Seascape genomics identify adaptive barriers correlated to tidal amplitude in the shore crab Carcinus maenas

Abstract

Most marine invertebrates disperse during a planktonic larval stage that may drift for weeks with ocean currents. A challenge for larvae of coastal species is to return to coastal nursery habitats. Shore crab (Carcinus maenas L.) larvae are known to show tidal rhythmicity in vertical migration in tidal areas and circadian rhythmicity in microtidal areas, which seems to increase successful coastal settlement. We studied genome‐wide differentiation based on 24,000 single nucleotide polymorphisms of 12 native populations of shore crab sampled from a large tidal amplitude gradient from macrotidal (~8 m) to microtidal (~0.2 m). Dispersal and recruitment success of larvae was assessed with a Lagrangian biophysical model, which showed a strong effect of larval behaviour on long‐term connectivity, and dispersal barriers that partly coincided with different tidal environments. The genetic population structure showed a subdivision of the samples into three clusters, which represent micro‐, meso‐ and macrotidal areas. The genetic differentiation was mostly driven by 0.5% outlier loci, which showed strong allelic clines located at the limits between the three tidal areas. Demographic modelling suggested that the two genetic barriers have different origins. Differential gene expression of two clock genes (cyc and pdp1) further highlighted phenotypic differences among genetic clusters that are potentially linked to the differences in larval behaviour. Taken together, our seascape genomic study suggests that tidal regime acts as a strong selection force on shore crab population structure, consistent with larval behaviour affecting dispersal and recruitment success.

1. INTRODUCTION

Local adaptation is the fine‐tuning of populations to their local environment via natural selection (Savolainen et al., 2013). Selective forces driven by environmental differences result in locally adapted populations with resident genotypes having higher fitness in their native habitat (Kawecki & Ebert, 2004). However, the diversifying effects of selection will be opposed by the homogenizing effect of gene flow (Nosil, 2009). In the marine environment, where seascape genetics and genomics have only recently helped “to navigate the currents” (Riginos et al., 2016), microgeographical adaptation (adaptation at a spatial scale smaller than dispersal; sensu Richardson et al., 2014) seems surprisingly common despite the high gene flow potential of many marine species (Sanford & Kelly, 2011). For instance, locally adapted populations have been described in microalgae and fish which have planktonic propagules that disperse passively for weeks in the water column (Barth et al., 2019; Sjöqvist et al., 2015). One of the potential processes driving this differentiation is purifying selection, which acts in each generation following dispersal. Purifying selection, also referred to as “phenotype–environment mismatch” (Marshall et al., 2010), can maintain allelic differences despite high gene flow. Further, the presence of genomic structural variation reducing recombination between co‐adapted alleles might favour the maintenance of local adaptation despite high gene flow (Barth et al., 2019; Cayuela et al., 2020; Le Moan et al., 2021). Alternatively, secondary contact after past isolation might facilitate the initial increase in frequency of adaptive variation despite gene flow (Ravinet et al., 2017; Rougemont et al., 2017). Finally, plasticity in gene expression is also an important factor in local adaptation (Kenkel & Matz, 2016). Phenotypic plasticity can provide a short‐term buffer during longer‐term genetic adaptation (Munday et al., 2013). The flip side of plasticity is that selection may be too weak to winnow less fit individuals, and therefore no gradual adaptation to a changing environment that may finally surpass the range of plasticity (Ho & Zhang, 2018). In contrast to these processes driven by selection, there also may be bias in gene flow when genetic variation leads to intraspecific variation in behaviour linked to dispersal, generating gene flow that is nonrandom with respect to individual variation (Edelaar & Bolnick, 2012). Any nonrandom gene flow can lead to local adaptation or it can lead to patterns that resemble local adaptation, but do not involve any fitness differences and/or evolutionary processes (i.e., natural selection). Looking at the question from the other side of the coin, there is also the issue “when is dispersal for dispersal?” (Burgess et al., 2016). The relationship between dispersal and fitness may not be as simple and direct as often assumed, and in fact in the marine environment dispersal has also often been viewed as an incidental by‐product of traits selected for other functions (Bonhomme & Planes, 2000).

The European shore crab (Carcinus maenas L.) is particularly well suited for assessments of the relationship between dispersal and adaptation, because of its particular life history and its long history as an object of scientific study including dispersal. The shore crab is also of global concern since it has successfully invaded several continents (Grosholz & Ruiz, 1996) and poses an ecological threat in many coastal areas (e.g., Pickering et al., 2017). Shore crabs have a lifespan of 3–5 years and a generation time of 2–3 years (Crothers, 1967). As adults they are mostly solitary and adult dispersal is limited to tens of kilometres. Larval dispersal is, however, possible—at least potentially—over large geographical distance as the larvae drift for 4–10 weeks in the water column depending on the water temperature (Dawirs, 1985; Mohamedeen & Hartnoll, 1989; Yamada et al., 2015). The population genetics/genomics of C. maenas are relatively well studied, but mostly focus has been on parts of the introduced range along the west Atlantic coast (Darling et al., 2014; Jeffery, DiBacco, Wringe, et al., 2017; Jeffery, DiBacco, Van Wyngaarden, et al., 2017; Roman, 2006; Tepolt & Palumbi, 2015). Interestingly, shore crab larvae in the native range show a cline in larval behaviour (vertical migration) along a gradient in tidal influence from macrotidal (>4‐m tidal range) areas along the British Isles to microtidal areas (<1‐m tidal range) in the Skagerrak area of the eastern North Sea (Moksnes et al., 2014; Queiroga et al., 2002). In meso‐ and macrotidal areas, shore crab larvae display an inherited endogenous vertical migration rhythm (Duchêne & Queiroga, 2001; Zeng & Naylor, 1996c), which when synchronized with local tides is believed to facilitate cross‐shelf transport and recruitment success (Queiroga, 1998; Zeng & Naylor, 1996a). Importantly, this tidal rhythmic behaviour was even displayed by newly hatched larvae from mothers that had been kept in a constant environment for months, but originally came from a macrotidal area (Zeng & Naylor, 1996c), suggesting that this behaviour is heritable (Zeng & Naylor, 1996b). This endogenous tidal rhythm is not present in larvae from microtidal areas, where instead larvae show an exogenous circadian rhythm with migration to deeper waters during the day, possibly reducing predation (Moksnes et al., 2003, 2014; Queiroga et al., 2002). As the velocity and direction of coastal ocean currents often vary with depth, the vertical position of pelagic larvae may critically affect their dispersal (Corell et al., 2012; Moksnes et al., 2014), and this adaptation in behaviour is therefore hypothesized to be an optimal strategy between avoiding inshore stranding, enhancing offshore dispersal of the larvae, predator avoidance and larval recruitment to shallow nursery areas (Moksnes et al., 2014).

