Evolution and disappearance of sympatric Coregonus albula in a changing environment—A case study of the only remaining population pair in Sweden

Abstract During the past 50 years, Fennoscandian populations of spring‐spawning Baltic cisco (Coregonus albula), sympatric to common autumn‐spawners, have declined or disappeared; for example, three out of four known spring‐spawning populations in Sweden are regarded as extinct. Over the same period, the climate has changed and populations have been subject to other anthropogenic stressors. We compared historic (1960s) and recent (1990–2000s) morphological data from the still‐existent sympatric cisco populations in Lake Fegen, Sweden. Phenotypic changes were found for spring‐spawners making them more similar to the sympatric autumn‐spawners that had remained virtually unchanged. Based on results for other salmoniform fishes, a phenotypically plastic response to increased temperature during early development appears unlikely. The recent material was also analyzed with microsatellite markers; long‐term effective population size in spring‐spawners was estimated to be about 20 times lower than autumn‐spawners, with signs of long‐term gene flow in both directions and a recent genetic bottleneck in spring‐spawners. We suggest the change toward a less distinct phenotype in spring‐spawners to reflect a recent increase in gene flow from autumn‐spawners. Time since divergence was estimated to only c. 1,900 years (95% CI: 400–5,900), but still the Fegen populations represent the most morphologically and genetically distinct sympatric populations studied. Consequently, we hypothesize that less distinct population pairs can be even younger and that spring‐spawning may have repeatedly evolved and disappeared in several lakes since the end of the last glaciation, concurrent with changed environmental conditions.

includes the "Medieval Warm Period" (900-1350 AD) and the "Little Ice Age" (1550-1850 AD) (Niedźwiedź et al., 2015). Within a contemporary timeframe, the increase in temperatures recorded since the mid-20th century (IPPC, 2014) marks the beginning of a gradually warmer period.
Numerous studies have dealt with anthropogenic driven climate change and its recorded or expected consequences on biodiversity, such as range expansions and extinction of species or local populations (McLean, Lawson, Leech, & Pol, 2016;Pecl et al., 2017). In addition, anthropogenic influences on the environment have been shown to induce phenotypic changes at higher rates compared with what is expected due to "natural" processes (Hendry, Farruiga, & Kinnison, 2008).
So far, the most detailed studies of sympatric ciscoes have been carried out in Lake Stechlin, Germany; in addition to physiological adaptations Ohlberger, Staaks, Petzoldt, Mehner, & Hölker, 2008), only slight morphological differences between these coexisting cisco forms could be determined, tentatively representing adaptations to different microhabitats (Helland, Vøllestad, Freyhof, & Mehner, 2009). Hence, besides marked differences in spawning time, it is intriguing that sympatric cisco populations lack the more apparent ecological niche separations typically seen in coexisting whitefishes and chars (e.g., piscivorous, benthivorous, and planktivorous morphs) often manifested by striking differences in body size and morphology related to the feeding apparatus (Markevich, Esin, & Anisimova, 2018;Snorrason et al., 1994;Svärdson, 1979). Delling et al. (2014) studied Baltic cisco populations, including the four known Swedish cases with sympatric populations. The spring-spawning form is extinct in three of these lakes, but archived scale samples made analyses of mtDNA variation possible. Both spring-and autumn-spawners from all four lakes and additional lakes (with only autumn-spawners) in south-central Sweden were fixed for a few closely related haplotypes not found elsewhere in Sweden.
Furthermore, these haplotypes seem more closely related to those in North American Coregonus sardinella. Contemporary microsatellite data for Swedish populations were also congruent with the distribution of mtDNA haplotypes, showing a striking dichotomy between samples from south-central (higher altitude) lakes compared with other parts of Sweden, regardless of spawning time (Delling et al., 2014).
The single extant population of spring-spawning cisco in Sweden ( Figure 1) inhabits Lake Fegen ( Figure 2). Together with the extinct populations from Lakes Ören, Stora Hålsjön, and Åsunden, it was formally described as Coregonus trybomi by Svärdson (1979). The species description was mainly based on spawning time, but it was noted that the Fegen spring-spawners exhibited a particularly distinct morphology. Most striking was a proportionally larger eye compared with other spring-spawners and the common autumnspawning form in Fegen and other lakes ( Figure 1). A larger eye has F I G U R E 1 Sympatric spring-spawning (top) NRM 53995 (127 mm Sl) and autumn-spawning (bottom) NRM 54000 (131 mm Sl) Baltic cisco, C. albula from L. Fegen. The spring-spawning population is characterized by (on average) larger eyes and larger heads also been reported for one winter-spawning population in Finland (Airaksinen, 1968), whereas the spring-spawning populations in Ören and Åsunden (no data exist from St Hålsjön) showed no difference in eye size compared with their sympatric autumn-spawners (Svärdson, 1979). Differences in vertebral counts have also been described, again most strongly expressed in Fegen, with springspawners having on average 3.3 fewer vertebrae compared with the sympatric autumn-spawners (Svärdson, 1979).

