1 Phenotypic plasticity is maintained despite geographical isolation in an African cichlid fish,
Accepted Article
Pseudocrenilabrus multicolor
Kirsten E. Wiens1, Erika Crispo2, and Lauren J. Chapman3
1
Department of Pathology, New York University School of Medicine, 550 First Avenue, New York, NY, 10016, Phone: (646)385-9734, Email: [email protected]
2
Department of Biology and Health Sciences, Pace University, One Pace Plaza, New York, NY, 10038, Phone: (212)346-1504, Email: [email protected]
3
Department of Biology, McGill University, 1205 Ave. Dr. Penfield, Montréal, QC, Canada, H3A 1B1, Phone: (514)398-8199, Fax: (514)398-5069, Email: [email protected]
3
Wildlife Conservation Society, 2300 Southern Blvd., Bronx, NY, 10460, USA
Correspondence: [email protected]
Running title: Plasticity in an isolated population
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/1749-4877.12029
2 Abstract Gene flow among populations in different selective environments should favor the evolution of
Accepted Article
phenotypic plasticity over local adaptation. Plasticity in development is a common response to long-term hypoxia in some widespread African fishes including Pseudocrenilabrus multicolor, a cichlid that exploits both normoxic (high oxygen) rivers/lakes and hypoxic (low oxygen) swamps. Previous studies have shown that fish from normoxic and hypoxic sites differ in many traits including gill size, brain size, and body shape, and that much of this variation reflects developmental plasticity. However, these earlier studies focused on areas in Uganda where gene flow between swamp and river or lake populations is high. In this study we tested the hypothesis that P. multicolor from a relatively isolated lake population (Lake Saka, Uganda) exhibit low
levels of plasticity in traits related to oxygen uptake. Multiple broods of P. multicolor from Lake
Saka were reared under low and high dissolved oxygen (DO), and traits related to gill size, brain mass, and body shape were quantified. Surprisingly, both gill size and brain mass showed high levels of developmental plasticity. We suggest that high levels of plasticity, particularly in the gill size of P. multicolor, may reflect low costs of maintaining the plastic response even in relatively isolated populations.
Keywords: hypoxia, riverine fish, gills, brain, geometric morphometrics
3 INTRODUCTION Phenotypic variation among populations may be explained by phenotypic plasticity, genetic
Accepted Article
variation, or a combination of both. It is predicted that high gene flow among populations should favor the evolution or maintenance of phenotypic plasticity over local adaptation (Sultan & Spencer 2002; Crispo 2008). Plasticity might not be adaptive to an individual, but instead plasticity might evolve as a by-product of selection on the most extreme phenotypes. The most extreme phenotypes would be those expressed by the most plastic individuals (or their offspring – i.e. the most plastic “genotypes”) after dispersal into a new environment (Via 1993; Via et al., 1995). Alternatively, plasticity could itself be adaptive if it provides a means for populations to track environmental change. Thus, in the absence of gene flow, individuals that are able to specialize or locally adapt to the environment should have a selective advantage. However, plasticity may be costly due to the sensory mechanisms necessary to regulate responses to environmental cues and the risk of misinterpretation (DeWitt et al. 1998). Further, if the benefits
of plasticity are small or rare, plasticity may be lost from a population due to genetic drift (Masel et al., 2007). Therefore the factors that affect the plasticity of a trait are complex, and the general objectives of this and previous studies have been to determine what drives the evolution of plasticity, and if this changes given the environmental context. Developmental plasticity is a common response to long-term hypoxia in some
widespread African fishes including Pseudocrenilabrus multicolor victoriae (Seegers 1990), a
cichlid that exploits both normoxic (high oxygen) rivers/lakes and hypoxic (low oxygen) swamps. Previous studies in a river-swamp system in western Uganda showed that fish from normoxic (river) and hypoxic (swamp) sites differ in many traits related to oxygen uptake and metabolic costs (e.g., gill size, brain size, Chapman et al. 2000; Chapman et al. 2008) and that
4 much of this variation reflects developmental plasticity (Chapman et al. 2000; Chapman et al. 2008; Crispo & Chapman 2010). In common garden rearing experiments, P. multicolor
Accepted Article
developed larger gills under low oxygen conditions, presumably because larger gills increase the surface area for oxygen uptake. Brains had a lower mass under low oxygen, presumably due to the metabolic cost associated with this organ (Chapman & Hulen 2001; Poulson 2001; Safi et al. 2005). Additionally, body shape was affected largely by changes in gill size associated with hypoxia exposure (Crispo & Chapman 2011). A characteristic feature of this system is that gene flow may have been possible between high and low oxygen environments, at least within the recent past. Therefore, high levels of plasticity might have evolved in the meta-population as a result of selection on “extreme” phenotypes (Sultan & Spencer 2002; Crispo 2008). The goal of the present study was to examine phenotypic plasticity of gill, brain, and
body shape of P. multicolor from Lake Saka, a crater lake in western Uganda, East Africa. Lake
Saka is unique in that it is geographically isolated from the lakes and rivers sampled in previous studies of P. multicolor. In addition, P. multicolor in this region show low neutral genetic
variability, based on microsatellites, compared to those in surrounding sites (Crispo & Chapman 2008). This could be due to a small founding population in Lake Saka and/or bottlenecks associated with the introduction of predatory Nile perch, coupled with low levels of gene flow from other populations. P. multicolor is found in ecotonal vegetated areas of this lake (L.
