments suggests the apparent tendency of male dickcissels to selectively learn the
353local dialect does not result from selection related to territorial interactions. Thus, some other
354hypothesis is necessary to explain the delayed song learning that promotes dialect
355development in this species. Our results also raise important questions about why male
17 356dickcissels respond equally aggressively to local and to foreign dialects while territorial males in
357so many other species (e.g., Lemon 1967; McGregor 1983; Podos 2007; Osiejuk et al. 2012;
358Bradley et al. 2013) respond more aggressively to local song than to foreign song.
359
360We are confident in our observed lack of difference in response to local and foreign song
361playback for two reasons. First, dickcissels responded strongly to playback and thus were not
362simply ignoring both treatments. Second, we readily detected significantly reduced aggression
363towards playback of neighbor song relative to local and foreign dialect treatments. The dear-
364enemy effect (Brooks & Falls 1975) explains reduced aggression between individuals who have
365mutually established a territorial boundary. This is a widely recognized effect in birds (Godard
3661993; Wilson & Vehrencamp 2001; Hyman 2005; Wei et al. 2011), especially during the peak of
367the breeding season (Briefer et al. 2008) as in our study, and it is an effect not limited to birds
368(e.g., Davis 1987). Besides increasing confidence that we had the power to detect difference in
369response among playback treatments, the results of our comparisons of responses to neighbor
370versus stranger song are the first indication of the dear-enemy effect in the dickcissel. We
371would like to point out that our neighbor song treatment songs were unmanipulated whereas
372the two songs from our primary hypothesis tests were both spliced together from multiple
373individuals. As explained in the methods, we avoided manipulating the neighbor song because
374we wanted to be sure the neighbor was recognized as a particular individual. Because of this
375difference in song preparation, we cannot exclude the possibility that the difference in
376responses to the neighbor songs and the two other treatment categories was due to the
377difference in manipulation. However, this seems highly unlikely. Manipulation of the song
18 378should, if done poorly, weaken its effectiveness as a signal. Thus, we would have expected a
379weaker response to the manipulated songs if the manipulation itself were impacting their
380effectiveness as a signal. However territorial male dickcissels responded most strongly to the
381two manipulated song treatments.
382
383Although a failure to support the deceptive mimicry hypothesis is not a surprise, our results are
384in contrast to the many studies that have found reduced aggression to foreign songs (e.g.,
385Lemon 1967; McGregor 1983; Podos 2007; Osiejuk et al. 2012; Bradley et al. 2013) as predicted
386by the recognition hypothesis. Our failure to support the recognition hypothesis in one species
387does not disprove the applicability of this hypothesis more widely, but the recognition
388hypothesis has occasionally been contradicted in other species (e.g., Hansen 1984; Colbeck et
389al. 2010), and it does raise the possibility that the recognition hypothesis is not a general
390explanation for the evolution of dialect conformity.
391
392If the selection pressure promoting delayed song learning in dickcissels does not arise from
393recognition by conspecific males, selection from female choice may drive conformity (Danner et
394al. 2011). In some systems, female preference for males conforming to the local dialect could
395serve to promote female mating with locally adapted males (Searcy et al. 2002). This
396mechanism is unlikely in the dickcissel, however, where both males and females typically
397disperse beyond the range of a typical dialect, males appear to learn their song after dispersal,
398and the geographic extent of dialects is almost certainly much smaller (Schook et al. 2008) than
399the extent of any locally adapted gene complexes. However, degree of conformity to the local
19 400dialect could be a function of male learning ability and thus could signal some aspect of male
401quality. In this scenario, selection could favor females who discriminate among males based on
402dialect conformity (Nowicki et al. 1998). Even if dialect conformity is not a signal related to male
403quality, it could still be a target of female aesthetic preference and thus could act as a sexual
404ornament (Prum 2010), or the clustering of similar signals could itself be attractive to females
405(Wagner 1998). We have not yet investigated the relationship between female choice and male
406dialect conformity in dickcissels.