Here we hypothesize that tidal regime exerts a strong selection pressure for different vertical migration behaviours of C. maenas larvae, which is probably modulated by gene expression differences in clock genes. We further hypothesize that larvae with the “wrong” tidal behaviour will have reduced recruitment success, which restricts along‐coast dispersal resulting in genetic population differentiation in an isolation‐by‐distance manner and between geographical areas with different tidal regimes. To address these hypotheses, we sampled 12 populations along a tidal gradient in the North Sea from macro‐ to microtidal conditions, and used a combination of approaches to examine (i) dispersal distances and connectivity of the shore crab metapopulation based on biophysical modelling, (ii) historical and contemporary patterns of gene flow inferred from ~24,000 single nucleotide polymorphisms (SNPs) generated with 2b‐RAD sequencing and a demographic model, and (iii) gene expression differences of putative clock genes.

2. MATERIAL AND METHODS

2.1. Sampling

Guided by observed behavioural differences between crab larvae of macro‐ and microtidal environments (Moksnes et al., 2014) and a previous genetic assessment of Carcinus maenas in central Europe (Roman & Palumbi, 2004), we carried out sampling at 12 sites (Figure 1) in September and October 2017 along a tidal cline from macrotidal (>4 m) in Wales (Bangor: 8.6 m) and the English Channel (Dunkirk: 7.0 m) to microtidal (<2 m) in the Kattegat (Odense: 0.22 m, www.tide‐forecast.com, Table 1). One site in the mesotidal area, Hvide Sand, is experiencing a low tidal amplitude due to an artificial barrier (Table 1). The sites span a geographical range of ~2000 km sea‐distance between Bangor and Odense. Sampling was performed at similar geographical distances among sites and males and females were collected in equal proportions at each site. For genotyping, 10 males and 10 females were assessed at each of the 12 sites, whereas for gene expression analyses, four males and four females were used from only six of the sites, including the full tidal gradient (Figure 1). The crabs were dredged (Bangor, Yerseke, Esbjerg, Hvide Sand and Sylt), caught in fyke nets (Texel) or caught in baited traps (Dunkirk, Hooksiel, Tyboron, Kristineberg, Varberg and Odense). For genotyping, one leg from each living individual was removed, transferred to ethanol and immediately stored at −20°C. For gene expression analyses, leg muscle tissue was directly transferred to RNAlater (Ambion, Life Technologies), kept overnight at 4°C, transported at −20°C and finally stored at −80°C until RNA extraction. Crabs were released at the site of capture when possible. We additionally sampled shore crab megalopae at the Kristineberg site (58.259°N, 11.450°E; Figure 1) in the morning of August 22, 2019 with a plankton net (500 µm) at 15–20 m depth to verify candidate gene expression in larvae.

FIGURE 1.

Map of Carcinus maenas sampling locations in the North Sea. Site acronyms with a star have been included in the gene expression analysis. Acronyms are as follows: O, Odense; V, Varberg; K, Kristineberg; T, Tyboron; I, Hvide Sand; E, Esbjerg; S, Sylt; H, Hooksiel; X, Texel; Y, Yerseke; D, Dunkirk; B, Bangor. Red lines indicate distinctions between micro‐ (<2 m) in the Skagerrak and Kattegat, meso‐ (2–4 m) in the Wadden sea and macrotidal (>4 m) in the English channel according to Jänicke et al. (2020). Red dotted lines indicate “adaptive barriers” as identified in Figure 4. Turquoise, purple and green dotted lines indicate barriers to dispersal identified from the multigenerational stepping‐stone biophysical modelling for the three behaviours: surface dispersal (green barrier), circadian behaviour (purple) and tidal behaviour (turquoise)

TABLE 1.

Information on sampling sites of Carcinus maenas included in the analysis and sorted along a geographical line from the Kattegat to the Irish Sea

Site	Acronym	Latitude (°N)	Longitude (°E)	Sampling date and time for gene expression (for those sampled)	Tidal range (m)
Odense	O	54.9849	10.6329	Afternoon October 9, 2017	0.22
Varberg	V	57.1108	12.2392		0.16
Kristineberg	K	58.2499	11.4464	Morning October 6, 2017	0.34
Thyborøn	T	56.7062	8.2209	Morning September 30, 2017	2.14
Hvide Sand	I	56.0044	8.1296		1.15
Esbjerg	E	55.4896	8.4107		2.15
Sylt	S	55.0404	8.4591		2.1
Hooksiel	H	53.6436	8.0839		3.35
Texel	X	52.9968	4.7766	Midday September 26, 2017	2.65
Yerseke	Y	51.488	4.0578		3.24
Dunkirk	D	51.0464	2.3744	Morning September 25, 2017	7.03
Bangor	B	53.2265	−4.1589	Morning October 23, 2017	8.56

2.2. Biophysical model of multigenerational larval dispersal

We used a biophysical model to explore the effects of different larval behaviour strategies on multigenerational dispersal distance and to identify possible barriers along the tidal gradient from the English Channel to the Kattegat area. Dispersal trajectories were simulated with a Lagrangian particle‐tracking model (TRACMASS, Vries and Döös, 2001) driven off‐line with flow fields (3‐hr resolution) from an ocean circulation model (NEMO‐Nordic, Hordoir et al., 2019). The model has a horizontal spatial resolution of 3.7 km, 84 vertical levels, a free surface and allows the grid boxes to stretch and shrink vertically to model the tides without generating empty grid cells at low tide. To mimic larval vertical behaviour, the vertical position of the trajectories was locked at predetermined depths, changing with tidal, circadian and ontogenetic periods (Moksnes et al., 2014).