The extinction of spring-spawners in Ören, Åsunden, and St
Hålsjön in recent decades  have been explained by extrinsic factors such as pollutants, eutrophication, and introduction of non-native fish species (Fjälling, 1988;Svärdson, 1979).
The present status of winter-and spring-spawning populations in Finland is largely unknown, but there are indications that they are declining or have disappeared (M. Himberg, Åbo Akademi, personal communication).
The rapid disappearance of three out of four known springspawning populations in Sweden and the similar situation in Finland has raised concerns and questions regarding the status of the still surviving sympatric population pair in Lake Fegen. Also this lake has been subject to anthropogenic impacts, for example, in terms of repeated introductions of predatory pikeperch, Sander lucioperca, since the 1940s, and eutrophication (Thörne & Carlsson, 2004). Survey gill net fishing in 2014 suggested a decline in the overall abundance of C. albula compared with 2003, and based on morphologic data obtained from a subsample (mainly from the NE basin), only some 2%-4% of the 2014 catch were classified as spring-spawners (B. Delling, unpublished). There are also some records (e.g., Lessmark, 1976) that the Fegen spring-spawners were once more widely distributed compared with the present distribution which is limited to the NE basin (Figure 2), indicating a decline in abundance over time (Thörne & Carlsson, 2004). However, the spring-spawning population has always been considered rare, and there is not sufficient monitoring data to establish a proportionally stronger decline compared with the sympatric autumn-spawners.
In this study, we investigated recent phenotypic changes in the sympatric Fegen populations in relation to possible effects of environmental factors. Specifically, we compared historic (1960s) and more recent (1990-2000s) morphological data from Fegen to changes in local temperatures over the same period, discussing a possible eco-phenotypic response directly or indirectly related to temperature by means of plasticity and (or) microevolution.
Microsatellites markers were used to study genetic structure and amounts of gene flow. We further estimated effective population sizes, and searched for signs of genetic bottlenecks, to investigate whether elevated gene flow could potentially explain observed phenotypic changes. The time since divergence between the two Fegen populations was also estimated. By combining data from detailed morphological and genetic analyses with results from previous studies of other sympatric Baltic cisco populations, an overall aim was to gain further understandings of the evolution, maintenance, and possible reasons for recent collapses of these two-population systems.

| MATERIAL AND ME THODS
Lake Fegen (23.5 km 2 ) is situated 132 m above the sea level in Southern Sweden (Figure 2). The average water depth is 7.5 m with a maximum of 38 m (NE basin). In addition to Baltic cisco, perch, Perca fluviatilis, and roach, Rutilus rutilus, represent the most abundant indigenous fish species (Thörne & Carlsson, 2004). These authors listed a total of 19 fish species, including adjacent watercourses emptying in the lake (see Appendix 1). Fegen has a history of fish introductions, including that of pikeperch which led to an established population in the late 1990s (see Appendix 1).
More detailed ecological data on the Fegen cisco populations in relation to, for example, niche segregation, are scanty. Stomach contents in April consisted almost solely of copepods in both populations (Lessmark, 1976). All ciscoes from Fegen show comparatively slow growth and small adult size, for example, compared F I G U R E 2 Lake Fegen (23.5 km 2 , 132 m above sea level, max depth 36-38 m) with surrounding waters in southern Sweden. Analyzed air temperatures were taken from the nearest weather station "SMHI Fagered" (marked with an asterisk). Spring-spawning C. albula are mainly found in the deepest NE basin, whereas autumn-spawners exist in all parts of the lake * Weather staƟon km 0 2.5 5

NEbasin
with ciscoes in the adjacent Lake Kalvsjön. In addition, the springspawners show slower growth than the sympatric autumn-spawners (Lessmark, 1976). Comparably slow growth has also been noted for bream, Abramis brama, a species that usually thrive under rich conditions, which has been interpreted as a sign of relatively poor nutrient status of the lake (Thörne & Carlsson, 2004).

| Samples
Primary material consisted of 376 Baltic ciscoes from Fegen collected during the years 1995-2008 (Table 1). Part of this material (n = 149) was included in the genetic study by Delling et al. (2014). For comparisons, we also used historical morphological data for 50 springspawners (SS) and 100 autumn-spawners (AS) from Fegen collected in 1960-1969(Svärdson, 1979. Additional specimens and data from other lakes were used to address certain methodological issues related to comparisons of historical and more recent morphological data (Table A3). All historical data, earlier published only as means and standard deviations by Svärdson (1979), were reanalyzed based on archived individual measurements from original protocols recovered at the Institute of Freshwater Research, Drottningholm.
The historical material was collected during spawning from "ripe and running" (rr) individuals that could be strictly classified as AS or SS (Svärdson, 1979). The primary recent material includes AS and SS collected while spawning (SSrr and ASrr), but also specimens found to be ripe but not running (SSr and ASr), and some fish with unknown or uncertain gonadal status that were immature and/or collected outside the spawning seasons (Table 1). As detailed below, depending on the analysis or question, different subsets of the recent material were assigned to SS or AS based on gonadal status (rr and r) and time of capture, or from DNA (microsatellites) and morphology with the suffix mg (morphometry and genetics).

