High sensitivity to ocean acidification in wild out‐migrating juvenile Pacific salmon is not impacted by feeding success

Abstract

Salmon populations are declining worldwide, with high mortality rates during juvenile marine migration presenting a bottleneck to recruitment. The ocean conditions along the main migratory route of juvenile salmon in British Columbia are characterized by high variability in CO₂, with the amplitude, duration, and frequency of ocean acidification events exacerbated by climate change. Similarly, the variability in ocean conditions affects the abundance and diversity of plankton prey, leading to areas of food paucity for juvenile salmon. We investigated the combined effects of ocean acidification (control and 3200 μatm CO₂) and food limitation (ad libitum, ½ ration, and food deprived) on the survival, condition, and gene expression profiles of juvenile Chum salmon (Oncorhynchus keta) to develop predictive biomarkers for CO₂ exposure and food deprivation. Ocean acidification caused a direct 3‐fold increase in mortality over 25 days of exposure, which was unaffected by food availability but differentially affected smaller fish. CO₂ exposure induced transcriptomic changes in a suite of genes associated with ion regulation, while food deprivation was associated with a differential expression of stress, immune, and mortality markers, as well as reduced condition factor. Our data indicate that CO₂ directly impairs ionoregulatory capacity to the point of failure in juvenile Chum salmon and that these effects cannot be compensated through increased energy from food. Applying our gene panels as biomarkers to a subset of fish with known exposure, we were able to accurately predict exposure to CO₂ and food deprivation (74% and 90%, respectively). By combining these gene panels with previously established biomarkers for other environmental stressors, the recent environmental stress history of wild fish can be determined and can be used in models to predict salmon returns, informing fisheries management and conservation efforts.

INTRODUCTION

Understanding how local conditions in the ocean and freshwater system shape the vulnerability to climate change and adaptive capacities in commercial fish is imperative to advancing ocean conservation efforts, fisheries and aquaculture management, and policies for the new Blue Economy. Anthropogenic CO₂ emissions are causing the oceans to warm, as well as acidify, leading to widespread changes to ecosystems (Fabry et al., 2008). Ocean acidification is caused by the absorption of excess anthropogenic CO₂ from the atmosphere by the surface ocean, changing the carbonate chemistry of seawater and lowering its pH from preindustrial levels of 250 μatm CO₂ (~pH 8.2) to projected global sea surface levels of 1000 μatm CO₂ (~pH 7.4) by the end of the century (Collins et al., 2019). In addition to the slow global increase of oceanic CO₂ content through sea surface gas exchange, local oceanographic processes cause large‐ and small‐scale variability in CO₂, particularly in dynamic coastal waters through upwelling and estuaries with low buffering capacity, high riverine input, and microbial respiration. In the Baltic Sea, for example, CO₂ levels above 2800 μatm were frequently measured in the summer months already a decade ago (Reusch et al., 2018). In the Pacific Northwest, heatwaves and upwelling also elevate levels well above those projected for the global ocean with climate change (Cavole et al., 2016; Schwing et al., 2000; Sydeman et al., 2013), and the area between the Salish Sea and the Queen Charlotte Strait has been highlighted as a potential bellwether and hotspot for extreme OA events that can occur rapidly and last for years with levels of 1000 μatm frequently measured (Evans et al., 2019, 2022, 2025; Holdsworth et al., 2021).

Direct effects of ocean acidification on fishes have been found to be strongly species and life‐stage specific, with the early life stages generally more vulnerable (Cattano et al., 2018). Pacific salmon (Oncorhynchus sp.) are a group of species of high cultural and economic value and are key ecological players in the North Pacific food web. Many Pacific salmon populations have exhibited strong declines in recent decades due to a wide range of factors such as fishing, logging, and habitat degradation, with the changes in ocean conditions driven by climate change a likely contributing factor to high mortality rates during early marine migration (Beamish & Mahnken, 2001; Cohen, 2012; Del Rio et al., 2019; McKinnell et al., 2014; Mueter et al., 2005; Sobocinski et al., 2018). Juvenile salmon migrating from the Fraser River watershed in British Columbia to the open North Pacific experience challenging conditions in the Tidal Mixing Zone between the Discovery Islands and the Johnstone Strait (Dosser et al., 2021; Evans et al., 2019; Nemcek et al., 2008; Peña & Fine, 2024; Tortell et al., 2012).