Chapman, unpubl. data). Lake Saka experiences hyper-eutrophication due to extensive clearing of surrounding forest and wetland. Oxygen availability remains high throughout the year (63 to 160% saturation, Binning & Chapman, 2009), relative to nearby papyrus swamp systems (Crispo & Chapman, 2008).
5 We predict that phenotypic plasticity of traits related to oxygen uptake and use, such as gill and brain size, will be low because (1) local adaptation is not constrained by gene flow from
Accepted Article
surrounding populations (Crispo & Chapman, 2008), and (2) dissolved oxygen conditions remain high throughout the year; and therefore, oxygen should not be limiting in ecotonal regions of the lake (Binning & Chapman, 2009). As an alternative, if plasticity is not costly and is an ancestral trait in cichlids, we might observe higher levels of plasticity. If this is the case, body morphology may also vary as an indirect response to environmental change, in order to accommodate changes in gill and brain size. Yet, the evolution of reduced plasticity might still occur due to drift (Masel et al. 2007).
MATERIALS AND METHODS
Laboratory experiment P. multicolor were collected from Lake Saka, using baited minnow traps, in June 2009 and transported live to McGill University, Montreal, Quebec. See Crispo & Chapman (2008) for a description of the collection site and dissolved oxygen measures. Binning & Chapman (2009) provide monthly mean values for dissolved oxygen concentration (DO) and water temperature in Lake Saka over a 14-month period (May 2007-June 2008) in an ecotonal area. Over that time, DO averaged 6.3±0.4 mg/L, (SE of monthly values, range = 4.6 to 9.2 mg/L; ~63 - 109% O2
saturation) in the morning and 10.6±0.4 ng/L in the afternoon (range 9.0 to 13.4 mg/L; ~109 160% O2 saturation). Water temperature reported by Binning & Chapman (2009) varied little over the 14-month period, typical of this equatorial region, averaging 22.3±0.3 oC in the morning
(range = 20.2 to 23.8 oC ) and 24.0±0.3 oC in the afternoon (range = 22.1 to 25.4 oC ).
6 We carried out a laboratory-rearing study to examine the extent of morphological plasticity in the Lake Saka P. multicolor population in response to high versus low DO
Accepted Article
environments. The experiments took place at McGill University after shipping the collected adults from Uganda. The experiment was conducted following previously described protocols (Crispo & Chapman 2010). Specifically, six full-sib families were raised from the Lake Saka collection site, where a family consisted of the family of one male-female pair. F1 individuals from each family were split between high and low DO treatments one week after release from the mouth (P. multicolor are mouth-brooders). Families were placed in separate 53-liter tanks,
for a total of six low DO tanks and six high DO tanks. Tanks were filtrated using Fluval underwater filters (Hagen). Low DO conditions were maintained with a dissolved oxygen control system (Point Four systems Inc., Coquitlam, British Columbia), and high DO conditions were maintained by bubbling air through the water column. In the low DO treatment, the mean DO concentration was 1.0 mg/L, and tank means ranged from 0.96-1.1 mg/L. In the high DO treatment, the mean DO concentration was 7.3 mg/L, and tank means ranged from 6.8-7.6 mg/L. The mean ambient temperature was 25.0°C, and tank means ranged from 24.4-25.2°C. At an age of one month, families were culled to 10 fish per tank. If fewer than 10 fish were present in a tank, we did not perform culling. The McGill University Animal Care Committee approved the protocol for the project. At an age of approximately one year, we harvested the two largest males from each
family for analysis of gill metrics, brain mass, and external body morphology. Initially this included 24 males (12 from each oxygen treatment). However, we removed one male (high DO treatment) with a mass well below the range of the other fish to prevent the possibility of skewed results due to an allometric growth relationship. We chose only the two largest males because the
7 development of subordinate fish was suppressed due to dominance hierarchy; and sex has been shown to influence gill shape in female P. multicolor (O’Connor et al. 2012). Previous studies
Accepted Article
on plasticity in this system also examined only the two largest males per tank (Crispo & Chapman 2008; Chapman et al. 2000, 2008). All remaining males and females were then
harvested to be included in external body morphology analyses (totals of 43 females and 27 males from low DO; and 33 females and 19 males from high DO). Fish designated as “females” for the body shape analysis also included juvenile males that had not yet developed secondary sex characteristics. All fish were euthanized using buffered tricaine methanesulfonate (MS-222; pH 7.0) and were preserved in 4% paraformaldehyde (buffered with phosphate buffered saline; pH = 7.0).
