Variation in anhydrobiotic survival of two eutardigrade morphospecies: a story of cryptic species and their dispersal

Authors


Correspondence
Søren Faurby, Ecology and Genetics, Department of Biological Sciences, University of Aarhus, Ny Munkegade, Building 1540, DK-8000 Aarhus C, Denmark. Tel: +45 89 423 335; Fax: +45 89 422 722
Email: soren.faurby@biology.au.dk

Abstract

Studies of geographic variation in anhydrobiotic tolerance may increase our understanding of the population dynamics of terrestrial meiofauna and the relative importance of local adaptation and microhabitat niche separation. Although anhydrobiosis in tardigrades has been studied extensively, few studies have dealt with intraspecific variation in survival and none of these included genetic data to validate the intraspecific nature of the comparisons. Such data are necessary when working with meiofauna as cryptic species are common. We analysed the anhydrobiotic survival and genetic variation in cytochrome oxidase subunit I of two eutardigrades (Richtersius coronifer and Ramazzottius oberhaeuseri) from Italy and Sweden to detect possible local adaptation. Survival was analysed as a multidimensional contingency table and showed that anhydrobiotic survival was higher in Sweden for Ra. oberhaeuseri whereas no significant geographic variation was found for Ri. coronifer. Our genetic analysis indicated the coexistence of two cryptic species of Ra. oberhaeuseri in Italy, only one of which was found in Sweden. It could not be determined whether the variation in Ramazzottius is intra- or interspecific due to the presence of these cryptic species. We suggest that geographic variation in anhydrobiotic survival may be a general phenomenon in tardigrades but further research is necessary to determine the degree of intraspecific variation. The genetic analysis showed indications of long-term isolation of the individual populations of Ri. coronifer but recent dispersal in one of the cryptic species of Ramazzottius. We found higher survival in Ra. oberhaeuseri than in Ri. coronifer. These results indicate a possible coupling between anhydrobiotic survival and dispersal rate.

Introduction

Tardigrades are microscopic metazoans known for their remarkable ability to reversibly stop their metabolism and survive unfavourable periods. This ametabolic state is called anhydrobiosis when it is induced by drought, and it represents an aspect of a wider class of resistance to several environmental stresses called cryptobiosis (Keilin, 1959). Most research on cryptobiosis has focused on the morphological, physiological and molecular mechanisms, while few studies have looked at ecological aspects.

Interspecific variation in anhydrobiotic survival has been confirmed in tardigrades (Wright, 1989; Rebecchi et al., 2006), rotifers (Ricci, 1998) and nematodes (Grewal, 2000; Shannon et al., 2005). Furthermore, it has been determined experimentally that energy reserves (estimated by the size of storage cells) decrease following a period of anhydrobiosis (Jönsson & Rebecchi, 2002), and ecological studies usually find different species in habitats with different exposure to drought (Ramazzotti & Maucci, 1983; Wright, 1991). This indicates a cost of anhydrobiosis and competitive exclusion in benign environments. Similar patterns are often found where different species express different degrees of stress tolerance (Hoffmann & Parsons, 2001).

Environmental heterogeneity across a species' distribution range should result in variation in the optimal investment in anhydrobiotic ability, and different genotypes are favoured in different areas of a species distribution, resulting in selection for local adaptation. However, the extent of local adaptation depends on the amount of gene flow from other populations. Limited gene flow can increase the potential because it increases the amount of genetic variation in each population, whereas too much gene flow removes any local adaptation and causes evolution towards genotypes with a high average fitness across environments (Lenormand, 2002). Dilution of local adaptation may be a larger problem in drier habitats, because populations there will have longer effective generation times and immigration of unadapted individuals may therefore be relatively more common.

The potential for local adaptation is also influenced by the distribution of environments favouring different genotypes. A large-scale cline increases the likelihood of local adaptations, whereas a checkerboard-like distribution with areas favouring one genotype in close proximity to areas favouring other genotypes would decrease local adaptations by promoting gene flow. Over large scales, changes in both precipitation and temperature regimes and on smaller scales, different degrees of shade, can influence the selection for anhydrobiotic tolerance by changing the likelihood, duration and extent of exposure to drought. The extent of local adaptation depends on the relative importance of these large- and small-scale effects.

