Seasonal immunoregulation in a naturally-occurring vertebrate
Fishes show seasonal patterns of immunity, but such phenomena are imperfectly understood in vertebrates generally, even in humans and mice. As these seasonal patterns may link to infectious disease risk and individual condition, the nature of their control has real practical implications. Here we characterize seasonal dynamics in the expression of conserved vertebrate immunity genes in a naturally-occurring piscine model, the three-spined stickleback.
We made genome-wide measurements (RNAseq) of whole-fish mRNA pools (n = 36) at the end of summer and winter in contrasting habitats (riverine and lacustrine) and focussed on common trends to filter habitat-specific from overarching temporal responses. We corroborated this analysis with targeted year-round whole-fish gene expression (Q-PCR) studies in a different year (n = 478). We also considered seasonal tissue-specific expression (6 tissues) (n = 15) at a third contrasting (euryhaline) locality by Q-PCR, further validating the generality of the patterns seen in whole fish analyses. Extremes of season were the dominant predictor of immune expression (compared to sex, ontogeny or habitat). Signatures of adaptive immunity were elevated in late summer. In contrast, late winter was accompanied by signatures of innate immunity (including IL-1 signalling and non-classical complement activity) and modulated toll-like receptor signalling. Negative regulators of T-cell activity were prominent amongst winter-biased genes, suggesting that adaptive immunity is actively down-regulated during winter rather than passively tracking ambient temperature. Network analyses identified a small set of immune genes that might lie close to a regulatory axis. These genes acted as hubs linking summer-biased adaptive pathways, winter-biased innate pathways and other organismal processes, including growth, metabolic dynamics and responses to stress and temperature. Seasonal change was most pronounced in the gill, which contains a considerable concentration of T-cell activity in the stickleback.
Our results suggest major and predictable seasonal re-adjustments of immunity. Further consideration should be given to the effects of such responses in seasonally-occurring disease.
KeywordsSeasonality RNAseq Teleost Three-spined stickleback Immunity Immunoregulation Ecoimmunology Wildlife
antigen presenting cell
false discovery rate
fragments per kilobase of exon per million fragments mapped
generalized additive model
gene set enrichment analysis
interbranchial lymphoid tissue
general linear model
major histocompatibility complex
principal components analysis
principal co-ordinates analysis
quantitative real-time PCR
structural equation model
river Stour (Sussex)
co-ordinated universal time
Seasonal immune function has often been observed in vertebrates , including humans [2, 3], but is relatively poorly understood. As with more studied circadian rhythms, though, there are fundamental implications for health [4, 5] (e.g., effects on vaccination and diseases linked to immune function). Taking a comparative approach  and considering the conserved genes of the vertebrate immune system, here we use transcriptomic measurements to reveal seasonal re-adjustments of immunity in a naturally-occurring teleost model. Crucially, through focussing on wild organisms exposed to real-world environmental extremes, we expected to discover more measurable variation than would be the case in a study of domesticated animals, where seasonal variation may be muted by anthropogenic influences.
We chose the three-spined stickleback (Gasterosteus aculeatus) as a subject because it has an annotated whole genome sequence  and occurs accessibly in highly seasonal natural habitats. Also, a considerable knowledge base exists for this species: it is a highly studied model organism [8, 9], and there are particularly detailed ecological studies relating to our main study area, mid Wales [10, 11]. We compared the transcriptomes of populations in late winter and late summer (outside of the breeding season, to reduce complexity) in ecologically divergent natural populations, reasoning that a focus on common responses would provide a way to filter overriding seasonal trends from locality-specific variation. We chose to primarily use global mRNA extracts from individual whole fishes rather than from isolated cell populations or tissues. This was because a fully reductionist approach to cell populations would be impractical, and because the majority of the teleost immune system is likely to be diffusely distributed in the gut, under the skin and mucosal surfaces and in association with the gills and liver (where, for example, complement proteins are mostly synthesized) [12, 13, 14, 15, 16].
By considering global (whole-fish) samples, we were thus able to take a holistic view of which immune system pathways are differentially expressed at seasonal extremes. We corroborated our transcriptomic analyses by targeted gene expression measurements in year-round samples of fishes from the original sites in a new annual cycle, and by tissue-specific analyses at a further site. Moreover, using network analyses we were able to ask what genes are important in regulating seasonal immune function and how do seasonally-biased immune networks interact with other seasonally-biased organismal processes?