We included three different larval behaviours in this modelling effort (tidal, circadian and surface). Particles, simulating larvae, were released from 1360 model grid cells spanning the tidal cline from the English Channel to the Kattegat. Grid cells adjacent to land were assumed to include shallow‐water habitat typical for adult shore crab and suitable for spawning and settling; a few offshore grid cells, with a mean depth ≤6 m, were also included. Release of a total of 600,000 particles occurred at the end of June to coincide with a peak recruitment period of juvenile crabs in July and August in Northern Europe (Breteler‐Klein, 1976; Moksnes, 1999). The release was repeated for three years (1989, 1996 and 2003), representing two opposite extremes and one neutral year in the North Atlantic oscillation cycle (Hurrell & Deser, 2010). The number of release time points between and within years was a compromise in the modelling effort. In a previous study of dispersal of C. maenas larvae (Moksnes et al., 2014), we showed that dispersal metrics (distance to shore after dispersal and recruitment success) mainly depended on larval behaviour and the geographical location of release. Release time played a minor role, with variation between years more important than between months during the reproductive season. In this study of the full connectivity between shallow areas we therefore only included interannual variation. The dispersal simulations included a fixed pelagic larval duration of 40 days, consisting of 25 days of zoeal stages and 15 days of the megalopal phase (Mohamedeen & Hartnoll, 1989). Trajectories of the tidal behaviour shifted depths according to the selective tidal stream transport hypothesis (Queiroga et al., 1994), and were located at the surface (0.3 m depth) during ebbing tides and at 20 m depth during flooding tides for the first 25 days. For the last 15 days of the larval phase, the trajectories were given the opposite vertical migration pattern. The tidal migration was synchronized with the principal lunar semidiurnal tidal constituent (M2; 12.42‐hr period), which is the dominant tidal signal in the North Sea (Hill, 1995). The tidal behaviour further changed its rhythm to be in phase with the local tide. Trajectories for the circadian behaviour shifted depth according to the sea‐breeze hypothesis (Shanks, 1995), and were located at the surface (0.3 m depth) during night and twilight conditions and at 20 m depth at daylight hours for the first 25 days with the opposite pattern for the last 15 days. Particles that dispersed according to the surface behaviour drifted at 0.3 m depth during the entire larval duration of 40 days.

From these trajectories the dispersal probabilities between all 1360 grid cells were calculated and summarized in a connectivity matrix (Jonsson et al., 2020). We also calculated the multigeneration connectivity where stepping‐stone dispersal is allowed over many single‐generation dispersal events, and summed over all possible dispersal routes (White et al., 2010). Such multigeneration connectivity may be used to infer the long‐term connectivity between populations (e.g., exploring barriers to gene flow) (Jahnke et al., 2018). Multigeneration connectivity was calculated by multiplication of the connectivity matrix across 64 generations, which seemed an appropriate time span to potentially disperse from the English Channel to the Kattegat (~1200 km).

2.3. DNA extraction and 2b‐RAD library preparation

DNA was extracted with RNAse treatment, and four out of 20 individuals from each locality were also used as technical replicates (i.e., replicated extraction, library preparation and sequencing). DNA extraction was followed by an ethanol/isopropanol precipitation. DNA quantity and quality were assessed using a Qubit ds DNA BR or HS AssayKit (Invitrogen–ThermoFisher Scientific) and on a 1% agarose gel.

2b‐RAD libraries (Wang et al., 2012) were prepared following a laboratory protocol available at https://github.com/z0on/2bRAD_denovo, using the restriction enzyme BcgI. Libraries were individually barcoded, and fragment selection was performed by excising the amplicon band from an agarose gel. Gel fragments were cleaned using a MinElute Gel Extraction Cleaning Kit (Qiagen) and pooled equimolarly (as assessed by the Qubit ds DNA HS AssayKit) into population sets of 24 individuals (20 individuals plus four technical replicates) per sequencing lane. In total, 240 crab samples from 12 sites and 48 technical replicates were sequenced in 12 libraries on the Illumina HiSeq2500 platform, generating 50‐bp single‐end sequences, at the Science for Life Laboratory (SciLifeLab)—Genomics, SNP&SEQ Technology Platform in Uppsala University, Sweden.

2.4. Bioinformatics pipeline

Bioinformatics were performed using the computer cluster “Albiorix” at the University of Gothenburg, Sweden. The analysis followed a modified de novo pipeline available at https://github.com/z0on/2bRAD_denovo. Reads were “demultiplexed” based on barcodes. Restriction sites were trimmed off, followed by quality filtering using the fastx‐toolkit (stating that 100% of bases should have a quality score of at least 20, which means a 1% error rate). After trimming and quality filtering, a total of 1,352,415,682 reads were retained across the 12 libraries. Individual trimmed fastq files were merged to collect tags found in at least two individuals with a minimum depth of five for genotyping. Reads that had more than seven observations but lacked reverse complement reads were removed. Tags were then clustered with cd‐hit (Li & Godzik, 2006) allowing for up to three mismatches, followed by the creation of a “reads‐derived reference” based on 30 fake chromosomes. Individual trimmed fasta files were mapped back to the references. Then, genotyping was performed using The Genome Analysis Tookit version 3.1‐1 (GATK) (McKenna et al., 2010). A first round of putative variants was generated using GATK’s UnifiedGenotyper, followed by base quality score recalibration (BQSR/BaseRecalibrator and PrintReads) based on a high‐confidence (>75th quality percentile) SNP set. The realigned and recalibrated reads were used to generate a second round using UnifiedGenotyper. We then used the variant quality score recalibration (VQSR) step, to generate an adaptive error model using the SNPs that were reproducibly genotyped across the technical replicates. When applying the recalibration we chose the tranche with 95% truth sensitivity, which had a Ti/Tv ratio of 1.784, to call all SNPs from the overall data set. The final filtration step was performed in vcftools (Danecek et al., 2011) to remove under‐sequenced samples (set to fewer than 95 000 sites, which removed three individuals) and to select biallelic loci genotyped in at least 90% of individuals, and with a maximum heterozygosity of 50%. This resulted in a final data set of 81,389 loci for 240 unique individuals.