| Morphological analyses
Based on Svärdson (1979), the main morphological differences between AS and SS in Fegen are vertebral counts (Vc), eye diameter (Ed), and head length (Hl), with SS possessing fewer vertebrae, a proportionally larger head and larger eye.
Both total length (Tl) and standard length (Sl) were measured (whereas only Tl existed for the historical materials). Sex for mature specimens (rr or r) was determined in the field or at the laboratory shortly after capture. For the remaining unclassed specimens, sex was determined during final morphological analyses.
The majority of recent specimens were analyzed in a frozen Measuring the size of the eye (or the eye cavity in a weakly ossified fresh or fixed small coregonid) is notoriously tricky, and repeated trials often result in deviations. However, using X-ray, the sclera of the eyeball in the anterior and posterior margin of the eye is clearly visible, and the horizontal eye diameter can be measured with a high degree of accuracy, using a reference length for calibration ( Figure 3). The number of vertebrae was also counted from digital X-ray images, starting with the Atlas vertebrae behind the head and ending with the three upturned vertebrae in the caudal skeleton (here counted as three separate vertebrae; Figure 3). In case of occasionally observed vertebrae fusions, the actual number of vertebrae was estimated from the number of neural spins or ribs.
No documentation seems to exist on how morphological measurements and counts were obtained for the historical materials, and no such specimens from Fegen have been preserved. Thus, only original protocols with individual data have been available for the purpose of this study. Besides the variables measured for the recent specimens, the historical materials also included data on gill raker and scale counts, snout length, body depth, snout to dorsal origin distance, and snout to pelvic origin distance (Svärdson, 1979).
TA B L E 1 Material of C. albula from L. Fegen analyzed with microsatellites and morphology, divided on "ripe and running" (rr) or ripe (r) spring-(SS) and autumn-(AS) spawners Note: "Uncertain" are fish without sex determination ("?") or with gonads with uncertain status. Material from 1995 and 2003 lack Tl. The 10 SSrr from 1995 listed as females (f, determined in the field) were not possible to sex determine after preservation.
During the course of the study, it became obvious that historical and recent morphological data were not fully comparable regarding Vc and Hl, which most likely reflects systematic methodological differences. Due to a lack of historical preserved material from Fegen, we measured and counted these variables for preserved historical material from Ören and compared those results with published data from the same populations and time period (Svärdson, 1979). As detailed in Appendix 1 (Table A4 with associated text), this allowed for adjustments so that proper comparisons between recent and historic data were possible. In brief, we added 1.36 vertebrae per fish and reduced head lengths by c. 1% (in proportion to Tl) in the historic data. The relation between Vc and Hl was also investigated to confirm that the larger head in SS was not the result of a proportionally shorter body due to fewer vertebrae ( Figure A1).
Variation in Hl, Ed, and Sl were subjected to a principal component analysis (PCA) using the software SYSTAT13. Comparing the resulting principal components to standard lengths, showed that the second component (PC II) could be treated as a size-independent morphometric "shape" variable ( Figure A2; Table A5). The morphometric PC II was used in combination with a corresponding "genetic component" based on factorial correspondence analysis on microsatellite genotypes (details below), to ordinate the total recent material. This resulted in two virtually nonoverlapping clusters, which could be divided into two groups by addition of a (somewhat arbitrarily placed) straight line ( Figure 4).
We compared historic data on eye diameter (Ed) from Svärdson (1979) with recently collected specimens classified as rr or r. Potential temporal changes in relative Ed within SS and AS, respectively, were tested for with ANCOVAs, using Hl as covariate to account for allometric growth. Since the historical specimens were larger on average, the recent material was restricted to specimens with Hl > 25.5 mm to allow for comparison of similarly sized fish (after exclusion of small recent specimens, no significant difference in Hl between historical and recent specimens remained within AS and SS; t test: p = .44 for SS, p = .62 for AS).
To study the degree of morphological distinction among historic SSrr and ASrr, all studied characters from the original protocols (Svärdson, 1979) were also subjected to PCA ( Figure A3; Tables   A6 and A7). This independent analysis was done to investigate whether adding further morphological characters was sufficient to unambiguously distinguish (i.e., without overlap) the two forms in the historic material.

| Molecular analyses
DNA was extracted from fin-clips preserved in 95% EtOH, and all individuals were genotyped for the same set of nine microsatellites as in Delling et al. (2014). We refer to that paper for details on markers and laboratory procedures (DNA extraction, PCRs).
A factorial correspondence analysis (FCA) was performed with Genetix 4.05.2 (Belkhir, Borsa, Chikhi, Raufaste, & Bonhomme, 2004) to visualize relative similarity among individual multilocus genotypes. Fstat 2.9.3.2 (Goudet, 1995) was used to compute unbiased estimates of expected heterozygosity, allelic richness, and F-statistics, and to evaluate deviations from Hardy-Weinberg proportions and genotypic equilibrium. structure 2.3.4 (Falush, Stephens, & Pritchard, 2003;Pritchard, Stephens, & Donnelly, 2000) was employed to identify the most likely number of genetic clusters (K) in the total material. K was increased from 1 to 5, with true K inferred following Evanno, Regnaut, and Goudet (2005) as implemented in structure Harvester (Earl & vonHoldt, 2012). structure was run without prior information on sampling or phenotypic information, assuming admixture and correlated allele frequencies between clusters. As recommended by Gilbert et al. (2012), we used 20 replicate runs per K with a burn-in of 100,000 steps followed by 200,000 MCMC replicates.
To identify putative non-neutral outlier loci that can yield biased parameter estimates in certain analyses, we used LOSITAN (Antao, Lopes, Lopes, Beja-Pereira, & Luikart, 2008) that implements the F ST simulation approach by Beaumont and Nichols (1996). Signs of genetic bottlenecks where tested for with Bottleneck 1.2.02 (Piry, Luikart, & Cornuet, 1999), applying the two-phase mutation model (Cornuet & Luikart, 1996) with settings recommended for microsatellites (i.e., 95% single-step and 5% multistep mutations and a variance of 12 for multisteps mutations; Piry et al. (1999)). Results were evaluated with a one-tailed Wilcoxon test for heterozygote excess (i.e., if present expected heterozygosity was higher than the one expected at mutation-drift equilibrium) using 10,000 replicates. Using the procedure implemented in Bottleneck, we also evaluated the presence of allele frequency mode shifts, as expected following a reduction in effective population size (Luikart & Cornuet, 1998). To search for indications of bottlenecks, we further computed a modification of Garza and Williamson's (2001) M-ratio as implemented in arlequin 3.5 (Excoffier, Laval, & Schneider, 2005).