Gill parameters We separated the four gill arches (hemibranches) of the left branchial basket by dissection. The left and right sides of each gill arch were laid flat on a microscope slide and photographed using a Lumenera Scientific Infinity camera attached to a dissecting microscope. We measured five gill metrics using ImageJ 1.43u (Fig. 1), including total gill filament length (TGFL, in mm), average gill filament length (AFL, in mm), total number of gill filaments (TNF), total hemibranch area (THA, in mm2), and total perimeter of hemibranchs (TP, in mm). These measures are representative of gill size, and show potentially different ways that gill size could shift in response to environmental conditions. The length of every fifth filament was measured on both sides of each hemibranch. To increase resolution, every second filament was measured for the first five and the last five filaments on each hemibranch (Fig. 1). Filament lengths were measured along the length of the filament from the tip to the base, using segmented lines for
8 curved filaments. If filaments were missing or not clearly visible, measurements were estimated using lengths of surrounding filaments. Variables were measured on each of the four gills, and
Accepted Article
multiplied by two for the total number for each fish (i.e., because only one side of the head was dissected per fish) following procedures used in earlier studies (e.g., Crispo & Chapman 2010). We performed analyses of covariance (ANCOVA) on each gill metric separately to see if
the DO regimes influenced individual gill characteristics, while controlling for variation among families and variation due to body mass. We were not able to perform a multivariate analysis (MANCOVA) because the random factor (family) could not be included in this type of test. All analyses were performed using Type III sums of squares in SPSS version 17.0. Tests included log10-transformed gill metrics as response variables, log10-transformed body mass as a covariate,
treatment (high or low DO) as a fixed factor, family as a random factor, and the interaction between family and treatment. The treatment-by-body mass interaction was non-significant (results not shown) and therefore not included in the model. Using a two-tailed t-test we found that fish bred under high DO had a greater mass than fish bred under low DO (p = 0.045; mean mass under high DO was 0.47 g and under low DO was 0.33 g). This may be expected as fish under oxygen stress may allocate fewer resources to growth. However, there was good overlap in the size range of the two groups. We calculated overall difference between treatments as the percent difference between oxygen treatments for each gill metric, using anti-logged means adjusted to a common body mass (across families and treatments; least-square means obtained from the ANCOVA). Percent difference was calculated as a function of the mean value under high-oxygen: 100*(average under low DO – average under high DO)/(average under high DO). Additionally, we measured the total gill filament length (as above) for the second gill arch (i.e. second from the operculum) for 10 male fish caught in the wild. These wild-caught fish
9 were collected in June 2006 during a previous expedition, euthanized using buffered tricaine methanesulfonate (MS-222; pH 7.0), and preserved in 4% paraformaldehyde (buffered with
Accepted Article
phosphate buffered saline; pH = 7.0). We were not able to analyze the brain mass or body morphology of these fish because the samples had been destroyed during the dissection of the gills. We performed an ANCOVA for log10-transformed TGFL, with treatment (high DO, low DO, wild) included as a fixed factor and log10-transformed body mass as a covariate. The treatment-by- body mass interaction was not significant and thus not included. This analysis was performed to examine how the gill size of lab-raised fish compared to those caught in the wild.
Brain mass To measure relative brain mass we extracted whole brain from the same laboratory-raised fish used for gill analysis. Standard dissection methods were used following Chapman et al. (2008).
Brains were preserved in 4% buffered paraformaldehyde. We obtained wet mass by blotting each brain on a Kimwipe to removed excess paraformaldehyde, and measuring each to the nearest 0.1 mg using an analytical balance (Mettler AC 100). We measured each brain once every day for five consecutive days, and used the average of the five measures for analysis. Brains weighed less on each consecutive day, which was likely due to handling and the fragility of the brains. To assure that the effect was consistent across all brains, we performed a Pearson’s Correlation test and found that measures of brain mass taken on each day strongly correlated with each other (0.9 < r < 1; p < 0.001 for all comparisons). The ANCOVAs and percent difference calculations for brain mass relationships were carried out in the same way as for the gill metrics, with the same factors and covariate.
10 Body morphology We analyzed morphology for all fish harvested over the course of the experiment. Photos and
Accepted Article
linear measures were taken prior to dissection. We included the males used for the gill and brain analyses, along with the females, additional males, and juveniles present in each tank. After preservation in 4% buffered paraformaldehyde, we measured fish weight to the nearest 0.1 mg, and head width over (HWOO) and behind the operculum (HWBO) to the nearest 0.1 mm using digital callipers (Mitutoyo, Model CD-6”B). We included HWBO because it may show variation in head width independent of gill size. We then placed each fish on a laminated grid mounted on a level structure and captured an image of its left side using a digital camera (Canon PowerShot G9). Total body length, from the tip of the snout to the hypural plate (in mm), was later measured with these photographs using ImageJ version 1.43u. We digitized 11 landmarks and 4 semilandmarks using tspDig version 2, following the procedures in Crispo & Chapman (2011) (Fig. 2). We then performed geometric morphometric analysis using tspRelw version 1.45. The program uses the Generalized Procrustes Method for superimposition. This analysis yielded partial, uniform, and relative warp scores and centroid sizes that were used in further analysis. Partial warps scores are vectors representing localized variation in landmark positions. Uniform warps are a type of partial warps that represent parallel changes in landmark positions. Relative warps are independent linear combinations of partial warp i.e. principal components of the shape coordinates; we therefore refer to the relative warps as principal components (PC). Centroid size (covariate) is a measure of the total of the distances from each landmark to the center-point of all the landmarks. We analyzed principal components (relative warps) that each explained more than 5% of the variation in the data, and present results from only those that showed significant differences between treatments.