2.5. Population genetic analyses

2.5.1. Population structure

Population genetic analyses were performed on the Rackham cluster of the Swedish National Infrastructure for computing (SNIC). The final data set was thinned in order to keep one SNP per RAD locus with maximal allele frequency (script thinner.pl with criterion = maxAF, https://github.com/z0on/2bRAD_denovo), resulting in 24,273 loci. Individual genetic variation was visualized by a principal component analysis (PCA) using the R package adegenet (Jombart, 2008). bayescan (Foll, 2012) was used to identify putative loci under selection with prior odds set to 100, using a relaxed threshold of q = 0.5 and a stringent threshold at q = 0.00011 (see Figure S1 for bayescan plot). In an attempt to annotate the 128 outlier loci, we first blasted the short 36‐bp 2bRAD fragments against the C. maenas Transcriptome Shotgun Assembly (TSA), and then used the blastn algorithm to query the longer‐sequence TSA hits. Sea distances without crossing land were calculated in the R package marmap (Pante & Simon‐Bouhet, 2013). Pairwise Weir and Cockerham's F _ST and linkage disequilibrium were calculated among sites using vcftools (Danecek et al., 2011) for outlier and putatively neutral loci. We explored the variation of allele frequency across geographical distance for the outlier loci using cline model fitting with the R package HZAR (Derryberry et al., 2014). For each outlier, we calculated the slope of the cline by calculating the differences in allele frequency estimated by the model on each side of the cline, and dividing this difference by the cline width. The slope values were then represented against the position of the cline centres estimated from the model using the R package ggplot2 (Wickham, 2016). We tested for isolation‐by‐distance (IBD) between the pairwise F _ST (of all loci, neutral loci and outlier loci) and sea distance as well as dispersal probability obtained with the biophysical model using Mantel tests with the R package ncf (Bjornstad, 2009) with 100,000 replicates. For visualization, connectivity networks based on F _ST (neutral and outlier loci) and biophysical dispersal probability (three behaviours: surface drifting, migrating in rhythm with day/night or in rhythm with the tides) were drawn using the R packages igraph (Csardi & Nepusz, 2006) and popgraph (Dyer, 2014). All edges are shown for estimated dispersal probabilities, while for F _ST several cut‐offs were explored, and those that resulted in still connected networks are shown (similar to the percolation threshold). For IBD and network analyses, we removed the site Bangor from the analysis, as it lies beyond the oceanographic model domain.

2.5.2. Demographic history

In addition to the contemporary connectivity patterns, we evaluated historical population connectivity using the software moments (Jouganous et al., 2017). Inferences were conducted between two pairs of sampling sites representative of the main genetic clusters: Dunkirk (D, macrotidal) vs. Sylt (S, mesotidal), and Sylt vs. Kristineberg (K, microtidal, Figure 1). The data were filtered to keep one random SNP per RAD tag with a minor allelic count of at least two among the three geographical sites. The folded version of the Site‐Frequency‐Spectrum (SFS) was used to compare contrasted models of demographic history. Four basic models of population splits were compared: Strict Isolation (SI), Isolation with Migration (IM), Ancestral Migration (AM) and Secondary Contact (SC). These models were further tuned to include changing effective population sizes either in the ancestral population (model “ac” for ancient change) or, after the split, in one of the daughter populations (model “rc” for recent change). In addition, models of heterogeneous effective population size (model “2N”) and heterogeneous migration rate (model 2 m in the model with gene flow: IM, SC, AM) resulting from the effect of selection at linked sites and intrinsic barriers to gene flow, were included (Table S1). The models were adjusted to the data following the three‐step optimization procedure developed by Portik et al. (2017) adapted to the software moments by Momigliano et al. (2021). The fit of each model was replicated 10 times to control for parameter convergence. In total, the fits of 44 demographic scenarios were compared using Akaike's information criterion (AIC) value, the Δ_AIC (AIC_best−AIC _i ), and the weighted AIC (see Rougeux et al., 2017). The estimates of divergence time were converted into years following Rougeux et al. (2017) using a mutation rate of 1 × 10⁻⁸ per generation, and a generation time of 3 years (Crothers, 1967).

2.6. Candidate gene approach for gene expression analyses

Using a candidate gene approach, we also looked at gene expression differences of insect candidate clock genes, probably involved in the differing crab larvae behaviour between tidal and atidal sites. We extracted RNA from ~100 mg of muscle tissue from eight individuals (four males and four females) each at six sites with the RiboPure kit (Life Technologies) according to the manufacturer's instructions. For the extraction of larval RNA, 50 megalopae were combined. We used the high‐capacity cDNA Reverse transcription kit (Applied Biosystems) with 500 ng of RNA as input material to synthesize cDNA according to the manufacturer's instructions.