Contemporary gene flow and migration between sympatric
spring-and autumn-spawning populations were assessed with Bayesass 3.0 (Wilson & Rannala, 2003). Each simulation was run for 10 × 10 6 iterations sampled every 2000 step, with the first 3 × 10 6 iterations omitted as burn-in. As recommended by Meirmans (2014), we used several (5) independent runs with different random seeds and calculated Bayesian deviance (Faubet, Waples, & Gaggiotti, 2007) to evaluate differences in convergence among runs.
Historical demographic parameters were assessed under an "Isolation with migration model" (Nielsen & Wakeley, 2001) using an ABC-approach (Approximate Bayesian Computation) with PoPaBc (Lopes, Balding, & Beaumont, 2009). Under this model, two subpopulations, diverged from a common ancestral population sometime in the past, may be connected by gene flow in both directions. Prior distributions for demographic parameters are listed in Table A8. Similar to in Lopes and Boessenkool (2010), mutations simulated for the microsatellites followed a stepwise model with rates drawn from a lognormal distribution of base 10 (prior) with a mean (−4) sampled from a normal distribution (hyper-prior) and a standard deviation set to a fixed value (0.5). A total of 10 × 10 6 data sets were simulated with PoPaBc. Out of these, 1,000 were retained (0.0001 rejection level) based on their closeness to the observed data at a set of 17 "summary statistics" (heterozygosity, variance in allele length, number of different alleles, kurtosis of allele's lengths, Shannon's diversity index, heterozygosity-based Nm estimator) computed within subpopulations and for the total material. Posterior parameter estimates with associated probability densities were assessed using nonlinear regression (neural network method adjusted for heteroscedasticity in R-package "abc"; Csilléry, François, & Blum, 2012).

| RE SULTS
As detailed below, ciscoes in Fegen fall into two forms according to (a) spawning time, (b) morphology, and (c) nuclear genetic markers, but none of these independent data sets can alone be used to divide all fish into well-separated groups. Combining morphologic and genetic data, however, significantly enforced the distinction between the two forms ( Figure 4), and since information on spawning time was missing for parts of the material, the below results are mainly based on this joint morphologic-genetic assignment (ASmg, n = 231, and SSmg, n = 145). Notably, only three out of 248 specimens sampled as ripe and running (rr) or ripe (r) deviated from its expected group, according to the division based on independent morphologicgenetic data ( Figure 4).

| Morphology
A comparison of recent and historical vertebral counts confirmed that SS have significantly fewer vertebrae than AS, with just minor year-to-year variation within each group ( Figure 5). In contrast, no such differences among recent rr and mg individuals within SS and AS, respectively, could be detected (not shown). Comparing vertebral counts from several lakes showed that SS often have lower counts compared with sympatric AS, but that this difference was most prominent in Fegen ( Figure 6).

| Molecular data
The analysis with structure based only on microsatellite genotypes yielded highest likelihood for two genetic clusters (K = 2; Figure   A7). No statistically significant deviations from expected Hardy-Weinberg proportions or pairwise genotypic equilibrium occurred within SSmg and ASmg, whereas a significant heterozygote deficiency across loci existed in the total material ( Table 2).
The average number of alleles observed, allelic richness, and expected heterozygosity was higher in ASmg. Among 122 alleles in total, 55 were unique to one of the groups (50 and 5 private alleles in ASmg and SSmg, respectively). Significant allele frequency differences among SSmg and ASmg occurred at all but one locus and in total, with an overall F ST of 0.07 (Table 2). In contrast, no allele frequency differences could be detected between ASmg individuals collected from the three main lake basins (not shown).
One locus (BWF1) was identified as a putative outlier, displaying higher differentiation between SSmg and ASmg than expected under selective neutrality (p = .013; Table A9, Figures A8 and A9). Hence, several of the analyses below (assuming neutrality) were based on eight loci only, to avoid potential bias.
Signs of a genetic bottleneck were seen for spring-spawners.
The Garza-Williamson M-ratio (Table 2) was consistently lower in SSmg than in ASmg (paired t test: p = .001). Observed gene diversity (H E ) in both populations was found to be lower than the levels expected under mutation-drift equilibrium (H EQ ), that is, a pattern in the opposite direction as anticipated for populations following a genetic bottleneck where H EQ is expected to be reduced faster than H E (Cornuet & Luikart, 1996) The five runs with Bayesass (8 loci) yielded almost identical parameter estimates and Bayesian deviances, and all chains were therefore merged before final analysis using tracer v. 1.6 (Rambaut, Drummond, Xie, Baele, & Suchard, 2018). Contemporary levels of gene flow into ASmg and SSmg were estimated to less than 0.01 per generation with a point modal estimate for immigration into AS (from SS) being twice as high as in the opposite direction (0.008 vs. 0.004) although with highly overlapping probability intervals (Table 3).   Table 3), assuming a generation interval of 4 years (Delling et al., 2014).