The immediate fitness of each individual, that is the number of offspring in the next generation, is mainly based on adaptation to the local environment but of equal or greater importance on larger timescales is the dispersal ability of individual genotypes. A few refugia for temperate organisms during the last ice age have been postulated in northern Europe (Stewart & Lister, 2001; Bettin et al., 2007) but, nevertheless, it is likely that the vast majority of the species of this region have dispersed from the Mediterranean area since the last ice age. Because higher anhydrobiotic capacity may increase the dispersal ability, it is possible that populations in recently colonized areas have better anhydrobiotic ability than populations in refugial areas even if the local environment in the recently invaded areas does not favour anhydrobiotic investment. Local adaptation can therefore be masked in such cases even when there is no current gene flow. Another factor influencing the dispersal rate is the reproductive mode as parthenogenetic individuals can found new populations more easily (Bell, 1982; Bertolani, Rebecchi & Beccaccioli, 1990). In this study, however, we do not have any confounding effects of reproductive mode as we are only looking at parthenogenetic animals.

Only two studies have attempted to analyse intraspecific variation in anhydrobiotic ability in tardigrades. Horikawa & Higashi (2004) found significantly higher survival in Milnesium tardigradum Doyère, 1840 from Japan compared with those from Indonesia. It is, however, unlikely that this study was in fact comparing intraspecific populations. Recently, M. tardigradum has been split into a number of species (Tumanov, 2006) and, given the very large climatic difference and geographic distance between Indonesia and Japan, the study by Horikawa & Higashi (2004) may indicate differences in anhydrobiotic survival between two cryptic species, rather than two populations within one species, of Milnesium.

Another study by Jönsson, Borsari & Rebecchi (2001) compared anhydrobiotic ability in populations of two eutardigrade species from lowland Sweden and highland Italy. This study showed no significant difference in anhydrobiotic survival between the two populations in either species, whereas body size, species and experimental groups (Petri dishes) were all significant correlates of anhydrobiotic ability. The data were analysed using a logistic regression approach requiring the relationship between size and survival to be monotonous (i.e. constantly negative or constantly positive). Other studies on the same populations have, however, found contrasting results on the relationship between size and anhydrobiotic survival, probably owing to a lower survival of juveniles and older individuals (Jönsson et al., 2001; Jönsson & Rebecchi, 2002). Jönsson et al. (2001) also reported differences in body size between the populations, and it is possible that a different usage of size without the assumption of a monotonous relationship with survival may yield another result.

The study by Horikawa & Higashi (2004) compared two population/species with very different environments and found, as expected, larger survival in the drought- and cold-adapted population. In the study by Jönsson et al. (2001), it is difficult to predict which population is expected to have the largest anhydrobiotic survival. Their climatic data indicate that the Italian population may experience slightly drier conditions, but the differences are small and the relationship between the climatic data and the microhabitat of the tardigrades may not be obvious. Here we report a re-analysis of the dataset by Jönsson et al. (2001) and an analysis of the relationship between anhydrobiotic ability and genetic similarity between populations based on extracted mitochondrial DNA from individuals of the same populations used in the original study.

Materials and methods

Species and collection

Two species of terrestrial tardigrades Richtersius coronifer (Richters, 1903) (Eutardigrada, Macrobiotidae) and Ramazzottius oberhaeuseri (Doyère, 1840) (Eutardigrada, Hypsibiidae) were collected in Sweden and Italy on several occasions. Tardigrades for the study on anhydrobiotic survival were all collected in March 2000 in both countries. Those used for the genetic study were collected for both species in Sweden in January 2007, while the Italian samples were collected in March 2000 and in February 2003 (Ra. oberhaeuseri) and in July 2001 (Ri. coronifer). In Sweden, both species were found at the same locality at Öland in southern Sweden. In Italy, the species were collected at two different but nearby sites in the Northern Apennines south of Modena. Richtersius coronifer were extracted from the moss Orthotrichum cupulatum Hoffman ex Bridel, 1801 in Sweden and Homalothecium sericeum (Hedw.) Schimp 1851 in Italy. Ramazzottius oberhaeuseri were extracted from the lichen Xanthoria parietina (L.) in both localities. For further details and description of the two habitats, see Jönsson et al. (2001).