Yet, studies on ocean acidification effects on physiology in the early life stages of Pacific salmon remain scarce. Negative effects on growth, metabolism, and behavior have been observed in hatchery‐reared larval pink salmon (Oncorhynchus gorbuscha; Ou et al., 2015) and juvenile coho salmon (Oncorhynchus kisutch; Williams et al., 2019) at 2000–2700 μatm CO₂. On the other hand, wild juvenile pink salmon caught in the Tidal Mixing Zone and exposed to 2000 μatm CO₂ for two weeks were unaffected by ocean acidification (Frommel et al., 2020). Hatchery‐reared Chinook salmon at the time of smoltification were also able to fully recover from a short‐term acid–base disturbance during 3 ½ weeks of exposure to 1400 μatm CO₂ (Frommel et al., 2024). This suggests a high tolerance to short‐term, moderate CO₂ exposure during the juvenile stage in some salmonids and may indicate a preselection for robust individuals during the intense bottleneck of the smolting phase, or a difference in vulnerability between hatchery and wild fish. Salmon hatcheries are often characterized (and challenged) by extremely high CO₂ levels (Fivelstad et al., 2003) and some countries recommend CO₂ limits for best animal welfare practices to combat high mortalities, such as Canada at 10 mg/L (~5800 μatm CO₂) (Moccia et al., 2020) and the United Kingdom at 20 mg/L (~11,600 μatm CO₂, or roughly 3.5× the levels used in this study) (RSPCA, 2021), thus salmon originating from hatcheries may be preselected to tolerate higher CO₂ levels than wild fish.

In addition to direct effects, the sensitivity of fish to ocean acidification has also been shown to be highly dependent on other synergistic factors, including food availability (Sswat et al., 2018; Stiasny et al., 2019). Along the BC coastal shelf, hydrographic conditions associated with climate change can alter primary productivity through shifts in nutrient concentrations and composition (Arrigo, 2005; Del Bel Belluz et al., 2021; Holdsworth et al., 2021; Strom et al., 2006), reducing the quantity and quality of plankton (Costalago et al., 2020; El‐Sabaawi et al., 2009; Mahara et al., 2021), with cascading effects up the food chain (Litzow et al., 2006). Juvenile salmon migrating through the Tidal Mixing Zone show poor feeding success, likely due to unsuitable size distribution of zooplanktonic prey (James et al., 2020), and reduced body condition. In addition to adverse oceanic conditions and low food supply, juvenile Pacific salmon migrate past active Atlantic salmon farms (Rechisky et al., 2021), where they may be exposed to a range of pathogens and parasites (Bateman et al., 2021; Brauner et al., 2012; Johnson et al., 2021; Mordecai et al., 2021; Nendick et al., 2011).

The effects of environmental stressors and the degree of exposure are challenging to assess in wild fish. Transcriptomic biomarkers in combination with physiological traits offer a powerful method to determine recent environmental stress history. Using the Fluidigm BioMark technology, 96 genes across 96 samples can be measured to determine the set of genes related to the stressor of interest. Biomarkers for stress in response to disease, salinity, temperature, and hypoxia have been established for salmon using this approach for monitoring the health of Pacific salmon by the Department of Fisheries and Oceans Canada (DFO) (Akbarzadeh et al., 2020; Elmer et al., 2023; Houde et al., 2019; Jeffries et al., 2014; Miller et al., 2017), but to date no genetic markers of CO₂ exposure and food limitation have been established.

The objective of this study was to understand the combined effect of ocean acidification and low food availability on the condition, survival, ionoregulation, and establish genetic biomarkers for these stressors in chum salmon (Oncorhynchus keta). In a controlled laboratory study, we reared juvenile wild chum salmon under the combined stressors of CO₂ and food deprivation for up to 25 days, an ecologically relevant duration of stress in this area. We hypothesized that CO₂ exposure in combination with food deprivation would synergistically affect survival, condition, and immune response in juvenile chum salmon, reflected by gene expression patterns in the gills. We challenged salmon with higher levels of CO₂ than previous studies (Frommel et al., 2020, 2024) to determine a threshold of ecologically relevant CO₂ tolerance. The transcriptional markers of CO₂ and food stress identified by this study provide new tools to assess salmon environmental responses, and will contribute two new biomarker panels to the salmon “Fit‐Chip” tool (described in Akbarzadeh et al., 2024), built to comprehensively assess the health and condition of wild and farmed salmon globally. End user groups of these salmon Fit‐Chips include fisheries, aquaculture, and hatchery managers, as well as researchers in government, industry, and environmental NGOs.