| D ISCUSS I ON
Our results reveal rapid phenotypic change in the spring-spawning (SS) Fegen population, manifested as an increased number of vertebrae and a decreased relative eye size from the 1960s to the 1990s.
In contrast, the sympatric autumn-spawning (AS) population has remained virtually unchanged. Analyses of microsatellite data show lower allelic richness and signs of a genetic bottleneck in the SS population. The same genetic data further reveal a clear difference TA B L E 2 Genetic variation in autumn-(n = 231) and spring-spawning (n = 145) C. albula from Lake Fegen (cf. Table 1 As shown in Figure 4, there is not a perfect match between the division of the material into two groups based on spawning time and combined morphologic-genetic data, respectively. Similar incongruences between morphologically defined groups and spawning time were also revealed in the historic data ( Figure A3; Tables A6 and A7).
Consequently, even if we use vernacular names stipulating spawning period for the two sympatric forms, we conclude that some exceptions are expected between assignments of individuals based on different types of data. Below we discuss different possible explanations for the rapid morphological change seen in Fegen, followed by comments on evolution and disappearance of sympatric Baltic cisco populations in general.

| Eco-phenotypic response to temperature?
Although fluctuations between years in average air tempera- the data on brown trout, Salmo trutta, vertebral counts (Vc) in relation to temperature (Tåning, 1952), are probably most relevant for comparison with our results (i.e., a salmoniform fish with similar temperature preferences and vertebral counts as C. albula). Tånig's results showed that it required an increase of as much as 5°C to increase the average Vc in trout by one. Lowest counts were found around 6°C, and a decrease in temperature also resulted in a slight increase of vertebral counts in six out of eight crosses, that is, a so-called V-shaped response (Lindsey, 1988

| Microevolutionary change?
The alternative to effects of temperature on phenotypic traits in Fegen SS is microevolution involving natural selection, random genetic drift, and/or increased gene flow from AS into SS. Leary, Allendorf, and Knudsen (1985)  For three-spined stickleback, Gasterosteus aculeatus, evolutionary rapid phenotypic changes explained by altered natural selection regimes have been documented (e.g., Marques et al., 2016). Further, Bhat et al. (2014) showed that invasion of Baltic cisco into a typical Fennoscandian two-population system of whitefish Coregonus sp.
(planktivorous and benthivorous) led to breakdown of reproductive isolation between the two whitefish forms that collapsed into a more morphologically and genetically homogenous population within just 15 years. Thus, some kind of microevolutionary explanation to the present observations appears feasible.
However, in Fegen, a potential scenario involving natural selection raises several questions. It is unclear to what extent the observed phenotypic differences between SS and AS are in fact adaptive and related to ecological niche separation. There is a tendency with lower vertebrae counts in SS for all studied population pairs ( Figure 6). A plausible explanation for this general pattern may involve competition between juvenile AS and SS, where allocation of growth in SS from body segments (vertebrae) into a larger head and feeding apparatus might give an advantage for the offspring when competing with AS that hatches somewhat earlier in spring.
According to Figure 2 in Airaksinen (1968), it seems that SS in Lake Ännättijärvi had larger heads than their sympatric AS, even though not clearly stated by the author. Among the four known population pairs in Sweden, however, a larger head in SS is only found in Fegen, whereas SS in Ören and Åsunden (no data for St Hålsjön) on average displayed c. 1% smaller heads in relation to Tl, compared with their sympatric AS (Svärdson, 1979). Eronen and Lahti (1988) studied the life cycle of the "winterspawning" (March) cisco in the Finnish Lake Kajoonjärvi. They suggested that low fecundity combined with large and protein-rich eggs could be adaptions related to later hatching and competition with sympatric AS, although they only compared data for AS from other lakes. Unfortunately, no comparable data on fecundity or egg size exist for the Fegen populations.
Larger eyes are commonly found in fishes in deeper water, although this pattern could partly reflect differences in growth rates (Pankhurst & Montgomery, 1994). Even if most spring-spawning cisco populations have been reported to spawn deeper than their sympatric autumn-spawners, (e.g., Svärdson, 1979) little is known regarding their spatial separation outside the spawning seasons. We also note that SS have been reported to have significantly larger eyes than their sympatric AS (cf. head length above) in only two cases (Fegen and Ännättijärvi) (Airaksinen, 1968;Svärdson, 1979).
A negative relationship between relative eye size and individual growth rate has been reported for O. mykiss (Pankhurst & Montgomery, 1994). However, the pronounced eye size difference in Fegen (Figures 1 and 7) is coupled with just a slightly lower growth rate in SS (Lessmark, 1976). This may suggest that genetic factors are The present microsatellite data indicates a "genetic bottleneck" in SS ( Figure 9, Table A10). Elevated genetic drift during such a decrease in effective population size is expected to affect the whole genome, which may create random phenotypic change (Falconer & Mackay, 1996). However, to disentangle the relative effect of drift and selection on phenotypic trait means requires experiments in controlled environments (e.g., Rogell et al., 2013).
Furthermore, it has not been possible to date the present bottleneck event and to check whether it occurred between the collecting dates of the historical and recent data sets or earlier, although the lake history and some results obtained here indirectly support a more recent bottleneck. From the 1950 to 1960s and onwards, Fegen has suffered from acidification, slight eutrophication, and introduction of pikeperch (Thörne & Carlsson, 2004).
Results from gill-netting surveys also indicate a decline in overall cisco abundance in more recent years (Linderfalk, 2014). Signs of retarded growth (Figures 7 and 8)  Another not exclusive explanation for rapid phenotypic change could be a recent and temporary influx of genes from AS into SS.
Our results indicate low and rather similar levels of long-term and contemporary gene flow (immigration rates) in both directions, despite a larger than 20-fold difference in long-term average effective population sizes (Table 3). Converting these estimates into corresponding numbers of effective migrants results in about 50 times more migrants from the small SS into the larger AS population than vice versa. Although this estimate is statistically uncertain, it may reflect that autumn-spawning is the ancestral condition (Delling et al., 2014) and that spring-spawners more commonly revert into "normal" autumn-spawning. At the same time, we note that among the 376 specimens included there are three individuals, all autumn-spawners according to morphology and microsatellites, that seemingly spawned at the "wrong time" (Figure 4), possibly indicating a higher current gene flow from AS into SS.
Pointing toward a recent boost in gene flow from AS to SS is also the fact that an earlier allozyme study (30 loci) including SS and AS from Fegen (Öst, Jansson, & Hamrin, 1990) (Table 1).
In conclusion, out of the possible microevolutionary explanations for rapid phenotypic change listed above, we consider a tem-