Seasonal expression bias of immune system genes occurs against a well-defined genome-wide seasonal signature
We analyzed the global (whole fish) transcriptomes of G. aculeatus from two contrasting habitats in mid Wales, River (Afon) Rheidol (RHD) and Lake (Llyn) Frongoch (FRN), in September 2012 and March 2013. To begin our analysis we considered, genome-wide, which genes were associated with seasonal expression bias. At FRN, 4464 genes were significantly differentially expressed from summer to winter with an individual cut-off (P = 0.05) and 1678 with a false discovery rate (FDR)-adjusted cut-off. At RHD, 4383 genes were significantly differentially expressed with an individual cut-off and 2067 with an FDR-adjusted cut-off. Genes that were seasonally differentially expressed at both localities tended, overwhelmingly, to show synchronous expression (in the same direction at the same season across sites).
We hypothesized that these synchronously differentially expressed genes would also be those contributing to overarching seasonal responses. Thus, we categorized such genes on the basis that they were significantly (P <0.05) differentially expressed, in the same direction, at both sites at the individual error rate (in practice a more stringent cut-off than an FDR-adjusted P = 0.05 for one locality). Following this criterion, 1263 genes were differentially expressed in a consistent direction (Additional file 1: Table S1 shows those with Ensembl annotations), 850 increasing expression during winter (winter-biased) and 413 increasing expression during summer (summer-biased).
The above analyses were carried out on expression data un-adjusted for individual size, as this variable was (intentionally) approximately balanced across winter and summer samples. However, as our sampling points bounded a non-recruiting population ageing in the interval between breeding seasons, we considered in more detail the potential influence of ontogenetic stage. It is likely, given the months (March to September) in which we recorded reproductive activity in the field, and taking into account slower biological ageing at lower temperatures (through measuring age in growing degree days), that the 0+ cohort in our approximately size-matched summer and winter samples would have included individuals widely overlapping in effective age (see Additional file 4: Figures S1-S2). Furthermore, as our sampling deliberately selected a wide range of fish sizes, it is probable that 0+ and 1+ cohorts  were represented, resulting in a very extensive overlap of effective ages between summer and winter samples. A close association between body size and age allows age to be partitioned from season in statistical models by the use of a size metric, such as body length, as a surrogate. This is validated by data from experiments in artificial outdoor habitats, where we found that time explains at least 57 % of the variation in length (see Additional file 4: Figure S1). In order to control for age (length) effects we applied general linear models (LMs) to each in turn of the 11455 genes in the GSEA dataset, including main effects for season, length, sex and site. We found that season was the dominant predictor of gene expression (see Additional file 5: Figure S3a). Consistent with this gene-by-gene analysis, a multivariate principal co-ordinates analysis (PCO) of the same data demonstrated marked differentiation across seasons against the two major axes (axis 1, P = 0.003; axis 2, P = 0.0004), but none for length, sex or site (see Additional file 5: Figure S3b). We also re-ran “global” GSEA analyses (against all KEGG and REACTOME pathways), first with genes ranked by confounder-adjusted seasonal effect, and then with genes ranked by confounder-adjusted length effects (ranking was based on parameter sign and effect size, η2, in the LMs above). We found a similar outcome in the analysis ranked by confounder-adjusted seasonal effect to in the unadjusted analysis shown in Fig. 1a, with the two analyses sharing 67 gene sets that were significantly seasonally enriched (FDR-adjusted P = 0.05 cut-off), including all of the immunological sets except leucocyte transendothelial migration (Additional file 6: Table S4). In contrast, there was a distinctive outcome in the analysis ranked by confounder-adjusted length effect, where only 8 enriched gene sets were shared with the analysis shown in Fig. 1a, including none of the immunological sets. Thus, the effect of season was a very dominant one, emerging clearly in analyses even without adjustment for ontogeny.
Season is a dominant and consistent influence on immune gene expression
We also asked how important seasonal influences on the expression of immune-associated genes were in comparison to other sources of variation (site, sex, body size). To answer this we again considered LMs fitted to expression data for all 3648 immune-associated genes, initially with main effects for season, site, sex and body length and then with all 2-way interactions involving season. As for the analyses of genome-wide expression above, the broad pattern in these models was for season to be the dominant influence on immune gene expression (Fig. 2b), compared to individual sex, body size or site. Also, the interactions of season with other terms tended to be small compared to the main effect of season, indicating consistent seasonal effects across site, sex and age. Moreover, PCO ordination of all ImmPort list genes (whether seasonally expressed or not) revealed clear differentiation between summer and winter samples along similar trajectories between sites (Fig. 2c). These observations are consistent with overarching temporal environmental drivers acting similarly on the immune system across different habitat types and life-history stages.