Searches for C. maenas clock genes were conducted using a protocol previously applied to discover circadian genes/proteins in other crustaceans (Biscontin et al., 2017; Christie et al., 2018). First, tblastn (http://blast.ncbi.nlm.nih.gov/Blast.cgi) was used to search through the “Transcriptome Shotgun Assembly (TSA)” database of “Carcinus (taxid: 6758).” Known circadian proteins, primarily those from the fruit fly Drosophila melanogaster, were used as query sequences in tblastn (15 circadian proteins from D. melanogaster and two circadian proteins from Danaus plexippus). In addition, we selected four genes that have been suggested to be tidally linked in the limpet Cellana rota (Schnytzer et al., 2018; see Table S2). Primers for gene expression analyses with quantitative real‐time polymerase chain reaction (qPCR) were designed with primer3 (http://primer3.ut.ee). Design conditions included primer length (18–23 bp), T _m (~60°C), GC content (≥50%) and a product size of 150–200 bp. In qPCRs we used the Fast Sybr Green Master mix (Applied Biosystems) as per the manufacturer's instructions and a thermal profile of 95°C for 20 s, followed by 40 cycles of 95°C for 1 s and 60°C for 20 s. All qPCRs were conducted in technical triplicates and each assay included three no‐template negative controls (NTCs). Primer sequences, percentage efficiencies (E) and regression coefficients (R ²) of the candidate and reference genes are reported in Table S3. Three reference genes, elongation factor (cm‐ef2), elongation factor‐1 (camelf1a) and Ubiquitin‐conjugating enzyme E2 L3 (camubce2) (Abuhagr et al., 2014; Alexander et al., 2018), were selected to normalize the transcription profiles of the 21 candidate genes for each sample. stepone software version 2.3 was used to calculate raw threshold cycle (Ct) values and to analyse the endogenous control genes. To analyse differences in gene expression among samples (n = 48) and sites (n = 6), the sample maximization method according to Hellemans et al. (2007) was established. The significance of gene expression differences among populations and clusters was tested with the rest 2009 software (Pfaffl et al., 2002), followed by Benjamini–Hochberg false discovery rate corrections (Benjamini & Hochberg, 1995). Linear mixed models (LMMs) were performed using the function “lmer” in the lme4 package in R (Bates et al., 2014), with tidal gradient (scored 1–6), latitude, sampling date (days after equinox) and sampling time (morning, midday or afternoon) as fixed effects that were fitted to the dependent variable (normalized relative quantity [NRQ] of each of the 21 candidate genes separately). As a random effect, we had intercepts and random slopes for sex at each site (12 levels). Visual inspection of residual plots did not reveal any obvious deviations from homoscedasticity or normality. By examining backward reduced random‐effect tables generated with the package “lmerTest” (Kuznetsova et al., 2017), we found that latitude and sex at each site were not a significant factor for gene expression differences in any of the candidate genes (Table S4). These two factors were removed for the final full model, which used tidal gradient, sampling time and sampling date as fixed factors using “lm” of the stats package. A top‐down approach was then used for each candidate gene separately to compare the fit of the fully fitted model to (i) a model without the tidal gradient, (ii) without date and (iii) without sampling time, with analysis of variance (ANOVA). This analysis allowed us to identify candidate genes that exhibited a significant transcriptional response to the three factors that might affect circadian clock gene expression. To assess trends of overall gene expression, we performed a PCA to reduce the data based on all candidate genes, all clock genes, core clock genes, clock‐associated genes and putative tidal cyclers. Finally, we used the PC1s as input for lm for overall changes in gene expression of each gene group. A heatmap was created with the gplots R package (Warnes et al., 2009) using population NRQ values standardized to the microtidal population from Odense and the genes cyc and pdp1.

3. RESULTS

3.1. Biophysical model of multigenerational stepping stone larval dispersal

The biophysical model of multigeneration, stepping‐stone dispersal predicted some potential barriers to gene flow in the North Sea–Skagerrak–Kattegat transition. The number, strength and geographical position of potential barriers depended on the simulated vertical larval behaviour (Figure 1; Figure S2). Larvae drifting in surface water during the full 40‐day period showed the highest level of connectivity, and stepping‐stone dispersal over 64 generations predicted only one asymmetric barrier on the mid‐Dutch coast (green barrier in Figure 1; Figure S2). Stepping‐stone dispersal for the circadian behaviour showed a strong barrier on the south Jutland coast (purple barrier in Figure 1), and a few weaker, asymmetric barriers along the Dutch, Belgian and French coasts (Figure S2). The model of tidal vertical behaviour predicted three strong barriers (Figure S2) on the north Jutland coast, in the German Bight and on the mid‐Dutch coast (turquoise barriers, Figure 1).

3.2. Seascape genomic analyses

The first two axes of the PCA with all 24,273 loci showed 2.23% of the total inertia and distinguished three genetic clusters (Figure 2a). PC1 (1.26%) separates individuals sampled in the microtidal area of the Kattegat and Skagerrak (O, V and K), while PC2 (0.96%) shows differentiation between the individuals from the mesotidal sites of the North Sea, and the macrotidal sites from Bangor and Dunkirk (B and D). bayescan identified 500 putative outlier loci at q = 0.5 and 128 outlier loci at q = 0.00011 (see Figure S1). These putative outlier loci represent 2% and 0.5% of all loci. Removing those outlier loci led to weaker differentiation among sites (Figure 2b), while the outlier loci also drew the same three distinct clusters (Figure 2c). Pairwise F _ST values among sites were low, ranging between 0 and 0.02 based on all loci (see Figure S3) and between 0 and 0.009 for neutral loci. In contrast, F _ST ranged between 0 and 0.363 for the 128 top outlier loci. F _ST was highest between the microtidal and the macrotidal cluster, which are also the most geographically distant sites. We did not find any differentiation related to sex (Figure S4). IBD between F _ST and sea distance was significant for all data sets (all loci: r ² = 0.81, neutral loci: r ² = 0.76 and outlier loci: r ² = 0.82, all p < 0.0001). There was no significant relationship between F _ST and multi(64)‐generational dispersal probability from the biophysical modelling for any of the three behaviours. The network analyses based on F _ST were in line with the PCAs, showing no clear patterns of population structure for neutral loci, but three clusters for outlier loci (Figure 3a,b). The networks for the biophysical dispersal probabilities among the sampling sites showed large differences in connectivity estimates among the three dispersal behaviours with no clear structure for the surface drifting larvae, but three clusters for larvae with tidal behaviour (Figure 3c–e).

FIGURE 2.

Principal component analyses for (a) all 24,273 loci (PC1: 1.26% inertia, PC2: 0.96% inertia), (b) all neutral loci, that is after removal of potential outlier loci identified in bayescan with very relaxed q = 0.5 (n = 500; PC1: 0.84% inertia, PC2: 0.8% inertia), and (c) the 128 putative outlier loci identified in bayescan with q = 0.00011 (PC1: 14.59% inertia, PC2: 10.84% inertia). Site acronyms and coloration as in Figure 1

FIGURE 3.