| Patterns and processes in Baltic cisco evolution
Within a broader time frame, environmental change is probably a key factor behind the evolution, maintenance, and collapses of ciscoes with displaced spawning periods that in several cases has resulted in sympatric populations. Trybom (1903) speculated that the temperature regime in Stora Hålsjön could explain the occurrence of sympatric ciscoes. Svärdson (1979) further argued that autumn-spawning is an adaptation among "cold water fish species" to maximize their first growth season, whereas spring-spawning (seen among "warm-water fishes") is generally regarded as more beneficial under warmer conditions. This suggestion was in line with his hypothesis of an ancient preglacial origin of the proposed species C. trybomi ("spring-spawning cisco") during the warmer Eemian interglacial period. However, supported by genetic and geological data, Delling et al. (2014) rather proposed a later and independent postglacial diversification in early-arriving ciscoes (the so-called Group I) to higher altitude lakes with past or present sympatric populations, utilizing ice-dammed lake complexes above the Baltic Ice Lake. Vuorinen et al. (1981) also found genetic evidence for independent postglacial diversification for cisco populations and further showed that Finnish lakes with spring-or winter-spawners displayed higher water temperatures during winter, compared to other lakes with only autumn-spawners. These Finnish (and Swedish) lakes are comparatively small, deep and situated at higher altitudes, which results in an earlier development of a permanent ice cover following the autumn circulation. The ice cover, in turn, insulates the lake from continued cooling during wind-induced circulation (Eklund, 1998(Eklund, , 1999. Higher water temperature during winter was suggested by Vuorinen et al. (1981) to be unfavorable for autumn-spawners, through elevated egg mortality caused by oxygen deficiency. In line with this hypothesis, a milder climate could be beneficial for autumnspawning, especially in smaller high altitude lakes, which indirectly (via intraspecific competition) may be unfavorable for a sympatric spring-or winter-spawning population.
We have estimated the divergence time between SS and AS in Fegen to a few 1,000 years only, but still the Fegen ciscoes seem to be the most morphologically and genetically differentiated sympatric population pair studied so far. The populations in Lake Ören showed very little morphological differentiation (Svärdson, 1979) and only slight genetic differentiation at 32 allozyme loci (Svärdson, 1988;Vuorinen, 1988). In contrast, the allozyme study including SS and AS from Fegen (Öst et al., 1990) and our present data have revealed clear differentiation between AS and SS (see above). In German Lake Stechlin, only subtle morphological differences were found among the sympatric ciscoes and F ST estimated across six microsatellites was just 0.004 (Schulz et al., 2006).
Assuming that levels of genetic and morphological differentiation are roughly related to time since divergence, the population pairs in Ören and Stechlin may have evolved much more recently than the more distinct ones in Fegen. Alternatively, the low levels of morphological and genetic divergence could reflect that the "speciation process" has already reversed in those lakes. Regardless, it is obvious that levels of differentiation vary considerably between lakes.
In line with an evident independent postglacial divergence, this is also what one could expect; that is, to a certain degree all sympatric population pairs represent individual unique cases.
Comparison of genetic differences between population pairs and estimates of time since divergence using methods and data sets that are not fully comparable (see above) should be done with caution, and so far few such cases in C. albula have been studied in detail.
However, we suggest that the evolution of sympatric Baltic cisco populations may be a fluctuating process; that is, spring-spawners may "come and go" as environmental conditions change. More recent and rapid anthropogenic impacts, including slight regional climate change, may have had an overall negative impact on these fragile systems, and it is possible that the declining SS population in Fegen has persisted a little longer than in other lakes as it is comparatively old and morphologically distinct, indicating a possible stronger niche differentiation toward AS.
Even though we suggest that these sympatric cisco populations could appear almost ephemeral in evolutionary terms, the extinction and decline of SS populations in recent decades represent a rapid loss of biodiversity within the far from fully understood C. albula complex. Further studies of ciscoes from additional Fennoscandian lakes, that is, in Finland, possibly still inhabited by sympatric C. albula populations, appear urgently needed to gain further knowledge on these interesting populations.