Adaptive immunity genes are summer-biased and innate immunity genes are winter-biased
The 244 consistently seasonally-biased genes from the ImmPort list were individually evaluated to identify those with core immunological functions (the ImmPort list tending towards inclusivity) (Additional file 7: Table S5). Such “core” genes in the summer-biased set included those involved in, or regulating, adaptive effector response pathways (rag1, rag2, cd8a, zap70, ccr7, il4, igh@ irf4b, foxp3b, rorc, satb1), corroborating the summer bias in lymphocyte responses suggested by GSEA analyses. One weakly expressed classical major histocompatibility class (MHC) IIa locus (from chromosome VII ) was also detected more strongly in summer, although this was not the case for other more highly expressed MHCIIa loci; the chromosome VII locus is hereafter referred to as mhcIIa. There were also summer-biased genes involved in immunological cell adhesion (itgb2) and toll-like receptor (TLR)-mediated signalling (tirap).
In the winter-biased set there was a lack of genes clearly promoting adaptive immunity. However, there were several genes involved in regulating or suppressing lymphocyte activity (orai1, apoea, tnfrsf21, bnip3, rnf128, itm2a, tgfbr2) [19, 20, 21, 22, 23, 24, 25]. In addition there were genes associated with innate immune cell activity (nfkbiz, zbtb16b, lsp1, cd302) and interleukin (IL) 1 family signalling pathways (three il1r gene cluster members, il1rap), and genes like those up-regulated by type I interferons in mammals (ifi44/ifi44l-like) or involved in TLR signalling pathways leading to the production of type I interferons (tbk1). Key elements of non-classical complement pathways (cfd, masp2) were also winter-biased.
Although selected on the basis of Cuffdiff outputs , all of the core immune genes were highly significantly seasonally-biased when analysed in confounder-adjusted LMs with terms for season, site, sex and body length.
A set of highly co-expressed winter- and summer-biased immune genes can be identified that may lie close to a regulatory axis for seasonal immunity
Finally, we constructed a small (bearing in mind sample size considerations) three-variable structural equation model (SEM) of the form shown in Fig. 5b. We used this to further assess the influence of individual winter-summer interfacing (key) genes from ARACNe Network 1 on the seasonal transition in immune function. In this analysis two of the variables were derived as the first components from separate principal components analyses (PCAs) of summer and winter-biased core immune genes (but excluding key interface genes). Each component thus represented the major axis of covariation within the respective summer or winter-biased gene set. The third variable was the expression of a key winter-summer interfacing gene, each of which, in turn, was evaluated in the model. All of the winter-summer interfacing immune genes, except tbk1, negated the direct effect of winter-biased on summer biased genes (these were significantly associated in a univariate model) and themselves showed significant associations, of opposite sign, with the summer and winter-biased genes. This supports the linking role of these genes indicated by the ARACNe analyses.
Consistent seasonality confirmed by year-round Q-PCR measurement of key genes over a new annual cycle
Tissue-specific expression of key genes suggests intense seasonality in the gill
We also considered seasonality of key genes (see above) within specific tissues (Fig. 6b, c) at a new locality on the River Stour in eastern England (STO). All of these genes, except for orai1, were primarily expressed in organs with known concentrations of lymphoid tissue (Fig. 6b). Furthermore, all of the many instances of significant tissue-specific seasonal bias (13/25 comparisons) occurred in the same direction as predicted by the whole-fish transcriptomic study (Fig. 6c). Outside of the thymus, expression of T-cell-associated genes (cd8a, foxp3b) was highest in the gill, lower in head kidney, spleen and intestine and negligible in skeletal muscle (Fig. 6b, c). This is consistent with a strong concentration of T-cell activity in the gill. Moreover, the summer bias of T-cell-associated genes was seen primarily in the gill (Fig. 6c). In the case of orai1, whose expression is important in mammalian T-cells  but not narrowly characteristic of them, high expression occurred in skeletal muscle (consistent with a known physiological importance in this tissue ) (Fig. 6b, c). This gene was, however, also robustly expressed (Fig. 6b) and winter-biased (Fig. 6c) in the organs with greatest T-cell-specific expression, thymus and gill, supporting a possible role in seasonal immunoregulation. Genes from innate signalling pathways (tbk1, il1r-like) tended to be winter-biased in all tissues (Fig. 6c). Overall, gill most closely reflected the pattern of seasonal bias seen in whole-fish mRNA pools (Fig. 6c).
Seasonal immune gene expression links to wider life history signatures
Some seasonal immune functions in vertebrates are controlled by photoperiodic time measurement [30, 31] and the circadian molecular clock may also have a role in co-ordinating circannual biological rhythms [2, 32]. A scan of the seasonally-biased genes for those involved in such processes revealed that timeless (a clock-associated gene) occurred within the summer-biased set. When added to the ARACNe networks above timeless was, remarkably, most strongly connected to key winter-summer interface genes, with more connections to winter-biased genes (Fig. 3b). However, other genes involved in clock machinery or photoperiodism did not show the same tendency and timeless is known to have physiological functions in mammals that are independent of any role in biological clocks .