Experimental design

After collection, moss and lichens samples were air-dried and kept under laboratory conditions [+20 °C, 40–50% relative humidity (RH)]. The air-dried moss and lichens were rehydrated within 4 weeks. The viability of the tardigrades was observed under a stereomicroscope. Active animals were collected and placed individually in plastic tubes with a fine mesh (0.03 mm) floor, one animal in each tube. The top was covered with cotton to reduce the rate of drying. The tubes were embedded in Petri dishes with clean wet sand, 50 tubes in each Petri dish (25 from each population of the same morphospecies), resulting in four Petri dishes with a total 100 specimens of each population. Anhydrobiosis was induced by drying the Petri dishes at 23 °C and 65% RH for 12 days (climate-controlled room).

After the dry period, the animals were rehydrated by wetting the sand with distilled water. Each specimen was checked for viability after 7 h, and animals showing no sign of movement were considered to be dead. All animals were then fixed in Carnoy's fluid (methanol:acetic acid, 3:1, v/V), stained in toto with acetic-carmine and then mounted in Faure-Berlese fluid. For all animals, body length was measured under an Olympus BX60 light microscope (Tokyo, Japan).

Analysis of survival

The data were analysed as a multidimensional contingency table with the package CoCo. (Badsberg, 2001) from the program R (R Development Core Team, 2005).

Four-dimensional (population, dish, size group, survival) contingency table analyses were performed for each morphospecies. For these analyses, the animals were grouped into size groups based on body lengths. Richtersius coronifer were grouped into <400, 400–500, 500–600, 600–700, 700–800 and >800 μm. Ramazzottius oberhaeuseri were grouped into <200, 200–250, 250–300, 300–350 and >350 μm.

As the ranges of the size groups were chosen arbitrarily, all analyses were repeated with other ranges to make sure that the effects were consistent. For Ri. coronifer ranges of 50 and 200 μm were tried, and for Ra. oberhaeuseri, ranges of 25 and 100 μm were tried. The results of these analyses will only be mentioned when the range of the size group changed the results.

For Ri. coronifer, three-dimensional tables (size group, survival, dish) were analysed to see whether the interaction between size and survival was present in each population, and for Ra. oberhaeuseri three-dimensional tables (population, survival, dish) were analysed for all size groups.

We tested the significance of all interactions in the full model with all possible interactions. This is a conservative approach as it increases the number of degrees of freedom in most tests. Because all analyses were performed on both species, we used a Bonferroni correction for multiple comparisons (Rice, 1989). Following Miller's (1981) suggestion, we made a separate probability statement for each interaction.

Genetic analysis

After tardigrade taxonomic identification and cleaning, total genomic DNA was extracted from single rehydrated individuals with the STE-buffer method (Maniatis, Fritsch & Sambrook, 1982). A fragment of cytochrome oxidase subunit I (COI) was amplified from six individuals from each species at each population with the primers LCO and HCO (Folmer et al., 1994) for Ri. coronifer and LCO (Folmer et al., 1994) and HCOoutout (Giribet & Edgecombe, 2006) for Ra. oberhaeuseri. PCR reactions were performed using Qiagen mastermix© (Hilden, Germany). The PCR settings for both species were a preheat step at 95 °C for 5 min, 30 cycles of denaturation at 94 °C for 10 s, annealing at 42 °C for 30 s and amplification at 72 °C for 30 s and a final extension step at 72 °C for 3 min.

The product was checked on 1% agarose gels and successful amplifications were cleaned using Roche's High pure PCR product purification kit© (Basel, Switzerland) and thereafter sequenced in both directions by Macrogen® company (Seoul, Korea).

Forward and reverse sequences were compared and edited in BioEdit (Hall, 1999). All sequences were compared with the database at NCBI using Blast (Altschjul et al., 1990). Alignment was performed using ClustalX (Thompson, Higgins & Gibson, 1994) and checked manually. All sequences were of similar length and no gaps had to be postulated. Divergences between all sequences were calculated with K2P using Mega v. 3.1. (Kumar, Tamura & Nei, 2004). Two published sequences of Ri. coronifer from Italy (AY598780 and AY598781; Guidetti et al., 2005) and one sequence of Ra. oberhaeuseri from Denmark (EF620418; Møbjerg et al., 2007) were included in the analyses.

Results

Survival analysis

Survivorship (%) of each population, size group and Petri dish are given in Table 1.