ACK N OWLED G M ENTS
We thank Anna Eklund (SMHI) and Rasmus Linderfalk (County

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
B.D. and S.P. designed the study. B.D performed the morphological work, whereas S.P. analyzed molecular data. B.D. and S.P. performed joint analyses of morphologic-genetic data and drafted the manuscript.

. FI S H FAU N A I N L A K E FEG E N
The information below comes from the extensive data compilation by Thörne and Carlsson (2004) with some additional input from Linderfalk (2014).

A D D ITI O N A L S PECI E S N OT I N CLU D E D I N TA B LE A 2 A B OV E
White bream, Blicca bjoerkna: Two specimens reported in 1967, most certainly misidentification (Linderfalk, 2014).
Brown trout, Salmo trutta: A few specimens recorded during electrofishing in inflowing watercourses.
Minnow, Phoxinus phoxinus: Recorded during electrofishing in inflowing watercourses, possibly also present on suitable habitats along shore.
European eel, Anguilla anguilla: This species is never caught in gill nets, but its regular presence in the lake has been confirmed (see also below).

H I S TO RY O F S TO CK I N G A N D I NTRO D U C TI O N S
Brook trout, Salvelinus fontinalis: Introduced in a few watercourses 1963-1969 and regarded as, at least temporarily, established at one locality in 1967. 1967-1989 and 1995-2014  Note: Relative abundances are given as percentages of total number or total weight. For the period 1967-1989, very low abundances are noted only as "X". Note that test fishing methodologies have changed over time, especially between 1989 and 1995. However, on the whole, data are consistent within the two time periods (1967-1989 and 1995-2014), justifying comparisons within (but not between) these periods.

TA B L E A 2 Species composition from survey gill net fisheries in periods
Rainbow trout, Oncorhynchus mykiss: Stocked in 1965 (100 specimens only), not established.
Whitefish: Stocked (1942) but probably already present in the lake. The whitefish in Fegen is a typical sparsely rakered form (18-21 gill rakers), most certainly benthos-feeding.
Pike: Intensive stocking  like in many other Swedish lakes during that period. This kind of stocking of young pike (0+ or 1+) have later been considered as largely inefficient.
European eel: Stocking of eel in Swedish lakes, including Fegen, have a long tradition to compensate for the loss of natural migrating elvers.
Pikeperch: Repeatedly stocked since 1945 but not confirmed as established until in late 1990s.

. CO M PA R I S O N O F M E TH O DS TO O BTA I N M O R PH O LO G I C A L DATA
Unfortunately, the historical Fegen material was not preserved. In Svärdson (1979) or in the original protocols, there are no details regarding methods on how morphological measurements and counts where obtained, but likely specimens were examined in fresh condition with vertebrae counted by means of dissection. Similar handwriting on all original protocols from the SLU archive (Table A3) indicates that the same staff took all measurements.
Comparisons between historic records and results from recent analyses of preserved material from Lake Ören revealed systematic differences with respect to number of vertebrae (Vc), head length (Hl), and eye diameter (Ed): • Historical Vc is lower compared with recent data. Possible explanations include (a) vertebrae are more easily counted using X-ray (as for recent material), (b) uncertainty over whether the Atlas vertebra was included, and (c) uncertainty regarding how the three last "upturned" vertebrae in the caudal skeleton ( Figure 3) were counted earlier. Note: All data from "SLU archive" only consist of individual records from original protocols. a Mean values with standard deviations (Runnström, 1941). b Incomplete collecting data but listed as possible PARATYPES in the NRM collection. c PARATYPES collected during spawning from the same spawning shoal as the HOLOTYPE for Coregonus trybomi.
• Historical Hl is slightly longer compared to recent data. A possible explanation could be that it was measured from the most anterior tip of head (tip of lower jaw, cf. Tl measurements) to most posterior margin of gill cover versus tip of snout to most posterior margin of the operculum. The operculum is the uppermost dermal bone out of three making up the functional gill cover.
• Historical Ed is slightly smaller compared to recent data. Possible reasons are mentioned below.

Comparisons of historical and contemporary values (based on preserved historical samples) resulted in slight adjustments of Vc (into Vca) and
Hl (into HLa). They also elucidated problems comparing different data sets. Details on comparisons are as follows:

Vertebral counts
Vertebral counts (Vc) were obtained by means of X-ray from a total of 47 preserved C. albula from Lake Ören (Table A3, NRM 16231-16234, 39286, 39287). Out of these, 11 were originally classified as SS whereas status (SS or AS) of the remaining 36 is uncertain. Counts were compared with the historical Ören data set (202 SS and 53 AS). The difference in Vc between historic AS and SS in Lake Ören is not statistically significant (t test) whereas a comparison of all the 47 X-rayed specimens with the 255 historical ones yielded a difference +1.36 vertebrae (i.e., higher average for the recently analyzed material). Consequently, 1.36 were added to individual vertebrae counts in the historical Fegen material, indicated as Vca (adjusted vertebral count). Note that this adjustment is only used in graphic presentations and discussion on trends of changes in Vc over time, that is, the correction is not included in any combined dataset that has been analyzed statistically.