Networks of genetic distance (based on F _ST [a and b]) and oceanographic distance (based on dispersal probability from the biophysical model over 64 generations [c–e]) constructed for 11 sampling sites (Bangor was not included as it falls out of the oceanographic model domain) of Carcinus maenas in the North Sea. Lines indicate connectivity in terms of gene flow (a and b) and dispersal probability (c–e) and nodes (dots) are shown at their geographical location

The genetic cline analysis of the 128 outlier loci identified barriers to gene flow with obvious clustering of loci with steeper allelic clines at two geographical locations: ~1000 km (between the geographically close sites of Dunkirk and Yerseke) and 1700 km (separating the Skagerrak from the North Sea) from Bangor (Figure 4). These two barriers to gene flow closely reflect the separation into micro‐, meso‐ and macrotidal areas (Figure 1). An LDna analysis (Kemppainen et al., 2015) did not indicate the presence of genomic islands nor did it indicate that the loci at the tidal breaks shown in Figure 4 are in linkage disequilibrium (Figure S5). The number of markers showing evidence of clear allele frequency clines was higher along the transition from meso‐ to microtidal (15 loci with slope >0.002) than from macro‐ to mesotidal (four loci). Interestingly, demographic inferences suggested that different demographic histories were involved during the formation of the two barriers (Table 2 and Figure 5). Inferences performed between the meso‐ vs. microtidal sites showed relatively strong support for a secondary contact (SC) model including an expansion in the ancestral population and heterogeneous effective size across the genome (SC_ac_2N, with W _AIC =0.97, Table 2). The estimated contact period (~14 thousand years ago in the best model) was six times shorter than the isolation period (estimated to be 96,000 years, Table 2). In contrast, the SC model (including a recent expansion and heterogeneous effective population size across the genome, SC_rc_2N) and the isolation with migration (IM) model (including both ancient and recent expansion and heterogeneous effective population size or migration rate along the genome, IM_ac_rc_2N and IM_ac_rc_2 M) had a similar support for the meso‐ vs. macrotidal sites (Δ_AIC < 2 for the three best models, Table 2). However, the parameter estimated for the SC model in the meso‐ vs. microtidal comparison were biologically unexpected (long divergence time >1 million years or isolation phase 50 times shorter than the secondary contact). Thus, an IM model is more likely to explain the divergence between the meso‐ vs. macrotidal site (Figure 5a). The two IM models differ only in the way genomic heterogeneity is captured and estimated similar parameter values (Table S1).

FIGURE 4.

Geography of the genetic clines based on the slopes and centres of the allele frequencies of the 128 putative outlier loci as a function of sea distance from the site Bangor (B), the site with the highest tidal range. The dashed lines indicate the identified two barriers that separate macrotidal sites (B and D), mesotidal sites (Y, X, H, T, I, S, E) and the microtidal sites in the Kattegat/Skagerrak (K, V, O). Geographical locations of the sampling sites are approximate. Site acronyms are as in Figure 1. The inset panels show the clines of the loci with the strongest slopes at both barriers over the same geographical distance as shown in the main panel. The colour of the clines indicates the respective slope of the outlier locus in the same colour

TABLE 2.

Summary of the demographic inferences showing the three best demographic histories inferred between populations from macro‐ vs. meso‐, and meso‐ vs. microtidal sites

Data set	Model	likeli.	AIC	ΔAIC	WAIC	P	TS	TSC
Macro vs. Meso	SC_rc_2N	−591	1201	0	0.480	0.23	232	10,864
Macro vs. Meso	IM_ac_rc_2N	−591	1202	1	0.177	0.29	108	—
Macro vs. Meso	IM_ac_rc_2m	−590	1202	1	0.177	0.16	116	—
Meso vs. Micro	SC_ac_2N	−617	1255	0	0.973	0.34	95	16
Meso vs. Micro	IM_ac_rc_2N	−619	1259	4	0.018	0.36	123	—
Meso vs. Micro	SC_ac_rc_2N	−619	1260	5	0.007	0.25	83	14

The table shows, in order of appearance, the population pairs, the best model (IM = isolation with migration, SC = secondary contact, rc and ac = recent and ancient change in effective size, 2N = heterogeneous effective size along the genome, 2m = heterogeneous migration rate along the genome); the AIC of the model, the difference of AIC with the best model (Δ_AIC), the weighted AIC (W _AIC), the proportion of the genome affected by low migration rate or low effective size (P), the time of split (T _s), and the time of secondary contact (T _sc) given in thousands of years. Other parameters estimated by the models are available in Table S1.

FIGURE 5.

Result from the most likely scenario associated with the population divergence between (a) the macro‐ vs. mesotidal environment, and (b) meso‐ vs. microtidal environment. The graphs on each rows show the observed SFS, the schematic representation of the best demographic scenario (IM_ac_rc_2N: isolation with migration with change in effective size in ancestral and derived populations and with heterogeneous effective size along the genome; SC_ac_2N: secondary contact with change in effective size in the ancestral population and heterogeneous effective size along the genome), the fitted SFS and the residual of the model (containing the distribution of the residuals)

In our blast search of the 36‐bp fragments of all 128 putative outlier loci, we only encountered 10 matches on the Carcinus maenas TSA database, and blasting these longer contigs resulted in 41 blastn hits (Table S5). The hits were for five crustacean species and for a parasitic barnacle of the European shore crab (Sacculina carcini). Only a few predicted genes, all from Homerus americanus, were encountered. This is perhaps not surprising, as there are few genome references for crustaceans. One of these genes was mucin‐3A‐like, a glycoprotein component of a variety of mucus gels in humans, which may also be involved in ligand binding and intracellular signalling (UniProt). Other annotated outlier genes include the transposable element‐derived protein 4‐like piggyBac, a WD repeat‐containing protein 19‐like, and most interestingly two sterol O‐acyltransferase 1‐like transcripts. This acyl‐transferase is also called ACAT1, and is one of the candidate genes we tested below, as differential expression of this genes has been shown in Cellana rota in a circa‐tidal manner. The potential function of interest may be gravity impact (Casey et al., 2015; Frigeri et al., 2008; Schnytzer et al., 2018). However, the primer pair designed to amplify C. maenas acat1 did not bind to any of the 10 TSA outlier sequences.

3.3. Gene expression of putative clock genes

We were able to identify—for the first time in C. maenas—the orthologues of most clock components known from Drosophila as well as from other crustaceans. We identified and verified 21 orthologues of clock‐related proteins, corresponding to insect core clock, clock‐associated and clock input proteins (see Table S2). We found an “ancestral clock” in C. maenas with both cry1 and cry2 present in the transcriptome and expressed in crabs at all locations and in both megalopae and adults.

After Benjamini–Hochberg corrections, four out of the 21 candidate genes were significantly up‐ or down‐regulated in at least one pairwise comparison of the six assessed sites (see Table S6). The linear model revealed a significant effect of the tidal gradient on gene expression of two genes: pdp1 and cyc (Table S7). Both genes were generally down‐regulated with increasing tidal amplitude and their patterns are very similar (Figure 6).