Table 1.  Anhydrobiotic survival of the tardigrades Richtersius coronifer and Ramazzottius oberhaeuseri according to the population, Petri dish and body size
SpeciesGroupSurvival
ItalySweden
  1. The number of survivors, the total number of animals in the group and the survival rate in per cent are given.

Ri. coroniferAll animals45/100(45%)34/100(34%)
Petri dish 110/25(40%)13/25(52%)
Petri dish 210/25(40%)5/25(20%)
Petri dish 313/25(52%)8/25(32%)
Petri dish 412/25(48%)8/25(32%)
<400 μm0/4(0%)2/18(11%)
400–500 μm6/18(33%)9/27(33%)
500–600 μm3/16(19%)5/15(33%)
600–700 μm,16/28(57%)5/19(26%)
700–800 μm15/22(68%)12/18(67%)
>800 μm5/11(45%)1/3(33%)
Ra. oberhaeuseriAll animals65/100(65%)67/99(68%)
Petri dish 123/25(92%)20/25(80%)
Petri dish 223/25(92%)13/25(52%)
Petri dish 313/25(52%)22/25(88%)
Petri dish 46/25(24%)12/24(50%)
<200 μm7/9(78%)3/5(60%)
200–250 μm27/36(75%)22/29(76%)
250–300 μm20/38(53%)27/42(64%)
300–350 μm10/16(63%)12/18(67%)
>350 μm1/1(100%)3/5(60%)

For Ri. coronifer there was no effect of population on survival, whereas both size and Petri dish had a significant effect, although the effect of Petri dish is only significant before correction (see Table 2). There was also a significant size difference between Petri dishes but not between populations. These results were concordant with other ranges of the size groupings.

Table 2.   Results of four-dimensional contingency table analyses of Richtersius coronifer and Ramazzottius oberhaeuseri
SpeciesInteraction−2log (Q)d.f.Corrected
P-value
  1. All P-values are Bonferroni's corrected (K=6).

Ri. coroniferSurvival–population18.272917NS
Survival–size70.8694355.28 × 10−6
Survival–Petri dish48.0547300.1167
Size–population45.378233NS
Size–Petri dish102.5407579.12 × 10−6
Ra. oberhaeuseriSurvival–population37.8647157.74 × 10−5
Survival–size32.109027NS
Survival–Petri dish80.6261250.00331
Size–population35.019526NS
Size–Petri dish69.5009450.0835

For Ra. oberhaeuseri there was no effect of size on survival, whereas both population (highest survival in Sweden) and Petri dish had a significant effect; the effect of Petri dish is, however, only significant before correction (Table 2). The interaction between size and population was non-significant.

The three-dimensional tables of Ri. coronifer showed that the interaction between size and population was present in both populations (Sweden: P=0.00618; Italy: P=0.00312).

The three-dimensional tables of Ra. oberhaeuseri indicated that the interaction between population and survival appears to be stronger in the larger animals (Table 3). It is non-significant in <200 μm, nearly significant in 200–250 and 250–300 μm, highly significant in 300–350 μm and nearly significant in >350 μm. A problem with this analysis is that the number of replicates and therefore the power of the test is different in the different size groups, but this would result in lower P-values in size groups with a higher number of replicates, which is not found and so the trend appears to be genuine.

Table 3.   Results of three-dimensional contingency table analyses of difference in survival between the two populations of Ramazzottius oberhaeuseri within the individual size groups
Size group (μm)Replicates−2log (Q)d.f.P-value
  1. Replicates give the number of animals in the two populations that belong to each size group.

<200141.18520.553
200–250687.84740.097
250–300768.70040.069
300–3503515.41240.004
>35065.06220.051

Genetic analysis

For Ra. oberhaeuseri, 770 nucleotides were amplified, of which 126 were variable. Four different haplotypes were identified (all present in Italy); three unique in Italy, and only one present also in Sweden. The haplotypes fall into two groups with 0.3–0.6% variation within each group but 18.5% difference between haplotypes from different groups.

The divergence between the published sequence from Denmark (EF620418; Møbjerg et al., 2007) and the haplotypes from this study varied between 4.0 and 14.0%. The sequence from the Danish specimens appears to be a mosaic with different parts belonging to the two different clusters. The sequence was amplified from a pool of several animals and it is likely that individuals from both clusters were found in this pool. The trace file from the sequence has been manually checked and in 34 bases there appear to be two tops for bases showing variation between the two clusters.