TLR signalling pathways show seasonal modulation
We have demonstrated seasonal re-adjustments of immune system gene expression in naturally-occurring freshwater teleosts. These occurred most intensely in the gill and were substantial (greater than variation between habitats and life-history stages) and over-arching (with consistent trajectories across habitats and life-history stages). In keeping with some previous suggestions about seasonal immune function in teleosts , we found that genes marking adaptive immune processes were summer-biased (expressed more strongly in summer), whilst certain innate immune genes were winter-biased. However, as set out below, our observations provide considerable new insights into the control of seasonal immune responses.
Transcriptomic analyses (based on whole-fish samples) indicated that summer-biased genes included many centrally involved in lymphocyte responses. For example, the recombination activating genes (rag1, rag2) and genes associated with particular adaptive cell populations: T-cells (zap70), cytotoxic T-cells (cd8a), helper T-cells (foxp3b, il4) and B-cells (igh@). In contrast, the set of winter-biased genes lacked those promoting adaptive effector responses. In all cases, winter-biased genes associated with T- or B-cell responses were regulatory or even suppressive in nature. This strongly suggests a regulatory control of adaptive immunity during winter, rather than, or additional to, a loss of function due to the kinetic consequences of low temperature in a cold-blooded organism. Furthermore, there were gene expression signatures of elevated innate immune functions in winter: including IL-1 signalling and non-classical complement pathways. A complex modulation of genes involved in innate TLR-mediated signalling occurred, with a predominant winter bias.
We designed the sampling for our transcriptomic study to, as far as possible, reduce correlation between season and ontogeny, and we also carefully considered, post-hoc, the possible role of ontogeny in generating apparent seasonal differences. To ensure that an extensively overlapping range of effective ages was present in our winter and summer transcriptomic samples, we deliberately selected a wide range of fish sizes within samples (to the extent that there were no significant differences in length between winter and summer samples). Through monitoring the growth of fishes in artificial outdoor habitats we confirmed that age predicted the majority of variation in length, and we adjusted for length, as a surrogate for age, in statistical models applied to transcriptomic data. Importantly, the much greater overall signature of season compared to length in statistical models applied to genome-wide and immune system-wide gene expression is not consistent with ontogeny being a major confounder in our study. Moreover, the balancing of size across seasonal samples, and the adjustment for length in our statistical modelling, also accounts for the possibility that growth allometries in different tissues (for example, proportionately increased muscle mass with size) may have biased results for whole-fish samples.
We also considered whether the patterns of gene co-expression in our transcriptomic data could give insights into the regulation of seasonal immune function. Information theory-based network analyses  of expression in seasonally-biased core immune system genes identified a small set of genes lying at the interface between summer- and winter-biased genes. These were highly networked (statistically associated) amongst themselves and also each highly networked within the seasonally-biased group to which they respectively belonged. Remarkably, several of the interfacing genes have roles in APC-T-cell immunological synapses (cd8a, zap70, orai1 and perhaps the summer-biased classical mhcIIa locus [27, 36, 37, 38]) and mutations leading to loss of function in their mammalian orthologues cause primary immunodeficiencies [27, 36, 37, 38]. Also amongst these interface genes, foxp3  is a master regulator of regulatory T-cell function and recombination-activating genes are central to the production of re-combined adaptive receptors . In mammals loss-of-function mutations in these genes respectively cause lethal autoimmunity and severe combined immunodeficiency [39, 40]. Other interfacing genes are involved in innate processes that might precede antigen presentation: innate signalling pathways (tbk1, il1r-like [41, 42, 43, 44]) and antigen internalization via phagocytosis or endocytosis (cd302, colec12 [45, 46]).
When we added the summer-biased clock-associated gene timeless to these networks, it proved to be closely associated with interface genes, and especially with winter-biased (mostly innate) interface genes. Whilst this could reflect some co-ordination via a seasonal oscillator, though, other genes involved in photoperiodism or circadian rhythms did not enter the networks in corresponding ways. It is also the case that timeless itself has a relatively poorly resolved role in the mammalian circadian clock and is known to have independent physiological functions  (consistent with the links to metabolic pathways discussed next).