FIGURE 6.

Effect of tidal amplitude on gene expression. Relative gene expression (normalized relative quantity [NRQ], standardized to individual D7F, a female from Dunkirk with generally low gene expression of many candidate genes) of genes significantly affected by the tidal gradient for (a) pdp1 and (b) cyc along the tidal gradient from microtidal to large tidal amplitudes, and (c) heatmap of relative gene expression (NRQ, standardized to microtidal population from Odense) for both genes at all sites, assigned to their micro‐, meso‐ or macrotidal location. Red indicates high expression. D, Dunkirk; B, Bangor; X, Texel; T, Thyborøn; K, Kristineberg; O, Odense

Other genes were significantly affected by the sampling date expressed in days after equinox (used as the reference, as at this date all sites experienced 12 h of daylight), and sampling time of the day (Table S7). The experiment was not designed to test for seasonal and daily expression patterns, but we have accounted for them in our linear model, and the identified genes may be interesting for further investigations into the circadian clocks in C. maenas. In a data‐reduction approach based on PCAs, tidal gradient had significant effects on expression of the core clock genes comprising tim, cyc, clk, cry2 and per (Table S8, Figure S6).

4. DISCUSSION

We set out with the hypothesis that tidal regime is a strong selection force on shore crab population structure, mostly driven by the fact that shore crab larvae show different vertical migration behaviours depending on the tidal amplitude. Vertical migration of Carcinus maenas larvae occurs with an endogenous tidal rhythm in tidal areas, and an exogenous circadian rhythm in microtidal areas (Moksnes et al., 2014; Queiroga et al., 2002). This tidal rhythm is heritable (Zeng & Naylor, 1996b), and we hypothesize that it may be regulated by clock genes, which are well described in insects and have also been found in some crustaceans (Biscontin et al., 2017; Christie et al., 2018). This behaviour has an impact on dispersal potential because according to the selective tidal stream transport hypothesis, vertical migration between layers of opposite current directions is thought to facilitate cross‐shelf transport and increase return of larvae to shallow nursery habitats (Duchêne & Queiroga, 2001; Zeng & Naylor, 1996a). Thus, “wrong” tidal behaviour may reduce recruitment success and thereby restrict dispersal between areas with different tidal regimes.

In our study we were able to show with oceanographic modelling of multigeneration, stepping‐stone dispersal that larval vertical migration had a large effect on predicted long‐term connectivity in the assessed geographical area. We found barriers to dispersal for all three modelled behaviours, with a barrier in the eastern Wadden Sea for both tidal and circadian behaviour, and with an additional barrier for tidal migration in the western Wadden Sea and at the border between the North Sea and Skagerrak (Figure 1).

Similar to previous population genetic assessments of the European and Mediterranean shore crab (Pascoal et al., 2009; Schiavina et al., 2014) we found weak F _ST (F _ST = 0–0.009 for neutral loci). Despite little population structure after removing 500 putative outliers (2% of all loci), we were able to detect a signal of IBD, suggesting that local recruitment patterns contribute to genetic differentiation in neutral loci. However, we did not find significant isolation‐by‐oceanography (IBO) based on multiple generation (64) dispersal probability for any modelled behaviour. Instead, the biophysical modelling indicates several strong barriers to dispersal (Figure 3) rather than a linear gradient in connectivity.

Using all ~24,000 SNPs and the 128 outlier loci, the shore crab samples clustered into three distinct groups (Figure 2a,c), driven by allele frequency clines at multiple loci whose centres are clustered at two localities (Figure 4). This suggests the presence of two main barriers to gene flow. These barriers spatially coincide with shifts in tidal regime (macro‐ vs. meso‐ vs. microtidal), but the two barriers seem to be shaped by different demographic histories (Figure 5). The barrier between micro‐ and mesotidal sites (i.e., between the Skagerrak and North Sea) seems to be a secondary contact zone between populations that have diverged in total isolation for about 90,000 years and then came in contact again (Table 2). The timing of secondary contact estimated here (14,000 years) approximately matches the ice retreat from the northern hemisphere (Hewitt, 2000), and is similar to the estimate obtained for the sand goby for secondary contact between North Sea and Baltic Sea populations (Leder et al., 2021).

A previous study exploring the genetic diversity of mitochondrial DNA in C. maenas described a genetic clade found only in populations from northern Europe, which could suggest the presence of a northern glacial refugium for the species (Roman & Palumbi, 2004). While it is not possible to infer the exact locations where isolation took place, one possibility to explain the secondary contact observed between the meso‐ and microtidal environment could involve a dual colonization of the North Sea – Baltic Sea system. Indeed, this area was mostly covered by land during the last glacial maximum, and was only accessible to marine life after the last deglaciation (about 15,000 years ago). Therefore, this secondary contact could correspond to the meeting point of two colonization waves, one coming from a northern refugium, and the other coming from the south, through the English Channel, as already previously suggested for this species (Marino et al., 2011). These contact zones are expected to be stabilized at environmental or physical barriers to gene flow through coupling effects (Barton, 1979; Bierne et al., 2011). In agreement with the coupling hypothesis, our biophysical dispersal modelling suggests a strong oceanographic dispersal barrier for larvae with tidal migration behaviour at the transition zone between meso‐ and microtidal environment (Figures 1 and 3e). Interestingly, the alleles of some loci which showed a sharp frequency cline at the transition zone to the microtidal environment were also found at high frequency in Bangor (e.g., locus in blue and purple in the inset to Figure 4).