For Ri. coronifer, 630 nucleotides were amplified, of which 130 were variable. Three different haplotypes were identified, of which one was unique for Sweden and two for Italy. The sequences from the different populations are 2.8–2.9% divergent and 22.9–23.4% different from the published sequence of the same species.

For both morphospecies, the frequencies of the individual haplotypes can be seen in Table 4 and the divergence between them can be seen in Table 5.

Table 4.   Number of individuals with the different haplotypes of Ramazzottius oberhaeuseri and Ricthersius coronifer from the two populations
 HaplotypeSweden, 2007Italy, 2001Italy, 2003Italy, 2000GenBank Acc. Number
Ra. oberhaeuseriRama1621EU251379
Rama2020EU251380
Rama3015EU251381
Rama4010EU251382
Total 666
Ri. coroniferCoro105EU251383
Coro201EU251384
Coro360EU251385
Total 66
Table 5.   Divergences in per cent according to the K2P model between the different haplotypes within each species
 Rama 2Rama 3Rama 4Rama NCBI
Ramazzottius oberhaeuseri
Rama 10.318.518.54.5
Rama 2 18.518.55.0
Rama 3  0.614.0
 Coro 2Coro 3Coro NCBI
  1. The haplotypes for the two published sequences of Ri. coronifer that are identical except for differences in length are called Coro NCBI. The haplotype from the published Ra. oberhaeuseri is called Rama NCBI.

Richtersius coronifer
Coro 10.22.822.9
Coro 2 2.923.1
Coro 3  23.4

Discussion

Our analysis shows that when the non-monotonous nature of the relationship between size and survival and the potential interaction between different factors are controlled for, there is in fact a difference in anhydrobiotic capacity between the two populations of Ra. oberhaeuseri. The four-dimensional analysis (Table 2) shows no effect of size on survival, but the three-dimensional contingency table analyses of difference in survival between the two populations of Ra. oberhaeuseri within the individual size groups do indicate an effect of size. The latter result is in agreement with Jönsson et al. (2001), and maybe an effect of size was masked in the four-dimensional table because the effect of population is only evident in larger animals. We find a significant interaction between Petri dish and survival for Ra. oberhaeuseri. The reason for this is unknown. It could indicate minute differences in the drying rate between Petri dishes but it could also be coupled with difference in animal size between Petri dishes and the likely effect of size on survival.

In contrast to the study of Jönsson et al. (2001), this analysis shows no significant difference between the sizes of the two populations of Ra. oberhaeuseri. One explanation is that the grouping of individuals into size classes reduces the power of the analysis and because the original results are only weakly significant, the difference in results can be caused by a lack of power. Another possibility is that the different results are caused by the inclusion of the size by Petri dish interaction. The rest of the results in this analysis are in agreement with Jönsson et al. (2001).

Based on the genetic divergence between the two clusters of Ra. oberhaeuseri, we suggest that they should be considered as cryptic species. Similarly, in Ri. coronifer it is reasonable to conclude that the sequences from this study and the GenBank sequences (published in Guidetti et al., 2005) belong to two different cryptic species. The population from GenBank is bisexual and amphimictic, whereas the two populations considered here carried out an automictic parthenogenesis (Rebecchi et al., 2003). At present, we cannot determine whether the sequences of Ri. coronifer from Italy and Sweden considered here represent an intra- or an interspecific variation. The difference between these two populations is, however, much higher than is usually found in intraspecific comparisons (Kerr et al., 2007).

The genetic analysis offers an interesting explanation for the variation in anhydrobiotic ability between the populations of Ra. oberhaeuseri and lack of the same in Ri. coronifer. In Ramazzottius, two different cryptic species were found in the Italian sample whereas only one was present in Sweden. Because the only haplotype found in Sweden was also found in Italy, the Swedish population is probably post ice age, founded from individuals from a southern refugium. If the founding individuals were then of the species and genotype with the largest anhydrobiotic investment, this could explain the significantly higher anhydrobiotic survival in the Swedish population. The NCBI sequence of Ramazzottius has several uncertainties, as already mentioned, but because it appears to be comprised of parts from both haplotype clusters, this indicates that both cryptic species are present in Denmark.