407
408We also cannot rule out the possibility that delayed song learning in male dickcissels is not
409currently adaptive. The dickcissel is in a monotypic genus and so we cannot compare patterns
410of song learning in congeners, but in the closely related genus Passerina (Klicka et al. 2007),
411males appear to also exhibit delayed song learning (reviewed in Payne 2006; Greene et al.
4122014). Thus, it may be that delayed song learning was favored in the past and has persisted as
413a phylogenetic legacy in the absence of modern selection.
414
415Although our original goal was to assess how selection might act to favor learning of local
416songs, we might also ask about the selection pressures leading males to respond aggressively to
417various types of song playbacks. Some researchers who have detected greater responses of
418territorial males to local than to foreign song playbacks have hypothesized that this differential
419response may be an adaptation to invest more in aggression towards males who pose the
420greatest territorial threat (Tomback et al. 1983; Wright & Dorin 2001; Nicholls 2008; Turcokova
421et al. 2011). If foreign birds are just passing through or are unlikely to attempt to settle, then
20 422investing in challenging those individuals might be selected against. However, in the dickcissel it
423is not uncommon for a male with a foreign dialect to settle in a neighborhood otherwise
424occupied by dialect conformers (our personal observations). These non-conformers are likely
425those males, well documented in dickcissels (Schartz & Zimmerman 1971; Zimmerman & Finck
4261989), who have abandoned unsuccessful territories elsewhere. Although dramatically non-
427conforming dickcissels are, in our experience, a minority (when both moderate and dramatic
428non-conformers were tallied together, they constituted 5.5% of 183 monitored territories this
429site), they may be common enough to provide a selection pressure that favors a territorial male
430who directs equal amounts of aggression to any conspecific male he encounters, regardless of
431dialect.
432
433Another component of this selection environment comes from the risk of mistakenly
434responding to the song of a heterospecific (Hamao 2016). If males commonly encounter other
435species with similar songs, and if responding aggressively to heterospecific singers carries a
436cost, then selection may push males to constrain the acoustic space that triggers a response in
437such a way that reduces the risk of responding to heterospecifics. This selection could then lead
438to reduced response to foreign conspecific dialects as a by-product (Hamao 2016). Dickcissels
439are the sole species in their genus, and sing a song that is, to our ears, quite distinct from the
440songs of other birds in the region. Thus it may be that dickcissels simply do not face much
441selection from mistaken response to heterospecific song, and thus have evolved to respond to
442song from a relatively broad range of acoustic space.
443
21 444Another non-exclusive alternative explanation is that males learn to recognize features of
445conspecific songs through interactions with conspecifics, and individuals who have encountered
446few dialects less readily recognize foreign dialects as representing a threat (Wright & Dorin
4472001). In many species, for instance those which do not migrate or in which individuals from
448different dialects migrate to different wintering grounds, individuals may have very little
449exposure to foreign songs and thus when they hear playback of a foreign song it may serve as a
450less effective stimulus of territorial aggression. In contrast, a species in which individuals
451frequently encounter singing conspecifics from many dialects during migration or winter may
452respond to a more diverse array of dialects when back defending a territory on the breeding
453grounds (Colbeck et al. 2010). Dickcissels appear likely to fall into this latter category. They
454aggregate in flocks of many thousands in winter and are reported to sing in these flocks
455(Temple 2002). However, it is important to point out that recognition of foreign dialects does
456not appear to require prior experience with the particular dialect in question even in species
457which show reduced aggression to foreign vs. local dialects. First, in those cases where
458response to playback of foreign conspecific dialects versus heterospecific song has been
459investigated, males typically show greater aggression to foreign conspecific song than to
460heterospecific song (Slabbekoorn & Smith 2002). Thus, there is some recognition of these
461foreign dialects, just a reduced response relative to the local dialect. Further, we know that in
462some species, naïve juveniles preferentially learn conspecific song over song of other species
463(Marler & Peters 1977) or even other sub-species (Nelson 2000). Thus, it seems that ability to
464recognize unfamiliar conspecific song may be widespread. So, if familiarity with foreign dialects
465increases the strength of response to foreign dialects, this might not be based on recognition
22 466per se but it could still result from some form of learning. For instance, individuals might learn
467to react more strongly if they have previously interacted with males singing a particular dialect.