11 We performed ANCOVAs to test for effects of treatment and family on head shape, while controlling for variation due to body length. Log10-transformed HWOO and HWBO were
Accepted Article
included as response variables in two separate analyses. In both analyses, treatment (high or low DO) was included as a fixed factor, family as a random factor, and log10-transformed total body length as a covariate. Sex effects were not significant (results not shown) and thus not included. The family-by-treatment interactions were also included. We calculated overall difference between treatments (relative to head size under high DO, as for the gills) as the percent difference between DO treatments for each head measure, using anti-logged means adjusted to a common body length. For the geometric morphometrics analysis, we conducted a multivariate analysis of
covariance (MANCOVA) to test for effects on overall body shape. For this analysis, we were able to include all partial and uniform warps as response variables. A drawback of multivariate models is that random factors cannot be included, so the effect of family is excluded in this analysis. We included treatment as a fixed factor and centroid size as a covariate. Sex effects were not significant (results not shown) and thus not included. The treatment-by-centroid size interaction was also removed because it was not significant (result not shown). To complement the above analysis, we used ANCOVAs to test for family and treatment
effects on specific components of body shape (principal components; PC), while controlling for variation among families and variation due to centroid size. We included PC 1 and 2 as the response variables because each explained more than 5% of the variation and showed significant – or approaching significant – differences between treatments. We included treatment as a fixed factor, family as a random factor, and centroid size as a covariate. Sex effects were not significant (results not shown) and thus not included. For PC 1 we included the treatment-by-
12 centroid size interaction, and removed all other non-significant interactions. We did not include the family-by-centroid size interaction. For PC 2 we initially included interactions with centroid
Accepted Article
size, but subsequently removed them when non-significant (results not shown). We also analyzed PC’s 3, 4 and 5, because each also explained more than five percent of the variation, but do not present the results because treatment effects were not significant.
Meta-analysis In addition to the overall goal of characterizing the plasticity of an isolated population of P. multicolor, we were interested in comparing morphology and phenotypic plasticity between the Saka population and other P. multicolor populations from the Mpanga River drainage, the latter
of which were already characterized in a previous study (Crispo & Chapman 2010). Different experimenters measured the gills for Lake Saka (K. Wiens) and the Mpanga populations (E. Crispo), which could bias the results. Thus, one experimenter (E. Crispo) re-measured the TGFL for the second hemibranch of the Lake Saka fish. The proportional difference was calculated for each hemibrach (re-measure – original measure/average of the two measures), and then an average proportional difference was obtained by averaging these 44 values (22 fish, two hemibranchs per gill arch). This proportional difference was used to standardize the original labraised Lake Saka values to the measures used in the previous study (Crispo & Chapman 2010). We conducted two ANCOVAs, one for TGFL and one for brain mass. Treatment and
population were included as fixed factor, family was included as a random factor, and log10transformed body mass was included as a covariate. In both cases, response variables (TGFL or brain mass) were log10-transformed. We initially included all possible interactions among
population, treatment, and body mass, but sequentially removed non-significant interactions.
13
RESULTS
Accepted Article
Gills In the univariate tests, treatment effects were highly significant for all gill metrics with the exception of total number of filaments (TNF) (Table 1). All gill metrics increased with body mass and, with the exception of TNF, were larger under low DO (when controlling for variation in body mass; Fig.3). The overall difference between treatments ranged from a low of 4.2% for TNF (not significant) to a high of 32.6% for total hemibranch area (THA) (Table 2). Body mass, family, and family-by-treatment effects were not significant (Table 1). In the analysis including the wild-caught fish, the treatment term was significant (F2,30 =
41.686, P < 0.001), and log10-transformed body mass was a significant covariate (F1,30 = 70.593,
P < 0.001). As expected, TGFL increased with body mass. Controlling for body mass, TGFL for wild-caught fish did not differ from that for fish raised under high DO, while TGFL for fish raised under low DO was larger than wild-caught and high DO-raised fish (Fig. 4). These results are expected based on the assumption that the high DO treatment represents oxygen conditions similar to those experienced in the wild.
Brains Treatment effects on brain mass were also significant (Table 1). Brains were smaller under low DO when controlling for variation in body mass. The overall difference between treatments was 15.2% (Table 2). Although brain mass tended to increase with body mass, this effect was not significant (Fig. 3). Additionally, family and family-by-treatment effects were not significant (Table 1).