It seems odd that the two Italian cryptic species of Ramazzottius are found in the same habitat if one species has better-developed anhydrobiotic ability than the other, because competitive exclusion of one of the species might be expected. The reason for this is unknown but a possibility is microhabitat variation and niche separation between different parts of the lichen. The significant interaction between body size and Petri dish, which was found in both species, could perhaps also be attributed to microhabitat variation that was accidentally transferred to the experiment.

Our results indicate that the divergence between the two populations of Ri. coronifer studied here may predate the most recent ice age. The difference between the two populations could indicate that they have been separated around 1.5 million years following the usual molecular clock of about 2% divergence in COI per million years (Brower, 1994). Several factors influence this dating but even a fivefold variation in mutation rates (which is higher than any published studies) indicates a divergence time between 0.3 and 7.5 million years and it is thus highly unlikely that the divergence is post ice age. The fact that both populations harbour a low genetic variation (all Swedish individuals had the same haplotype and only two haplotypes were found in Italy separated by a single nucleotide) supports the notion of these being reproductively isolated. It seems likely that Ri. coronifer survived the glaciations in situ but genetic data from other possible southern European refugia in Spain and the Balkans are necessary to confirm or reject this. Tardigrades are known for high survival under cold conditions and population survival in situ during the ice ages has also been postulated in Antarctica (Gibson et al., 2007). These are, however, the first molecular data to support in situ survival of tardigrades in glaciated areas. It is interesting that Ri. coronifer apparently survived the glaciations in Sweden whereas Ra. oberhaeuseri did not, because drought and frost are very similar processes (Ramløv & Westh, 1992) and Ri. coronifer has significantly lower anhydrobiotic survival than Ra. oberhaeuseri (Jönsson et al., 2001).

Our analysis found cryptic species in both Ra. oberhaeuseri and Ri. coronifer. The presence of mixed populations of two morphologically very similar Ramazzottius species has been found previously in Italy (Bertolani & Rebecchi, 1988; Rebecchi & Bertolani, 1988), and was interpreted as an absence of gene flow among populations. As stated above, the populations of Ri. coronifer used in this study are parthenogenetic (Rebecchi et al., 2003) whereas the individuals from NCBI are sexual and amphimictic (Rebecchi et al., 2003; Guidetti et al., 2005; R. Guidetti pers. comm.). A deep split between the parthenogenetic and amphimictic sexual forms of Ri. coronifer has already been noted in an allozyme study (Rebecchi et al., 2003). Our study is not capable of dating this split as partial saturation between the sequences is likely to have occurred but 10 million years seems to be a conservative estimate. The development of parthenogenesis is therefore likely to predate the climate changes during the Pleistocene glaciations.

Knowlton (1993) has suggested that up to 10 times the number of currently described marine species remain to be found and that the majority of these will be cryptic. She further notes that the reason why cryptic species are likely to be so common in the sea is that many species are difficult to observe under natural conditions, making a number of behavioural and microhabitat characters hard to attain for systematic studies. Marine and terrestrial tardigrades pose similar problems with regard to microhabitat, as observations under a microscope may not be easily transferable to their natural environment. Despite great efforts by taxonomists, it therefore seems likely that large numbers of undescribed cryptic species remain in tardigrades.

Our study has documented geographic variation in anhydrobiotic survival in the two populations of Ramazzottius and such a variation has thus been found in two out of three different tardigrades investigated so far. No study has so far been able to determine whether this variation is intra- or interspecific. To analyse this, studies of species with known genetics are needed. There are not yet sufficient data to determine the relative importance of small- and large-scale climatic variation in determining optional anhydrobiotic investments. The limited variation from this study and that by Horikawa & Higashi (2004), however, indicates that large-scale variation may be the most important driving force.

Our studies found interesting correlations between cryptobiotic ability and dispersal ability because the genetic similarity of the two populations of Ra. oberhaeuseri is much larger than that for Ri. coronifer. Furthermore, our results indicate a northern refugium for Ri. coronifer.

Acknowledgements

S.F. was supported by grants from Augustinusfonden and Oticonfonden. P.F. was supported by a grant from the Danish Natural Sciences Research Council (272-06-0534). L.R. was supported by FAR 50% from the University of Modena and Reggio Emilia. K.I.J. was supported by the Carl Trygger Foundation and The Crafoord Foundation. We would like to thank Else Bomholt Rasmussen and Camilla Håkansson for laboratory assistance, and Diane Nelson and an anonymous reviewer for helpful suggestions on an earlier draft.

Ancillary