468Or, individuals may not require experience with specific dialects, but exposure to various
469dialects may expand the multivariate acoustic space that induces aggressive responses, even if
470the particular dialect they are confronting is novel. However, in one of the few species where
471there is published evidence relevant to this hypothesis, migratory Puget Sound white-crowned
472sparrow (Zonotrichia leucophrys pugetensis) individuals not only recognize foreign song but also
473can temporarily borrow foreign phrases during migration (DeWolfe & Baptista 1995), and yet
474birds from this population still react more strongly to local than to foreign song (Nelson & Soha
4752004).
476
477Because dickcissels experience both territorial threats from males singing foreign songs and
478exposure to many foreign dialects in the non-breeding season, and because they likely face
479little selection from mistaken response to heterospecifics, we cannot rule out any of these
480potential explanations outlined above explaining why dickcissel males respond equally
481aggressively to foreign and local dialects while most other species respond more aggressively to
482local dialects. An effective comparison of these hypotheses could come from a cross-species
483meta-analysis of strength of response to local and foreign song playbacks combined with
484sufficient natural history information regarding the similarity of heterospecific songs, the
485tendency for males to disperse across dialect boundaries, and the tendency to hear a diversity
486of conspecific dialects.
487
23 488We have found evidence consistent with the existence of the dear-enemy effect in dickcissels,
489ruled out two adaptive hypotheses that might explain delayed song learning in this species, and
490we have raised important questions regarding the factors that lead male dickcissels to respond
491equally aggressively to both foreign and local dialect playback. Although meta-analysis could
492help answer questions regarding why some bird species show reduced response to foreign song
493versus local song, there remain major unanswered questions across songbirds and within
494dickcissels. Despite studies in a variety of species showing that subsets of songs can signal
495species identity and thus facilitate identification of conspecifics, the processes producing
496recognition and classification of song, and leading to differential responses to these different
497songs, remains poorly known across songbird species and within dickcissels.
498
499Acknowledgments
500Support for this project came from a Whitman College Perry Award and the Kansas State
501University Biology National Science Foundation (NSF) Research Experience for Undergraduates
502(REU) program (DBI-0552930 and DBI-1156571). Access to research sites was provided by Konza
503Prairie Biological Station. We would like to thank B. Sandercock and the members of his lab for
504their generous support of our dickcissel research over the past decade. B. Sandercock and B.
505Snyder both generously facilitated our participation in the REU program. Additional foreign
506songs were recorded by K. Ballinger, C. Foo, H. Kahl, E. Ross and P. Williams. S. Nakagawa
507provided assistance with the calculation of average effect sizes. This work was approved by the
508Whitman College Institutional Animal Care and Use Committee (protocol number: 042114wctp)
509and complies with the current laws of the United States.
24 510
511Literature Cited
512
513Baker, M. C. 1982: Vocal dialect recognition and propulation genetic consequences. Am. Zool. 22, 561-
514 569.
515-. 1994: Loss of function in territorial song: comparison of island and mainland populations of the singing
516 honeyeater (Meliphaga virescens). Auk 111, 178-184.
517Baker, M. C. & Cunningham, M. A. 1985: The biology of bird-song dialects. Behav Brain Sci 8, 85-133.
518Beecher, M. D. & Brenowitz, E. A. 2005: Functional aspects of song learning in songbirds. Trends Ecol.
519 Evol. 20, 143-149.
520Beecher, M. D., Campbell, S. E. & Nordby, J. C. 2000: Territory tenure in song sparrows is related to song
521 sharing with neighbours, but not to repertoire size. Anim. Behav. 59, 29-37.
522Bradley, D. W., Molles, L. E. & Waas, J. R. 2013: Local–foreign dialect discrimination and responses to
523 mixed-dialect duets in the North Island kōkako. Behav. Ecol. 24, 570-578.