METHODS

Fish catch and husbandry

Juvenile chum salmon were caught with a purse seine net off of Bold Point in the Discovery Islands on June 5, 2021. Fish were immediately transported back to the Hakai Institute Quadra Island Ecological Observatory (~30 min) by boat in a large tote, closely monitoring water temperature and oxygen during transport. At the field station, salmon were distributed into 18 tanks (260 L each) at an initial density of 32 fish per tank. Tanks were set to partial flow‐through (0.7 L/min) and recirculation of filtered, UV sterilized seawater pumped from 20 m depth from the adjacent Hyacinthe Bay. Water was cooled and maintained at 14.5°C, and a preset CO₂–air mix (Alicat Scientific mass flow controllers MC‐500SCCM‐D‐DB9S, MCR‐1000SLPM‐D‐DB9S) was continuously bubbled (7.5 L/min) into each experimental tank to achieve either ambient (530 μatm CO₂) or acidified (3200 μatm CO₂) conditions, which were randomly allocated. Fish were provided with one of three food concentrations: ad libitum (4% body weight), ½ ration (2% body weight), and food deprived (0%). The resulting treatments (two CO₂ and three food concentrations) were conducted in triplicate. Ad libitum and ½ ration treatment groups were hand‐fed twice daily with locally caught frozen krill (Euphausia pacifica), a common prey item of juvenile chum (Brodeur, 1991). Temperature and pH were monitored in real‐time (Walchem and Honeywell systems) and temperature, salinity, pH, and dissolved oxygen were measured daily in each tank with a hand‐held multimeter (YSI). Ammonia was measured every other day, and levels remained below 0.01 mg/L. Discrete water samples from individual tanks were taken every week and analyzed for pCO₂ and TCO₂ using a non‐dispersive infrared absorbance gas analyzer (L1840A LI‐COR) constrained in a Burke‐o‐Lator (BoL, Dakunalytics, LLC) (Pocock, 2023). From these measurements, the in situ carbonate system parameters (pCO₂ and pH_Total) were calculated with an Excel version of CO2SYS version 2.5 using the dissociation and solubility constants from Lueker et al. (2000), Dickson (1990), and Uppstrom (1974). For a summary of the carbonate system and water parameters, see Table 1. Daily mortality was monitored, and weight and length of each dead fish were recorded.

TABLE 1.

Water parameters (mean ± SD) over the duration of exposure of juvenile chum salmon to different CO₂ levels (feeding rations combined): temperature (T), salinity, pH_Total, pCO₂, TCO₂, and total alkalinity (TA).

	T (°C)	Salinity	pHTotal	pCO2 (μatm)	TCO2 (μmol/kg)	TA (μmol/kg)
Control	14.78 ± 0.14	28.58 ± 0.70	7.90 ± 0.04	531.94 ± 57.20	1885.53 ± 36.63	1998.04 ± 29.20
High	14.78 ± 0.15	28.56 ± 0.74	7.17 ± 0.02	3214.31 ± 140.17	2116.50 ± 29.97	2018.32 ± 29.22

Sampling

A subsample of 18 fish was taken as a baseline at the beginning of the experiment before the initiation of treatments. After 12 and 25 days of exposure to their respective treatments, 10 fish per tank were sampled on each day in random tank order by hand‐net and euthanized with an overdose of tricaine methanesulfonate (MS‐222, 500 mg/L). Fish were blotted dry and weighed, and fork length was measured with calipers, from which Fulton's K was calculated as a measure of condition. From a severed caudal peduncle, blood was drawn into a heparinized hematocrit tube (in triplicate) from the caudal vein and spun down in a hematocrit centrifuge for determination of hematocrit (ratio of volume of packed red blood cells to total blood volume). The second gill arch on the left side of the fish was dissected out and frozen in liquid nitrogen for later gene expression analysis. Tools were sterilized with EtOH and flame between each fish. The second gill arch on the right side was removed, wrapped in aluminum foil, and frozen in liquid nitrogen for determination of gill Na⁺K⁺‐ATPase (NKA) activity.