During winter there was a genome-wide signature indicative of elevated metabolic processes and metabolite transfer and organismal stress, and in summer a signature of growth and developmental processes. Again using network analyses of our transcriptomic data, we finally asked how seasonal changes in immunity might be related to this background. We found that the winter-summer interfacing (key) immune genes identified above were especially highly connected to genes involved in non-immune seasonal variation, further emphasizing their relevance in the seasonal control of immunity. Genes involved in metabolism and oxidative stress interconnected densely with winter-biased innate genes, and amongst these especially to the winter-summer interfacing genes tbk1 and il1r-like, and also to il1rap. On the other hand, genes involved in general organismal stress responses linked differently to winter-biased immune responses: primarily via tnfrsf21, a protein that triggers apoptotic pathways  and restrains T-cell  and B-cell  responses. In comparison, genes involved in non-immune summer signatures (growth and development) networked primarily to summer-biased adaptive genes, especially to the summer-biased interface genes cd8a and zap70. These observations suggest an unexpectedly strong link between growth processes and adaptive immunity and that one, or both, may favour permissive conditions for the other. Taken together, the above patterns indicate that multiple organismal processes are likely to interact with the seasonal regulation of immunity, additional to the possible influence of any “hard-wired” circannual oscillator. It might be expected, then, that predictable seasonal influences will be modified by less predictable non-cyclical temporal variations in environmental stressors .
To validate our transcriptomic analyses we returned to our original study localities, an upland lake and river in mid-Wales, and also considered artificial outdoors habitats stocked from the lake site. Using Q-PCR measurements we confirmed (with very strong statistical support) seasonality in a panel of the key immunity genes predicted to be winter- or summer-biased. This year-round monthly analysis considered whole-fish samples (n = 478) and was carried out in a new annual cycle that lacked unusually cold winter or spring weather. Furthermore tissue-specific analyses (discussed below) at an entirely new locality (a euryhaline estuarine site in eastern England) found that all tissues showed seasonal expression changes and these changes all occurred in the same direction as in the whole-fish studies at our original sites. Thus, overall we considered 3 very divergent localities (upland lake, river, estuarine) across 2 years and found compelling evidence to support a general pattern such as that indicated in our initial transcriptomic measurements.
In addition to our analyses of whole-fish mRNA pools, we also confirmed tissue-specific expression patterns through Q-PCR measurements of key immunity genes at a new estuarine locality (considering head kidney, spleen, thymus, gill, intestine and muscle). As indicated above we found many significant tissue-specific seasonal expression differences and all of these were in the direction predicted by our other whole-fish studies. The most pronounced seasonal expression profile occurred in the gill (and this profile most closely reflected seasonal change at the whole-fish level). Furthermore, the gill contained the most intense concentration of T-cell activity outside of the thymus, with elevated expression of T-cell associated genes such as cd8a and foxp3b, and expression of these genes was seasonal in the gill but non-seasonal in the thymus. These observations are consistent with the known responsiveness of immune gene expression in the teleost gill to environmental stimuli , and also with the recent discovery and characterization of extensive, T-cell rich, interbranchial lymphoid tissue (ILT) in teleost fishes [12, 51, 52, 53]. Our results suggest the possibility that ILT may have an important role in seasonal immune function.
Finally, and whilst the present study is intended to characterize the seasonal dynamics of gene expression, rather than identify environmental causation, we briefly consider what external agents may drive the responses that we observed. In highly seasonal temperate zone habitats, such as the ones we consider here, each of temperature, diet, photoperiodic responses, pathogen exposures, or other biotic or abiotic manifestations of the environment, could be involved to unknown degrees. In the future, by matching detailed field observations with mesocosm studies and laboratory experiments, we expect to dissect the relative contributions of these influences to seasonal immune variation and to the immune phenotype more generally.
Our results suggest that in wild teleosts, during winter conditions, adaptive immune activity declines in a manner that involves the expression of regulatory genes affecting lymphocyte function. This is indicative of a controlled, strategic response rather than a simple kinetic tracking of environmental temperature. Seasonal change is most prominent in the gill, suggesting ILT may be important in such responses. Further broad attention to seasonal immune function is certainly warranted, given the likely practical relevance – through effects on infectious disease susceptibility and inflammatory status – to health in humans and domesticated animals and to fitness in natural populations.
Sampling and habitats
Samples of three-spined sticklebacks (Gasterosteus aculeatus L.) for transcriptomic analysis were taken at 9:00–12:00 h (UTC) in September 2012 and in March 2013, outside of the breeding season and respectively prior to the autumnal and vernal equinoxes. Specimens were collected at two contrasting sites in the Ceredigion area, mid Wales, U.K. (8–10 individuals/site/sampling point). One site (FRN) was a 7.2 ha upland lake, Lake (Llyn) Frongoch, 13.7 km from the sea at an elevation of 280 m (52.3599,–3.8776). The other (RHD) was a non-tidal minor channel of the River (Afon) Rheidol, 3.5 km from the sea at an elevation of 10 m (52.4052,–4.0372).