524Briefer, E., Rybak, F. & Aubin, T. 2008: When to be a dear enemy: flexible acoustic relationships of
525 neighbouring skylarks, Alauda arvensis. Anim. Behav. 76, 1319.
526-. 2010: Are unfamiliar neighbours considered to be dear-enemies? PLoS ONE 5, e12428.
527Brooks, R. J. & Falls, J. B. 1975: Individual recognition by song in white-throated sparrows. I.
528 Discrimination of songs of neighbors and strangers. Can J Zool 53, 879-888.
529Colbeck, G. J., Sillett, T. S. & Webster, M. S. 2010: Asymmetric discrimination of geographical variation in
530 song in a migratory passerine. Anim. Behav. 80, 311-318.
531Danner, J. E., Danner, R. M., Bonier, F., Martin, P. R., Small, T. W. & Moore, I. T. 2011: Female, but not
532 male, tropical sparrows respond more strongly to the local song dialect: implications for
533 population divergence. Am Nat 178, 53-63.
25 534Davis, M. S. 1987: Acoustically mediated neighbor recognition in the North American bullfrog, Rana
535 catesbeiana. Behav. Ecol. Sociobiol. 21, 185-190.
536DeWolfe, B. B. & Baptista, L. F. 1995: Singing behavior, song types on their wintering grounds and the
537 question of leap-frog migration in Puget Sound white-crowned sparrows. Condor 97, 376-389.
538Godard, R. 1993: Tit for tat among neighboring hooded warblers. Behav. Ecol. Sociobiol. 33, 45-50.
539Greene, E., Muehter, V. R. & Davison, W. 2014: Lazuli Bunting (Passerina amoena). In: The Birds of North
540 America Online. (Poole, A., ed). Cornell Lab of Ornithology, Ithaca.
541Hamao, S. 2016: Asymmetric response to song dialects among bird populations: the effect of sympatric
542 related species. Anim. Behav. 119, 143-150.
543Hansen, P. 1984: Neighbor-stranger song discrimination in territorial yellowhammer Emberiza citrinella
544 males, and a comparison with responses to own and alien song dialects. Ornis Scandinavica 15,
545 240-247.
546Hyman, J. 2005: Seasonal variation in response to neighbors and strangers by a territorial songbird. Ethol
547 111, 951-961.
548Klicka, J., Burns, K. & Spellman, G. M. 2007: Defining a monophyletic Cardinalini: a molecular
549 perspective. Mol Phylogenet Evol 45, 1014-1032.
550Krebs, J. R. & Kroodsma, D. E. 1980: Repertoires and geographic variation in bird song. In: Advances in
551 the Study of Behavior, 11. (Rosenblatt, J. S., Hinde, R. A., Beer, C. & Busnel, M. C., eds). Academic
552 Press, New York. pp. 143-177.
553Kroodsma, D. E. 1989: Suggested experimental designs for song playbacks. Anim. Behav. 37, 600-609.
554Lemon, R. E. 1967: The response of cardinals to songs of different dialects. Anim. Behav. 15, 538-545.
555Marler, P. & Peters, S. 1977: Selective vocal learning in a sparrow. Science 198, 519-521.
556Marler, P. & Peters, S. 1982: Developmental overproduction and selective attrition: new processes in the
557 epigenesis of birdsong. Developmental Psychobiology 15, 369-378.
26 558McDonald, J. H. 2014: Handbook of Biological Statistics, 3rd edn. Sparky House Publishing, Baltimore,
559 Maryland.
560McGregor, P. K. 1983: The response of corn buntings to playback of dialects. Z. Tierpsychol. 62, 256-260.
561Nakagawa, S. & Cuthill, I. C. 2007: Effect size, confidence interval and statistical significance: a practical
562 guide for biologists. Biological Reviews 82, 591-605.
563Nakagawa, S. & Santos, E. A. 2012: Methodological issues and advances in biological meta-analysis. Evol
564 Ecol 26, 1253-1274.
565Nelson, D. & Poesel, A. 2013: Song sharing correlates with lifetime social pairing success but not territory
566 tenure in the Puget Sound white-crowned sparrow. Behavioral Ecology & Sociobiology 67, 993-
567 1000.