In contrast, our biophysical modelling does not suggest an oceanographic barrier spatially coinciding with the other genetic break present in the English Channel (i.e., the transition between meso‐ and macrotidal). Demographic modelling suggested isolation with migration over ~100,000 years at this genetic barrier, which is in line with previous estimates of ~470,000 years for the European ancestors of the two introduced lineages now found on the east coast of the USA (Jeffery, DiBacco, Wringe, et al., 2017). Altogether, the absence of complete historical isolation and the lack of an oceanographic barrier suggests that adaptation, and notably adaptation to tidal environments, contributes to the genetic differences observed between macro‐ and mesotidal sites. While our data suggest that tidal amplitude plays a major role in both barriers, other environmental drivers could also contribute. Temperature has been shown to play an important role in shaping population differentiation in several crustaceans, such as in American lobster (Benestan et al., 2016) and northern shrimp (Jorde et al., 2015), and a genomic island has been identified in the shore crab which might be associated with rapid temperature adaptation (Tepolt et al., 2021; Tepolt & Palumbi, 2020). The geographical area assessed here is located in the middle of the native distribution of C. maenas, which makes it less likely that physical factors such as temperature and salinity play a major role in shaping population genetic structure. Indeed, we do not find any evidence that salinity seems to drive the observed population structure of C. maenas, despite a strong salinity gradient from the North Sea going into the Skagerrak and Kattegat (see Figure S7). In the North West Atlantic, one barrier to connectivity has been described in Nova Scotia at a temperature gradient for multiple species, including C. maenas (Jeffery, DiBacco, & Wyngaarden, 2017; Stanley et al., 2018). However, their identified multispecies barrier also coincides with a tidal regime shift from macro‐ to mesotidal (Flemming, 2012), and we here argue for the importance of differences in recruitment success in larvae adapted to different tidal regimes, affecting dispersal, rather than temperature as also suggested by Pringle et al. (2011).

Finally, the detected differences in gene expression for two clock genes—probably involved in regulating circadian behaviour of crab larvae—could be part of the explanation for the maintenance of the two identified barriers to gene flow associated with tidal regime. Based on the data obtained here, it appears that shore crabs possess a circadian clock that relies on many of the same molecular components as other, better understood, invertebrate circadian clocks. While the clock machinery for circadian rhythms is relatively well known (e.g., Crane & Young, 2014), little is known about which genes may be involved in tidal rhythms. The few studies investigating both circadian and lunar cycles (either for 12.4‐h or 29‐day periods) indicate that the two systems work independently and that lunar‐related rhythms may not be generated by the canonical circadian clock genes (Bulla et al., 2017). Our results here confirm that circadian genes and putative tidal cyclers occur in transcriptomes of C. maenas, and that they are expressed in both adult and megalopae phases. With linear models, we identified two genes, the core clock gene cyc and the clock‐associated gene pdp1, that may help explain behavioural differences among crab larvae from microtidal and tidal sites (macro‐ and mesotidal). In Drosophila constant high or low PDP1 levels disrupt locomotor activity rhythms, and although its function as a circadian oscillator is unknown, this suggests that PDP1 regulates oscillator output (Benito et al., 2007). This hypothesis is supported by our data, which showed that crabs that experience tides and do not follow a circadian rhythm (but a tidal rhythm) have a lower level of pdp1 transcripts. Additionally, acat1 may be a promising candidate gene for tidal cycling, as it was previously detected as a candidate “tidal cycler” in Cellana rota, and was among our few annotated outlier loci. However, we could only detect a weak effect of tidal gradient on gene expression of acat1 (Table S4).

A driving question for our research was to investigate the proximal causes for the adaptation of differences in timing of vertical migration of crab larvae according to tidal regime. Overall, the line of evidence in this study is a large step towards understanding the contribution of behaviour to local adaptation. While this study is an integrativeassessment of different approaches, there are nevertheless gaps in our line of evidence that make conclusions uncertain. For the gene expression analysis, we are able to show a correlation of gene expression with tidal gradient for two candidate clock genes, but we are not able to determine if the difference is driven by a plastic response or if it is heritable, for instance due to allelic differences. Moreover, we infer patterns of expression of clock genes in larval stages from analyses of adult crabs. However, we show that the clock genes are also expressed in megalopae, and while sensing and reacting to changes in light and tide may be particularly relevant in crab larvae, both light and tide have also been shown to alter behaviour and activity in adult shore crabs (Naylor, 1958; Queiroga et al., 1994; Zeng & Naylor, 1997). In the future, the availability of a C. maenas reference genome and using larger genomic coverage, combined with experimental work establishing the direct link between different larval behaviours and gene expression over time (particularly circadian and circa‐tidal), hold exiting potential to more fully understand the role of larval behaviour on the population structure of shore crabs. In particular, such assessments would help to conclude with more certainty that there are indeed locally adapted larval behaviours in C. maenas along a tidal gradient, which are mediated by genetic differences in clock gene alleles or their regulation. Additionally, such work could establish whether the identified clock genes (pdp1 and cyc) do indeed directly or indirectly control larval behaviour.

Finally, selection for larval behaviour that increases return to near‐coast habitats may explain regional differences and local adaptations of behaviour synchronized to regional hydrodynamic transport (Moksnes et al., 2014). In line with the perspective in Burgess et al. (2016), the pelagic larval stage may not only increase life‐long fitness through access to food and reduced predation in the water column, but also be under selection for increased dispersal. Indeed, larval behaviours (e.g., synchronized with the tide) promote return to adult habitats, which may also reduce along‐shore dispersal, and facilitate the evolution of locally adapted larval behaviours.

CONFLICT OF INTERESTS

The authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS

P.R.J., P.O.M. and M.J. designed the research and performed the sampling. A.L.M. performed the demographic modelling, G.A.M. carried out the gene expression experiments, M.J. performed the bioinformatic and population genetic analyses and P.R.J. performed the biophysical modelling. M.J. led the writing of the paper, and all authors contributed to the writing process.

Supporting information

Appendix S1

Table S5

Jahnke, M. , Moknes, P.‐O. , Le Moan, A. , Martens, G. A. , & Jonsson, P. R. (2022). Seascape genomics identify adaptive barriers correlated to tidal amplitude in the shore crab Carcinus maenas . Molecular Ecology, 31, 1980–1994. 10.1111/mec.16371

DATA AVAILABILITY STATEMENT

The 2bRAD raw sequences (fastq) are openly available on NCBI’s sequence read archive (SRA) (BioProject PRJNA797386, accession nos.: SAMN25002278–SAMN25002565). Key downstream data files, such as the final filtered vcf files for all, neutral and outlier loci, and key new scripts for population genetics and dispersal analyses can be found at: 10.17044/scilifelab.17836025. The scripts for the 2b‐RAD bioinformatic analysis are available at: https://github.com/z0on/2bRAD_denovo; and scripts for demographic inferences are available at: https://github.com/alanlm‐speciation/moments_optimization.