Additional specimens for corroborative tissue-specific quantitative real-time PCR (Q-PCR) gene expression studies (September 2012, n = 5; March 2013, n = 10) were collected from a site (STO) on the river Stour in Sussex, U.K. (51.9544, 1.0222). The STO site was in a small, tidal side-channel of the main river at an elevation of 1 m and 2.2 km inland from the tidal sluice opening into the main estuary.
U.K. meteorological office records indicate that March 2013 encompassed extended winter conditions and was the coldest U.K. March since 1962 and joint second coldest since 1910 . Weather patterns in September 2012 were unremarkable for the time of year. Water temperatures at the study sites varied across an approximate range of 13–20 °C in September and 0–5 °C in March; the FRN and STO samples in March were collected from habitats with superficial ice formation.
To avoid the confounding of variation between September and March with individual ontogeny, all samples were selected to contain a wide (extensively overlapping) range of sizes. Sample characteristics are summarised in Additional file 8: Table S6 (and there was no significant winter-summer difference in length for any of the locality-specific sample sets). Given considerations of timing and environmental temperature, the (widely overlapping) potential effective age variation in our samples is set out in Additional file 4. The use of body size as an age indicator is validated by data from our outdoors artificial habitats (see below), where time explains a minimum of 57 % of variation in individual body length over a 12 month study interval, even given a heterogenous starting population of wild fishes that varied in length by a factor of up to × 1.6 (see Additional file 4: Figure S1). Thus, it was possible to partition the effects of age in statistical models (described below) through the inclusion of body length as a surrogate term.
In addition, and also for the purpose of corroborative Q-PCR gene expression measurements, we considered samples of fishes from FRN (~10 individuals/month), RHD (~10 individuals/month) and 12 outdoors artificial 300 L habitats (~20 individuals/month) from October 2013 to September 2014. The artificial habitats were located on the Aberystwyth university campus and stocked in August-September 2013 with post-larval fishes from FRN which were given 2 × anti-parasitic Praziquantel treatments (24 h at 4 mg l−1; FlukeSolve, Fish Treatment Limited) to prevent Gyrodactlylus epizootics and maintained on a diet of frozen mini bloodworm (Tropical marine centre) adequate for normal growth. Water temperature within the artificial habitat units was uncontrolled, or was controlled a small increment (2 °C) above the ambient temperature in adjoining uncontrolled units. U.K. meteorological office climate summaries  confirm that this 2013–2014 sampling period occurred across average autumn temperatures, above average winter and spring temperatures, lacking frost conditions, and a summer period lacking extremely hot weather.
All animal maintenance and sampling of animals in the field followed U.K. Home Office (HO) regulations and local (Aberystwyth University) ethical procedures.
Whilst parasite infections are not considered explicitly in the present study, the river (RHD) and lake (FRN) populations studied supported divergent and predictable macroparasite communities (with limited seasonality) (unpublished data), whose differential influences are likely to emerge primarily in the site effect of analyses described below. There was no evidence of infection or pathology in any of the specific organs used for tissue comparisons.
Sample handling, nucleic acids processing and library preparation
Sticklebacks were captured individually using a dip net and immediately killed by concussion and de-cerebration and stored in RNAlater™ at ambient temperature. On return from the field (within 1–2 h) the samples were transferred to 4 °C overnight and then to -80 °C for long-term storage. Immediately prior to RNA extraction, sticklebacks were thawed at 4 °C, dabbed dry with tissue and weight (mg) and standard length (mm) recorded. For transcriptomic studies, RNA from whole fishes was extracted using the Isolate II RNA mini kit (Bioline): whole individual fishes were homogenized in lysis buffer using a 5 mm stainless steel bead (Qiagen, 69989) in a Qiagen TissueLyser LT system and a standard aliquot of the homogenate passed through the manufacturer-recommended protocol. RNA extracts were subjected to standard quality control diagnostics and individually barcoded cDNA libraries (mRNA focussed) for 36 fishes were prepared using the TruSeq RNA Sample Preparation Kit v2 (Illumina).
For Q-PCR, RNA was extracted from the 2013–2014 monthly samples as above, whilst RNA from the STO samples was extracted using the RNAqueous-Micro Total RNA Isolation Kit (Life technologies). All samples were DNase treated prior to conversion to cDNA with the High Capacity RNA-to-cDNA Kit (Life technologies).