14
Body morphology
Accepted Article
There was a significant effect of treatment on head width over the operculum (HWOO), with wider heads in the fish raised under low DO when controlling for variation in body length. However, there was no effect of treatment on head width behind the operculum (HWBO), suggesting that the variation in HWOO was due to widening of the gills under low DO. Both measures increased with body length (Fig. 3), but there was no effect of family on either measure. The family-by-treatment effects were significant for both HWOO and HWBO, meaning that the level of plasticity (i.e. treatment effects) varied among families – this could reflect genetic variation and/or tank effects (Table 1). Sex effects were not significant and were thus not included in the model. The multivariate analysis of partial and uniform warps revealed strong effects of
treatment and centroid size on body shape (Treatment F26,94 = 3.675, P < 0.001; Centroid size
F26,94 = 7.346, P
Accepted Article
Pseudocrenilabrus multicolor
Kirsten E. Wiens1, Erika Crispo2, and Lauren J. Chapman3
1
Department of Pathology, New York University School of Medicine, 550 First Avenue, New York, NY, 10016, Phone: (646)385-9734, Email: [email protected]
2
Department of Biology and Health Sciences, Pace University, One Pace Plaza, New York, NY, 10038, Phone: (212)346-1504, Email: [email protected]
3
Department of Biology, McGill University, 1205 Ave. Dr. Penfield, Montréal, QC, Canada, H3A 1B1, Phone: (514)398-8199, Fax: (514)398-5069, Email: [email protected]
3
Wildlife Conservation Society, 2300 Southern Blvd., Bronx, NY, 10460, USA
Correspondence: [email protected]
Running title: Plasticity in an isolated population
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/1749-4877.12029
2 Abstract Gene flow among populations in different selective environments should favor the evolution of
Accepted Article
phenotypic plasticity over local adaptation. Plasticity in development is a common response to long-term hypoxia in some widespread African fishes including Pseudocrenilabrus multicolor, a cichlid that exploits both normoxic (high oxygen) rivers/lakes and hypoxic (low oxygen) swamps. Previous studies have shown that fish from normoxic and hypoxic sites differ in many traits including gill size, brain size, and body shape, and that much of this variation reflects developmental plasticity. However, these earlier studies focused on areas in Uganda where gene flow between swamp and river or lake populations is high. In this study we tested the hypothesis that P. multicolor from a relatively isolated lake population (Lake Saka, Uganda) exhibit low
levels of plasticity in traits related to oxygen uptake. Multiple broods of P. multicolor from Lake
Saka were reared under low and high dissolved oxygen (DO), and traits related to gill size, brain mass, and body shape were quantified. Surprisingly, both gill size and brain mass showed high levels of developmental plasticity. We suggest that high levels of plasticity, particularly in the gill size of P. multicolor, may reflect low costs of maintaining the plastic response even in relatively isolated populations.
Keywords: hypoxia, riverine fish, gills, brain, geometric morphometrics
3 INTRODUCTION Phenotypic variation among populations may be explained by phenotypic plasticity, genetic
Accepted Article
variation, or a combination of both. It is predicted that high gene flow among populations should favor the evolution or maintenance of phenotypic plasticity over local adaptation (Sultan & Spencer 2002; Crispo 2008). Plasticity might not be adaptive to an individual, but instead plasticity might evolve as a by-product of selection on the most extreme phenotypes. The most extreme phenotypes would be those expressed by the most plastic individuals (or their offspring – i.e. the most plastic “genotypes”) after dispersal into a new environment (Via 1993; Via et al., 1995). Alternatively, plasticity could itself be adaptive if it provides a means for populations to track environmental change. Thus, in the absence of gene flow, individuals that are able to specialize or locally adapt to the environment should have a selective advantage. However, plasticity may be costly due to the sensory mechanisms necessary to regulate responses to environmental cues and the risk of misinterpretation (DeWitt et al. 1998). Further, if the benefits
of plasticity are small or rare, plasticity may be lost from a population due to genetic drift (Masel et al., 2007). Therefore the factors that affect the plasticity of a trait are complex, and the general objectives of this and previous studies have been to determine what drives the evolution of plasticity, and if this changes given the environmental context. Developmental plasticity is a common response to long-term hypoxia in some
widespread African fishes including Pseudocrenilabrus multicolor victoriae (Seegers 1990), a
cichlid that exploits both normoxic (high oxygen) rivers/lakes and hypoxic (low oxygen) swamps. Previous studies in a river-swamp system in western Uganda showed that fish from normoxic (river) and hypoxic (swamp) sites differ in many traits related to oxygen uptake and metabolic costs (e.g., gill size, brain size, Chapman et al. 2000; Chapman et al. 2008) and that
4 much of this variation reflects developmental plasticity (Chapman et al. 2000; Chapman et al. 2008; Crispo & Chapman 2010). In common garden rearing experiments, P. multicolor
Accepted Article
developed larger gills under low oxygen conditions, presumably because larger gills increase the surface area for oxygen uptake. Brains had a lower mass under low oxygen, presumably due to the metabolic cost associated with this organ (Chapman & Hulen 2001; Poulson 2001; Safi et al. 2005). Additionally, body shape was affected largely by changes in gill size associated with hypoxia exposure (Crispo & Chapman 2011). A characteristic feature of this system is that gene flow may have been possible between high and low oxygen environments, at least within the recent past. Therefore, high levels of plasticity might have evolved in the meta-population as a result of selection on “extreme” phenotypes (Sultan & Spencer 2002; Crispo 2008). The goal of the present study was to examine phenotypic plasticity of gill, brain, and
body shape of P. multicolor from Lake Saka, a crater lake in western Uganda, East Africa. Lake
Saka is unique in that it is geographically isolated from the lakes and rivers sampled in previous studies of P. multicolor. In addition, P. multicolor in this region show low neutral genetic
variability, based on microsatellites, compared to those in surrounding sites (Crispo & Chapman 2008). This could be due to a small founding population in Lake Saka and/or bottlenecks associated with the introduction of predatory Nile perch, coupled with low levels of gene flow from other populations. P. multicolor is found in ecotonal vegetated areas of this lake (L.