568Nelson, D. A. 2000: A preference for own-subspecies' song guides vocal learning in a song bird. Proc.
569 Natl. Acad. Sci. USA 97, 13348-13353.
570Nelson, D. A. & Marler, P. 1994: Selection-based learning in bird song development. Proc. Natl. Acad. Sci.
571 USA 91, 10498-10501.
572Nelson, D. A. & Soha, J. A. 2004: Perception of geographical variation in song by male Puget Sound
573 white-crowned sparrows, Zonotrichia leucophrys pugetensis. Anim. Behav. 68, 395-405.
574Nicholls, J. A. 2008: Site specificity in advertisement calls and responses to playbacks of local and foreign
575 call variants in satin bowerbirds, Ptilonorhynchus violaceus. Behav. Ecol. Sociobiol. 62, 831-841.
576Nowicki, S., Peters, S. & Podos, J. 1998: Song learning, early nutrition and sexual selection in songbirds.
577 Am. Zool. 38, 179-190.
578Osiejuk, T. S., Bielecka, A. & Skierczynski, M. 2012: What exactly is 'local song' in a population of ortolan
579 buntings with a common dialect? J. Ethol. 30, 133-142.
27 580Payne, R. B. 1981: Population structure and social behavior: models for testing the ecological
581 significance of song dialects in birds. In: Natural selection and social behavior: recent research
582 and new theory. (Alexander, R. D. & Tinkle, D. W., eds). Chiron Press, New York.
583-. 1983: The social context of song mimicry - song matching dialects in indigo buntings (Passerina
584 cyanea). Anim. Behav. 31, 788-805.
585Payne, R. B. 2006: Indigo Bunting (Passerina cyanea). In: The Birds of North America Online. (Poole, A.,
586 ed). Cornell Lab of Ornithology, Ithaca.
587Payne, R. B. & Payne, L. L. 1997: Field observations, experimental design, and the time and place of
588 learning in bird songs. In: Social influences on vocal development. (Snowdon, C. & Hausberger,
589 M., eds). Cambridge University Press, Cambridge, UK. pp. 57-84.
590Podos, J. 2007: Discrimination of geographical song variants by Darwin's finches. Anim. Behav. 73, 833-
591 844.
592Prum, R. O. 2010: The Lande-Kirkpatrck mechanism is the null model of evolution by intersexual
593 selection: implications for meaning, honesty, and design in intersexual signals. Evolution 64,
594 3085-3100.
595Rohwer, S. 1982: The evolution of reliable and unreliable badges of fighting ability. Am. Zool. 22, 531-
596 546.
597Rosenberg, M. S., Rothstein, H. R. & Gurevitch, J. 2013: Effect sizes: conventional choices and
598 calculations. In: Handbook of meta-analysis in ecology and evolution. (Koricheva, J., Gurevitch, J.
599 & Mengersen, K., eds). Princeton University Press, Princeton, USA.
600Schartz, R. L. & Zimmerman, J. L. 1971: The time and energy budget of the male dickcissel (Spiza
601 americana) Condor 73, 65-76.
602Schook, D. M., Collins, M. D., Jensen, W. E., Williams, P. J., Bader, N. E. & Parker, T. H. 2008: Geographic
603 patterns of song similarity in the dickcissel, Spiza americana. Auk 125, 953-964.
28 604Searcy, W. A. & Andersson, M. 1986: Sexual selection and the evolution of song. Annual Review of
605 Ecology and Systematics 17, 507-533.
606Searcy, W. A., Nowicki, S., Hughes, M. & Peters, S. 2002: Geographic song discrimination in relation to
607 dispersal distances in song sparrows. Am. Nat. 159, 221-230.
608Slabbekoorn, H. & Smith, T. B. 2002: Bird song, ecology and speciation. Philos. Trans. R. Soc. Lond. Ser.