Next generation sequencing and differential expression analysis
Individually barcoded Truseq sequencing libraries were sequenced using 4 lanes of an Illumina HiSeq2500 sequencer at IBERS, Aberystwyth University. Libraries for 2 summer individuals and 2–3 winter individuals from each locality were run on each lane (thus balancing different sampling units across lanes). Following removal of adaptors, the output paired end reads (~110 bp) were quality-controlled using FastQC  and the leading 10 bp of all reads trimmed prior to analysis via the Cufflinks suite of programmes . Reads were mapped to the stickleback genome (Broad, gasAcu1) using Tophat and de novo assembled into transcripts using Cufflinks. Transcripts from all samples were merged with Cuffmerge, using the USCS genes annotation (for gasAcu1) as the reference annotation. Differential gene expression analyses were run for each of the sites separately using Cuffdiff with parameters set for geometric library normalization, pooled dispersion estimation, a false discovery rate of 0.05, a minimum alignment count of 10, and using multi-read correction and bias correction. For subsequent analyses, FPKM data for individual loci were generated with Cuffnorm. Predicted genes with <0.5 FPKM mean expression or >50 % undetectable expression were excluded from all analyses below.
Q-PCR gene expression measurements
Primers used for quantitative real-time PCR (Q-PCR) measurements
Ensembl gene number
F - CCACCCTGTACTGCAATCGA
R - CCGCCTGCTGTTTTCTTTTG
F - TCTGAACACAGTCATGGGGAGA
R - CCAGGATGAGCTGACTTTCCA
F - GCACCTCGGCTCTGTTGTC
R - CCATGAGGGCGAAGAGGTGTA
F - AGACGGAGCAGCTGTTCGA
R - GCATATCTCATCATATCTGACGACAT
F - GAACGCGAGAACTGCAAGAAC
R - GGGACGCTGGTGAAGTTGAA
F - CACTTTAGCGGAGCTGTTGGA
R - AGAAAAGGAAGTCCGGAACCA
F - CCCTCAAACGGAGACTTTACGT
R - GGTGCCGCTGAGCTCTTC
Analyses of RNAseq data
As a small preliminary exercise to assess the relevant information content of the Broad gasAcu1 assembly for our wider study, we arbitrarily selected a panel of 31 immunological genes-of-interest whose existence would be expected to be conserved in a lower vertebrate genome and searched for these in the Ensembl stickleback database and the gasAcu1 genome assembly. From this list, 26 (84 %) were associated with predicted database genes (annotated), 3 (10 %) were detectable in the genome assembly but not annotated in the database and 2 (6 %) were absent from the database and genome assembly. In all cases fragments of the genes (confirmed by sequencing) were amplifiable by PCR using primers designed from the assembly sequence, or, in the case of the two missing genes, fragments (also confirmed by sequencing) were amplified using primers designed from conserved regions in multi-species (teleost) alignments. After filtering out low expression genes (see above), the RNAseq dataset contained 20947 predicted loci. Of these, 16575 (79 %) matched genes in the Ensembl G. aculeatus database. These data suggested that although a proportion of real genes were missing in the stickleback genome assembly, sufficient information was present to take a broad view of genome-wide expression patterns.
Genes were classified as summer- or winter-biased if they were significantly biased in the same direction at both RHD and FRN at an individual error rate (P < 0.05) based on Cuffdiff output; this represents a combined individual P <0.0025, in practice a more stringent cut-off than a False discovery rate (FDR)-adjusted P = 0.05 threshold for a single locality. For unannotated predicted loci that were seasonally-biased we performed a series of standardized Blast (tblastx) searches that identified a small number of additional genes with high confidence.
For analyses involving comparisons to curated gene sets, we used Ensembl Biomart  to convert the identifiers for annotated stickleback genes to the HGNC symbol for an estimated orthologous human gene. Where there were multiple predicted human orthologues (typically related in broad function), a single estimated orthologue was randomly retained per stickleback gene. Similarly, where more than one stickleback gene shared the same predicted human orthologue, only one of these was randomly retained in the gene list. Thus only annotated stickleback genes with a corresponding predicted human orthologue and HGNC symbol were considered in these analyses.
Immune-associated genes were initially defined by homologous relationships with genes from the relatively inclusive ImmPort comprehensive list of immune-related genes . Seasonally-biased genes from this list were individually, manually assessed for core immune functions (evidence of direct involvement in immune effector or regulatory activity) using links from the gene list report in DAVID 6.7 and from GeneCards v. 4.0.
Gene set nrichment analysis (GSEA v2.1.0) [59, 60] was used to investigate whether a priori defined gene sets showed significant expression differences between winter and summer samples. Separate GSEA analyses were carried out for FRN and RHD (Fig. 1a), using ranked winter-summer expression changes (GSEAPreranked), and comparing to all KEGG (c2.cp.Kegg.v.5.0.symbols.gmt) and REACTOME (c2.cp.Reactome.v.5.0.symbols.gmt) pathway gene sets within the MsigDB database . FDR-adjusted P values from this analysis were combined for the two localities by Fisher’s method  and combined FDR P values < 0.05 were considered further. In parallel, more targeted GSEA analyses were carried out using smaller numbers of selected REACTOME and GO gene sets to represent different immunological processes and also wider organismal processes (growth, responses to stress, metabolism, reproduction). Applying hypergeometric distribution tests (Fisher’s exact test) for overlap between gene sets, these selected gene sets were additionally used to probe the sets of winter-biased and summer-biased genes identified above, and sets of genes from modules identified by network analyses (see below).