Chapman, unpubl. data). Lake Saka experiences hyper-eutrophication due to extensive clearing of surrounding forest and wetland. Oxygen availability remains high throughout the year (63 to 160% saturation, Binning & Chapman, 2009), relative to nearby papyrus swamp systems (Crispo & Chapman, 2008).
5 We predict that phenotypic plasticity of traits related to oxygen uptake and use, such as gill and brain size, will be low because (1) local adaptation is not constrained by gene flow from
Accepted Article
surrounding populations (Crispo & Chapman, 2008), and (2) dissolved oxygen conditions remain high throughout the year; and therefore, oxygen should not be limiting in ecotonal regions of the lake (Binning & Chapman, 2009). As an alternative, if plasticity is not costly and is an ancestral trait in cichlids, we might observe higher levels of plasticity. If this is the case, body morphology may also vary as an indirect response to environmental change, in order to accommodate changes in gill and brain size. Yet, the evolution of reduced plasticity might still occur due to drift (Masel et al. 2007).
MATERIALS AND METHODS
Laboratory experiment P. multicolor were collected from Lake Saka, using baited minnow traps, in June 2009 and transported live to McGill University, Montreal, Quebec. See Crispo & Chapman (2008) for a description of the collection site and dissolved oxygen measures. Binning & Chapman (2009) provide monthly mean values for dissolved oxygen concentration (DO) and water temperature in Lake Saka over a 14-month period (May 2007-June 2008) in an ecotonal area. Over that time, DO averaged 6.3±0.4 mg/L, (SE of monthly values, range = 4.6 to 9.2 mg/L; ~63 - 109% O2
saturation) in the morning and 10.6±0.4 ng/L in the afternoon (range 9.0 to 13.4 mg/L; ~109 160% O2 saturation). Water temperature reported by Binning & Chapman (2009) varied little over the 14-month period, typical of this equatorial region, averaging 22.3±0.3 oC in the morning
(range = 20.2 to 23.8 oC ) and 24.0±0.3 oC in the afternoon (range = 22.1 to 25.4 oC ).
6 We carried out a laboratory-rearing study to examine the extent of morphological plasticity in the Lake Saka P. multicolor population in response to high versus low DO
Accepted Article
environments. The experiments took place at McGill University after shipping the collected adults from Uganda. The experiment was conducted following previously described protocols (Crispo & Chapman 2010). Specifically, six full-sib families were raised from the Lake Saka collection site, where a family consisted of the family of one male-female pair. F1 individuals from each family were split between high and low DO treatments one week after release from the mouth (P. multicolor are mouth-brooders). Families were placed in separate 53-liter tanks,
for a total of six low DO tanks and six high DO tanks. Tanks were filtrated using Fluval underwater filters (Hagen). Low DO conditions were maintained with a dissolved oxygen control system (Point Four systems Inc., Coquitlam, British Columbia), and high DO conditions were maintained by bubbling air through the water column. In the low DO treatment, the mean DO concentration was 1.0 mg/L, and tank means ranged from 0.96-1.1 mg/L. In the high DO treatment, the mean DO concentration was 7.3 mg/L, and tank means ranged from 6.8-7.6 mg/L. The mean ambient temperature was 25.0°C, and tank means ranged from 24.4-25.2°C. At an age of one month, families were culled to 10 fish per tank. If fewer than 10 fish were present in a tank, we did not perform culling. The McGill University Animal Care Committee approved the protocol for the project. At an age of approximately one year, we harvested the two largest males from each
family for analysis of gill metrics, brain mass, and external body morphology. Initially this included 24 males (12 from each oxygen treatment). However, we removed one male (high DO treatment) with a mass well below the range of the other fish to prevent the possibility of skewed results due to an allometric growth relationship. We chose only the two largest males because the
7 development of subordinate fish was suppressed due to dominance hierarchy; and sex has been shown to influence gill shape in female P. multicolor (O’Connor et al. 2012). Previous studies
Accepted Article
on plasticity in this system also examined only the two largest males per tank (Crispo & Chapman 2008; Chapman et al. 2000, 2008). All remaining males and females were then
harvested to be included in external body morphology analyses (totals of 43 females and 27 males from low DO; and 33 females and 19 males from high DO). Fish designated as “females” for the body shape analysis also included juvenile males that had not yet developed secondary sex characteristics. All fish were euthanized using buffered tricaine methanesulfonate (MS-222; pH 7.0) and were preserved in 4% paraformaldehyde (buffered with phosphate buffered saline; pH = 7.0).