609 B-Biol. Sci. 357, 493-503.
610Temple, S. A. 2002: Dickcissel (Spiza americana). In: The Birds of North America, No. 703. (Poole, A. &
611 Gill, F., eds). The Birds of North America, Inc., Philadelphia.
612Tomback, D. F., Thompson, D. B. & Myron Charles, B. 1983: Dialect discrimination by white-crowned
613 sparrows: reactions to near and distant dialects. Auk 100, 452-460.
614Turcokova, L., Pavel, V., Chutny, B., Petrusek, A. & Petruskova, T. 2011: Differential response of males of
615 a subarctic population of Bluethroat Luscinina svecica svecica to playbacks of their own and
616 foreign subspecies. J Ornith 152, 975-982.
617Wagner, R. H. 1998: Hidden leks: sexual selection and the clustering of avian territories. Ornithological
618 Monographs, 123-145.
619Wei, M., Lloyd, H. & Zhang, Y. 2011: Neighbour–stranger discrimination by yellow-bellied tit Parus
620 venustulus: evidence for the “dear-enemy” effect. J Ornith 152, 431-438.
621Wilson, P. L. & Vehrencamp, S. L. 2001: A test of the deceptive mimicry hypothesis in song-sharing song
622 sparrows. Anim. Behav. 62, 1197-1205.
623Wright, T. F. & Dorin, M. 2001: Pair duets in the yellow-naped Amazon (Psittaciformes: Amazona
624 auropalliata): responses to playbacks of different dialects. Ethol 107, 111-124.
625Zimmerman, J. L. & Finck, E. J. 1989: Philopatry and correlates of territorial fidelity in male dickcissels.
626 North American Bird Bander 14, 83-85.
627
29 628
629
30 630Table 1. Tests from generalized linear mixed models for the overall effects of time period (pre-
631playback, playback, post-playback) on each of the following three response variables, each in a
632separate analysis: number of songs, number of flights, and average distance from the speaker.
633Each analysis included territory identity as a random effect.
634
response variables F df df p # songs 6.19 5 189 0.002 # flights 26.40 2 189 <0.001 average distance 19.93 2 189 <0.001
random effect of territory Z p
(from tests of): # songs 2.11 0.035 # flights 1.92 0.056 average distance 2.97 0.003 635
636
31 637Table 2. Comparisons of mean measures, derived from generalized linear mixed models, of
638dickcissel response (number of songs, number of flights, and average distance from the
639speaker) among three time periods (before playback, during playback, after playback). We
640assigned positive values to Hedge’s d for cases of greater aggression during playback than
641before or after playback, or greater aggression after playback than before playback.
642
variable mean SD n mean SD n t df p Hedge’s variance
d of d before playback during playback # songs 10.477 11.200 64 7.263 7.912 64 2.2 189 0.03 0.330 0.032 # flights 1.365 1.800 64 4.621 5.120 64 3.2 189 <0.001 0.843 0.034 distance 6.064 1.704 64 4.900 1.704 64 5.5 189 <0.001 0.679 0.033 before playback after playback # songs 10.477 11.200 64 12.643 13.416 64 1.2 189 0.24 0.174 0.031 # flights 1.365 1.800 64 1.677 2.128 64 1.0 189 0.30 0.157 0.031 distance 6.064 1.704 64 4.924 1.704 64 5.4 189 <0.001 0.665 0.033 during playback after playback # songs 7.263 7.912 64 12.643 13.416 64 3.2 189 0.002 0.485 0.032 # flights 4.621 5.120 64 1.677 2.128 64 4.7 189 <0.001 0.746 0.033 distance 4.900 1.704 64 4.924 1.704 64 0.1 189 0.91 0.014 0.031 643
644
645
32 646Table 3. Tests with generalized linear mixed model analyses of the effects of treatment (foreign,
647local, and neighbor song) and trial number (first or second trial for each male) on number of
648flights during playback, and mean distance to the speaker (during and after playback). Each
649analysis included territory identity as a random effect.