Head length and eye diameter
In total, 36 C. albula (SS/AS) from Lake Ören collected in 1958 were analyzed in the same way as the contemporary material. Measurements included Tl, Hl, Hls (assumed head length, sensu Svärdson (1979), eye diameter measured with a digital caliper (Edd) and from X-ray (Edx). Data were compared with historic data for 100 specimens (SSrr) collected from Ören in 1957.
Measuring the head sensu Svärdson (1979)  Measuring the eye from radiographs gives results comparable to mechanical measurements but with a somewhat smaller standard deviation (Table A4). When comparing historical and contemporary values for Ed, there is a weak tendency for larger eyes in the contemporary material irrespective of method. This can represent a true difference, but also a difference in measurement methods. Therefore, we chose not to adjust Ed in the historical dataset. Silfvergrip (1996) used the term "segmental effect" to describe a situation where body length and other morphometric characters are correlated to the number of body segments, that is, vertebrae. Hypothetically, a high number of vertebrae should result in a longer body and consequently a proportionally shorter head.

R E L ATI O N B E T WE E N V E RTE B R A L CO U NT S A N D H E A D LE N G TH I N C . ALB UL A
We checked for a possible segmental effect in recent ASmg and SSmg males and females separately, applying the method described below (exemplified with data from SSmg males). The vertebral column runs through the whole body and from radiographs it was estimated that about four vertebrae lie behind the operculum, that is, within the range of Hl.  Figure A1 and elucidate hypothetical relationships between Hl and Vc, assuming a 100% segmental effect (when vertebrae are added or excluded from the average specimens). The graphic presentation ( Figure A1)  Note: Standard deviation (±) is given for contemporary eye measurements (Hls = tip of lower jaw to posterior margin of gill cover, Edx = eye diameter obtained from X-ray).
stronger tendency for a segmental effect in SSmg (close to significant in males) than in ASmg. Further, the difference in proportional Hl between SS and AS is influenced by, but far from fully explained by, a segmental effect. This is most easily seen when comparing SS and AS having 56-58 vertebrae, a range represented in both forms.

. P C A FO R R ECE NT M ATE R I A L FRO M L A K E FEG E N
Adjustments for size in a morphologically heterogeneous sample are often not fully doable applying a pooled among-group regression as it will generate size-dependent residuals (Bookstein et al., 1985). In a principal component analysis, PC I is often referred to as "size," whereas PC II and further components are less correlated to size and therefore referred to as "shape" (Bookstein et al., 1985).
In the recent data set, growth of head in relation to standard length and growth of eye in relation to head length, respectively, appear isometric ( Figures 7 and A6 below). In Figure A2a, there is strong correlation between PC I and size (Sl) as expected (see above), whereas Figure   A2b reveals no obvious correlation between PC I and PC II, the former explaining about 86% of the total variation (Table A5). The selection of ripe and running specimens only, divided into spring-and autumn-spawners, confirms size independence for PC II also for these subsets of recent specimens ( Figure A2c).  TA B L E A 7 Character loadings on principal components I-V for six meristic characters from the historical data set of C. albula from Lake Fegen (Fegen SSrr and ASrr in Table A1)

I S O L ATI O N WITH M I G R ATI O N M O D E L : PA R A M E TE R S A N D PR I O R S
F I G U R E A 4 Graphical depiction of individual Baltic cisco genotypes from Fegen following factorial correspondence analysis with GENETIX (9 microsatellites, n = 376). SSrr (n = 89) in blue, ASrr (n = 63) in orange. Note that scores for Factor 1 are also used in Figure 4 (together with corresponding morphological information)

. D I FFE R E N CE I N VC B E T WE E N SS A N D A S I N H I S TO R I C A L A N D R ECE NT M ATE R I A L
Irrespective of the different methods for counting vertebrae (historic vs. recent; see above), the absolute difference in average numbers between the two groups AS and SS in L. Fegen has decreased over time; from about 3.3 vertebrae in the 1950-1960s to about 1.5 in the

1990-2000s.
To test if the difference in Vc between SS and AS has changed significantly over time, the historic and contemporary data sets were randomly resampled 10,000 times. As shown by nonoverlapping 95% confidence intervals ( Figure A5), the reduced difference in Vc between SS and AS over time is highly significant.
F I G U R E A 5 Difference in average vertebrae counts between autumn and spring-spawning individuals in Fegen with 95% confidence intervals (based on 10,000 resamplings)

. M O R PH O M E TRY O F R ECE NT A N D H I S TO R I C A L M ATE R I A L FRO M L A K E FEG E N
F I G U R E A 6 Eye diameter plotted against head length (upper panels) and head length plotted against total length (lower panels) for historical (n = 145) and recent (n = 376) materials of C. albula from Fegen, divided by sex (AS in orange, SS in blue). Linear regression lines with 95% confidence bands are shown for each group separately. The dashed black lines shows hypothetical isometric growths based on average eye diameter, head length, and total length (black dots). Head lengths in the historical material have been slightly adjusted (0.938 × Hl = Hla) to allow direct comparisons (see "Comparison of methods to obtain morphological data"). Note that four of the graphs are also shown in Figure 7 (Ed against Hl for rr historic and mg recent) Ed against Hl (mm) females top, males below

. G E N E TI C B OT TLE N ECK S
TA B L E A 9 Result from outlier analysis with lositan. Prob. is the proportion of simulated F ST -estimates smaller than the observed one TA B L E A 1 0 Tests with Bottleneck for deviations from expected heterozygosity at mutation-drift equilibrium assuming two-phase (TPM) and stepwise (SMM) models of mutation (Cornuet & Luikart, 1996)