Gill parameters We separated the four gill arches (hemibranches) of the left branchial basket by dissection. The left and right sides of each gill arch were laid flat on a microscope slide and photographed using a Lumenera Scientific Infinity camera attached to a dissecting microscope. We measured five gill metrics using ImageJ 1.43u (Fig. 1), including total gill filament length (TGFL, in mm), average gill filament length (AFL, in mm), total number of gill filaments (TNF), total hemibranch area (THA, in mm2), and total perimeter of hemibranchs (TP, in mm). These measures are representative of gill size, and show potentially different ways that gill size could shift in response to environmental conditions. The length of every fifth filament was measured on both sides of each hemibranch. To increase resolution, every second filament was measured for the first five and the last five filaments on each hemibranch (Fig. 1). Filament lengths were measured along the length of the filament from the tip to the base, using segmented lines for
8 curved filaments. If filaments were missing or not clearly visible, measurements were estimated using lengths of surrounding filaments. Variables were measured on each of the four gills, and
Accepted Article
multiplied by two for the total number for each fish (i.e., because only one side of the head was dissected per fish) following procedures used in earlier studies (e.g., Crispo & Chapman 2010). We performed analyses of covariance (ANCOVA) on each gill metric separately to see if
the DO regimes influenced individual gill characteristics, while controlling for variation among families and variation due to body mass. We were not able to perform a multivariate analysis (MANCOVA) because the random factor (family) could not be included in this type of test. All analyses were performed using Type III sums of squares in SPSS version 17.0. Tests included log10-transformed gill metrics as response variables, log10-transformed body mass as a covariate,
treatment (high or low DO) as a fixed factor, family as a random factor, and the interaction between family and treatment. The treatment-by-body mass interaction was non-significant (results not shown) and therefore not included in the model. Using a two-tailed t-test we found that fish bred under high DO had a greater mass than fish bred under low DO (p = 0.045; mean mass under high DO was 0.47 g and under low DO was 0.33 g). This may be expected as fish under oxygen stress may allocate fewer resources to growth. However, there was good overlap in the size range of the two groups. We calculated overall difference between treatments as the percent difference between oxygen treatments for each gill metric, using anti-logged means adjusted to a common body mass (across families and treatments; least-square means obtained from the ANCOVA). Percent difference was calculated as a function of the mean value under high-oxygen: 100*(average under low DO – average under high DO)/(average under high DO). Additionally, we measured the total gill filament length (as above) for the second gill arch (i.e. second from the operculum) for 10 male fish caught in the wild. These wild-caught fish
9 were collected in June 2006 during a previous expedition, euthanized using buffered tricaine methanesulfonate (MS-222; pH 7.0), and preserved in 4% paraformaldehyde (buffered with
Accepted Article
phosphate buffered saline; pH = 7.0). We were not able to analyze the brain mass or body morphology of these fish because the samples had been destroyed during the dissection of the gills. We performed an ANCOVA for log10-transformed TGFL, with treatment (high DO, low DO, wild) included as a fixed factor and log10-transformed body mass as a covariate. The treatment-by- body mass interaction was not significant and thus not included. This analysis was performed to examine how the gill size of lab-raised fish compared to those caught in the wild.
Brain mass To measure relative brain mass we extracted whole brain from the same laboratory-raised fish used for gill analysis. Standard dissection methods were used following Chapman et al. (2008).
Brains were preserved in 4% buffered paraformaldehyde. We obtained wet mass by blotting each brain on a Kimwipe to removed excess paraformaldehyde, and measuring each to the nearest 0.1 mg using an analytical balance (Mettler AC 100). We measured each brain once every day for five consecutive days, and used the average of the five measures for analysis. Brains weighed less on each consecutive day, which was likely due to handling and the fragility of the brains. To assure that the effect was consistent across all brains, we performed a Pearson’s Correlation test and found that measures of brain mass taken on each day strongly correlated with each other (0.9 < r < 1; p < 0.001 for all comparisons). The ANCOVAs and percent difference calculations for brain mass relationships were carried out in the same way as for the gill metrics, with the same factors and covariate.
10 Body morphology We analyzed morphology for all fish harvested over the course of the experiment. Photos and
Accepted Article
linear measures were taken prior to dissection. We included the males used for the gill and brain analyses, along with the females, additional males, and juveniles present in each tank. After preservation in 4% buffered paraformaldehyde, we measured fish weight to the nearest 0.1 mg, and head width over (HWOO) and behind the operculum (HWBO) to the nearest 0.1 mm using digital callipers (Mitutoyo, Model CD-6”B). We included HWBO because it may show variation in head width independent of gill size. We then placed each fish on a laminated grid mounted on a level structure and captured an image of its left side using a digital camera (Canon PowerShot G9). Total body length, from the tip of the snout to the hypural plate (in mm), was later measured with these photographs using ImageJ version 1.43u. We digitized 11 landmarks and 4 semilandmarks using tspDig version 2, following the procedures in Crispo & Chapman (2011) (Fig. 2). We then performed geometric morphometric analysis using tspRelw version 1.45. The program uses the Generalized Procrustes Method for superimposition. This analysis yielded partial, uniform, and relative warp scores and centroid sizes that were used in further analysis. Partial warps scores are vectors representing localized variation in landmark positions. Uniform warps are a type of partial warps that represent parallel changes in landmark positions. Relative warps are independent linear combinations of partial warp i.e. principal components of the shape coordinates; we therefore refer to the relative warps as principal components (PC). Centroid size (covariate) is a measure of the total of the distances from each landmark to the center-point of all the landmarks. We analyzed principal components (relative warps) that each explained more than 5% of the variation in the data, and present results from only those that showed significant differences between treatments.