X F d.f. p # of flights during playback overall model 5.18 3, 60 0.003 treatment 3.88 2, 60 0.026 trial # 8.65 1, 60 0.005 territory ID (random) Z = 0.77, p = 0.441 mean distance during playback overall model 4.36 3, 60 0.008 treatment 4.81 2, 60 0.012 trial # 3.86 1, 60 0.054 territory ID (random) Z = 2.02, p = 0.044 mean distance after playback overall model 2.84 3, 60 0.046 treatment 2.80 2, 60 0.069 trial # 3.21 1, 60 0.078 territory ID (random) Z = 1.00, p = 0.315 650 651
652
33 653Table 4. Comparisons of mean measures of dickcissel response (number of flights during
654playback, average distance from the speaker during playback, and average distance from the
655speaker after playback) to a first playback trial versus a second playback trial. We assigned
656positive values to Hedge’s d for cases of greater aggression during the first trial.
657
first second variable mean SD N mean SD N t df p Hedge’s variance
d of d # flights 5.909 4.435 32 3.415 2.834 32 2.8 60 0.006 0.662 0.066 distance during 4.642 1.375 32 5.169 1.375 32 2.0 60 0.054 0.379 0.064 distance after 4.588 1.680 32 5.268 1.680 32 1.8 60 0.08 0.400 0.064 658
659
660
34 661Table 5. Comparisons of mean measures of dickcissel response (number of flights during
662playback, average distance from the speaker during playback, and average distance from the
663speaker after playback) to playback of foreign song versus non-neighbor local song. We
664assigned positive values to Hedge’s d for cases of greater aggression to foreign song.
665
foreign local variable mean SD N mean SD N t df p Hedge’s variance
d of d # flights 5.701 4.278 22 5.218 3.991 21 0.4 60 0.69 0.115 0.093 distance during 4.550 1.332 22 4.625 1.329 21 0.2 60 0.83 0.055 0.093 distance after 4.643 1.670 22 4.551 1.668 21 0.2 60 0.85 -0.054 0.093 666
667
35 668Table 6. Comparisons of mean measures of dickcissel response (number of flights during
669playback, average distance from the speaker during playback, and average distance from the
670speaker after playback) to playback of neighbor song versus playback of either foreign (F) or
671non-neighbor local (L) song.
672
foreign or local neighbor variable mean SD N mean SD N t df p Hedge’s variance
d of d # flights (F) 5.701 4.278 22 3.046 2.584 21 2.5 60 0.01 0.733 0.099 # flights (L) 5.218 3.991 21 3.046 2.584 21 2.1 60 0.04 0.634 0.100 distance during (F) 4.550 1.332 22 5.541 1.329 21 2.8 60 0.007 0.723 0.097 distance during (L) 4.625 1.329 21 5.541 1.329 21 2.6 60 0.01 0.676 0.101 distance after (F) 4.643 1.670 22 5.591 1.668 21 2.0 60 0.054 0.558 0.097 distance after (L) 4.551 1.668 21 5.591 1.668 21 2.1 60 0.04 0.612 0.100 673
674
36 675Figure legends
676Figure 1. Examples of playback songs used in a single neighborhood. Note the high note
677conformity within phrases but the variability in phrase number and phrase spacing between the
678unmanipulated ‘neighbor’ songs, as is common in dickcissels. Note also the close match in
679notes between the neighbor and local treatments and the radical differences when compared
680to the foreign songs.
681
682Figure 2. Forest plot of average standardized differences (Hedge’s d, ± 95% confidence limit)
683between mean dickcissel responses (number of songs [in first comparison only], number of
684flights, closest approach distance) before versus during playback, during first versus second
685trials, during and after playback of foreign versus local song, and during and after playback of
686neighbor versus either foreign or local song. Hedge’s d values between 0.5 and 0.8 are
687considered moderate to large. Thus, dickcissels responded strongly to playback trials, though
688more strongly in first than in second trials. Further, there was no detectable difference in
689response to playback of foreign versus local songs, though both foreign and local songs elicited
690much more vigorous responses than did playback of neighbor song.
691
692
37 693Fig 1
694
38 695
696Fig 2
697
39