The information-theory (mutual information, MI) based programme ARACNe2 (Algorithm for the construction of accurate cellular networks)  was applied to predict networks of interactions between gene products. Simulation studies  indicate this approach retains useful accuracy at sample sizes of the order utilized here (n = 36 fishes with transcriptomic data). Working with log2-transformed data we constructed the following networks: for seasonally-biased core immune genes alone (Network 1); for the full genome-wide set of seasonally-biased genes (Network 2); for the full set of ImmPort list immune-associated genes, whether these were seasonally-biased or not (Network 3); for the core-seasonally-biased immune genes and, in addition, groups of genes from the sets we used to represent wider organismal processes (selecting those that tended to be seasonally-biased in GSEA analyses) (Network 4). In Network 1 and 4 all genes were set as hubs, whilst in Networks 2 and 3 the seasonally-biased core immune genes were set as hubs. Networks were constructed with the adaptive partitioning algorithm, using a mutual information (MI) threshold estimated by a pre-processing run. For the networks shown, P thresholds were set at 1 × 10−5 (Networks 1 and 3), 1 × 10−4 (Network 2) or 1 × 10−6 (Network 4), with correction for the number of markers in the case of Networks 1 and 4. All networks shown were bootstrapped (2000 resamples; significance cut-off for reported edges, P = 1.0 × 10−6). Cytoscape 2.8 was employed to visualize networks (initially using force-directed layouts, from which the layout of nodes was sometime modified for clarity of presentation) and to calculate network statistics (Network Analyzer plugin). Betweenness centrality  was calculated to represent the centrality of nodes within a network (and thus the tendency of indirect connections across the network to route via that node). Eccentricity, the maximum path length connecting a node to any other node in the network, was also calculated to (inversely) reflect nodes that lie at the centre of a network. Both of these quantities might be indicative of the regulatory influence of individual nodes by reflecting their tendency to be co-expressed (and thus perhaps co-regulated) with many other nodes .
Principal co-ordinates analysis (PCO) of log2 transformed FPKM gene expression values was employed to ordinate individuals across other study variables (LabDSV package, R). First principal component scores from principal components analyses (PCAs) on the correlation matrix were used to represent the major axis of covariation within genes sets in some analyses. The R package GeneOverlap was used to compute overlap statistics amongst gene sets (significance tests based on a hypergeometric distribution, odds ratios and Jaccard similarity indices). For general linear model (LM) analyses of bulk sets of genes, models were fitted to log2 transformed FPKM data and statistics extracted using the lm and associated functions in R. For analyses focussing on smaller numbers of genes, equivalent models were run with transformations applied on a case-by-case basis based on standard model diagnostics. Small structural equations models (path analysis) and generalized additive models (GAMs) were respectively implemented in the R packages Lavaan and mgcv. All analyses with R used version 3.1.0.
Availability of supporting data
Sequencing data will be available in the European Nucleotide Archive under primary accession number PRJEB13319. Other supporting data are available as additional files.
This work was supported by the Leverhulme Trust (grant number RPG-301); related grants from the Fisheries Society of the British Isles (FSBI) and the Natural Environment Research Council (grant number NE/L013517/1), UK are also gratefully acknowledged. We are very grateful to the late Dr Rob Wootton and to Dr Chris Williams (Environment Agency, UK) for support, to Dr Rob Darby, Rory Geohagen and Gareth Owen (IBERS, Aberystwyth University) for technical assistance, and to Dr Matt Hegarty and co-workers (IBERS, Aberystwyth University) for carrying out next generation sequencing.
- 54.Anonymous. Why was the start to spring 2013 so cold? Synopsis report CSc 05. Exeter, U.K: Met Office; 2013.Google Scholar
- 55.Climate summaries. Met Office, Exeter, U.K. 2016. http://www.metoffice.gov.uk/climate/uk/summaries. Accessed 15 Apr 2016.
- 56.56. Andrews S. FastQC: a quality control tool for high throughput sequence data. 2016. http://www.bioinformatics.babraham.ac.uk/projects/fastqc. Accessed 15 Apr 2016
- 61.Fisher RA. Statistical methods for research workers. Edinburgh: Oliver & Boyd; 1925.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.