11 We performed ANCOVAs to test for effects of treatment and family on head shape, while controlling for variation due to body length. Log10-transformed HWOO and HWBO were
Accepted Article
included as response variables in two separate analyses. In both analyses, treatment (high or low DO) was included as a fixed factor, family as a random factor, and log10-transformed total body length as a covariate. Sex effects were not significant (results not shown) and thus not included. The family-by-treatment interactions were also included. We calculated overall difference between treatments (relative to head size under high DO, as for the gills) as the percent difference between DO treatments for each head measure, using anti-logged means adjusted to a common body length. For the geometric morphometrics analysis, we conducted a multivariate analysis of
covariance (MANCOVA) to test for effects on overall body shape. For this analysis, we were able to include all partial and uniform warps as response variables. A drawback of multivariate models is that random factors cannot be included, so the effect of family is excluded in this analysis. We included treatment as a fixed factor and centroid size as a covariate. Sex effects were not significant (results not shown) and thus not included. The treatment-by-centroid size interaction was also removed because it was not significant (result not shown). To complement the above analysis, we used ANCOVAs to test for family and treatment
effects on specific components of body shape (principal components; PC), while controlling for variation among families and variation due to centroid size. We included PC 1 and 2 as the response variables because each explained more than 5% of the variation and showed significant – or approaching significant – differences between treatments. We included treatment as a fixed factor, family as a random factor, and centroid size as a covariate. Sex effects were not significant (results not shown) and thus not included. For PC 1 we included the treatment-by-
12 centroid size interaction, and removed all other non-significant interactions. We did not include the family-by-centroid size interaction. For PC 2 we initially included interactions with centroid
Accepted Article
size, but subsequently removed them when non-significant (results not shown). We also analyzed PC’s 3, 4 and 5, because each also explained more than five percent of the variation, but do not present the results because treatment effects were not significant.
Meta-analysis In addition to the overall goal of characterizing the plasticity of an isolated population of P. multicolor, we were interested in comparing morphology and phenotypic plasticity between the Saka population and other P. multicolor populations from the Mpanga River drainage, the latter
of which were already characterized in a previous study (Crispo & Chapman 2010). Different experimenters measured the gills for Lake Saka (K. Wiens) and the Mpanga populations (E. Crispo), which could bias the results. Thus, one experimenter (E. Crispo) re-measured the TGFL for the second hemibranch of the Lake Saka fish. The proportional difference was calculated for each hemibrach (re-measure – original measure/average of the two measures), and then an average proportional difference was obtained by averaging these 44 values (22 fish, two hemibranchs per gill arch). This proportional difference was used to standardize the original labraised Lake Saka values to the measures used in the previous study (Crispo & Chapman 2010). We conducted two ANCOVAs, one for TGFL and one for brain mass. Treatment and
population were included as fixed factor, family was included as a random factor, and log10transformed body mass was included as a covariate. In both cases, response variables (TGFL or brain mass) were log10-transformed. We initially included all possible interactions among
population, treatment, and body mass, but sequentially removed non-significant interactions.
13
RESULTS
Accepted Article
Gills In the univariate tests, treatment effects were highly significant for all gill metrics with the exception of total number of filaments (TNF) (Table 1). All gill metrics increased with body mass and, with the exception of TNF, were larger under low DO (when controlling for variation in body mass; Fig.3). The overall difference between treatments ranged from a low of 4.2% for TNF (not significant) to a high of 32.6% for total hemibranch area (THA) (Table 2). Body mass, family, and family-by-treatment effects were not significant (Table 1). In the analysis including the wild-caught fish, the treatment term was significant (F2,30 =
41.686, P < 0.001), and log10-transformed body mass was a significant covariate (F1,30 = 70.593,
P < 0.001). As expected, TGFL increased with body mass. Controlling for body mass, TGFL for wild-caught fish did not differ from that for fish raised under high DO, while TGFL for fish raised under low DO was larger than wild-caught and high DO-raised fish (Fig. 4). These results are expected based on the assumption that the high DO treatment represents oxygen conditions similar to those experienced in the wild.
Brains Treatment effects on brain mass were also significant (Table 1). Brains were smaller under low DO when controlling for variation in body mass. The overall difference between treatments was 15.2% (Table 2). Although brain mass tended to increase with body mass, this effect was not significant (Fig. 3). Additionally, family and family-by-treatment effects were not significant (Table 1).
14
Body morphology
Accepted Article
There was a significant effect of treatment on head width over the operculum (HWOO), with wider heads in the fish raised under low DO when controlling for variation in body length. However, there was no effect of treatment on head width behind the operculum (HWBO), suggesting that the variation in HWOO was due to widening of the gills under low DO. Both measures increased with body length (Fig. 3), but there was no effect of family on either measure. The family-by-treatment effects were significant for both HWOO and HWBO, meaning that the level of plasticity (i.e. treatment effects) varied among families – this could reflect genetic variation and/or tank effects (Table 1). Sex effects were not significant and were thus not included in the model. The multivariate analysis of partial and uniform warps revealed strong effects of
treatment and centroid size on body shape (Treatment F26,94 = 3.675, P < 0.001; Centroid size
F26,94 = 7.346, P