Na⁺K⁺‐ATPase (NKA) activity

Gill NKA activity was measured with a modified protocol from McCormick (1993). Briefly, 20 mg of tissue were added to 1 mL of 0.5% SEID buffer (SEI buffer with 0.5% sodium deoxycholate), homogenized in a refrigerator for 5 min, and centrifuged for 5 min at 5000 RCF at 4°C. The supernatant was collected for determination of ATPase activity and protein levels. ATPase activity was measured in triplicate in a 96‐well microplate at 340 nm with and without ouabain as an inhibitor. For protein levels, samples were diluted 5:1 with SEID and measured in a 96‐well plate in triplicate with 250 μL of Bradford reagent per 10 μL of sample at 595 nm. Gill NKA activity was calculated by the differences between the presence and absence of ouabain and standardized to the protein levels.

Gene expression

Gene expression analysis of gill tissue was completed for 80 individuals. The method for gill tissue homogenization employed TRIzol (Ambion, Foster City, California, USA) and BCP reagent (Sigma‐Aldrich, Oakville, Ontario, Canada), followed by RNA extraction using the MagMAX‐96 Total RNA Isolation kits (Ambion). RNA was normalized to 62.5 ng/mL, and cDNA synthesis was performed using SuperScript VILO synthesis kits (Invitrogen), as previously described (Akbarzadeh et al., 2018, 2020, 2021; Houde et al., 2019). Negative controls for cDNA synthesis included no reverse transcriptase (no RT), no RNA, and cDNA.

Gene expression was first examined for 96 curated biomarkers associated with different environmental stressors, smolt stage, and morbidity status based on previous studies (Akbarzadeh et al., 2018, 2020, 2021, 2024; Houde et al., 2019). For gene expression analysis on the Fluidigm BioMark HT microfluidics platform, target cDNA amplicons for each gene were enriched using the specific target amplification (STA) method, as previously described (Akbarzadeh et al., 2018, 2020; Houde et al., 2019). Then the amplified samples and the 96 TaqMan assays (primers and probes) were run on one 96 x 96 gene expression dynamic array following the Fluidigm platform instructions (Fluidigm Corporation, CA, USA).

The dynamic array included serial dilutions (1, 1/5, 1/25, 1/125, 1/625, and 1/3125) of a cDNA pool composed of all Chum salmon samples. The qPCR conditions were as described previously (Akbarzadeh et al., 2018, 2020, 2021, 2024; Houde et al., 2019). Data were extracted using the Real‐Time PCR Analysis Software (Fluidigm), and the Ct thresholds were manually scored for each assay across all the chips.

PCR efficiency for each assay was calculated using (10¹/slope − 1) × 100, where the slope was estimated by plotting the Ct over the serial dilutions of cDNA. To determine the optimal normalization gene(s), the expressions of all 96 genes were evaluated across all samples using two independent software tools: geNorm (https://genorm.cmgg.be) and NormFinder (https://moma.dk/normfinder-software). These tools identified the most suitable genes with the best stability (lowest SD). The expression of the samples was then normalized to the geometric expression of HTATIP, H2F2, and EEF2 (the genes that were the most suitable based on the three software tools), using the ∆∆Ct method (Livak & Schmittgen, 2001) with the pooled samples as the calibrator in GenEx (ver. 7). The gene expression values were then log‐transformed to log2(2^−∆∆Ct) for all further analyses. Primer sequences for genes significantly affected by food deprivation or CO₂ exposure are shown in Appendix S1: Table S1.

Data analyses

Mortality, weight, condition, and gill NKA activity were analyzed using mixed model ANOVAs with CO₂, day, and treatment as fixed effects, and tank included as a random factor. Effect size statistics were performed with the natural logarithm of the response ratio (LnRR) with a 95% CI (Hedges & Olkin, 1985). The normalized gene expression data were subjected to a t‐test (Mann–Whitney non‐parametric test for some biomarkers, due to failed normality assumption) to discover the genes significantly different between ad libitum and food‐deprived groups, and between the CO₂‐exposed and control fish. A receiver operating characteristic (ROC) analysis was also conducted to find the most effective biomarkers between treatments, which were used for the principal components analysis (PCA). For the Random Forest‐based classification model (RF), the entire normalized gene expression data for the selected significant biomarkers was split into analytical training (two‐thirds samples) and testing sets (one‐third samples). The model was trained via the training set with 10,000 trees (Toth et al., 2019). Then, the performance, including sensitivity, specificity, and prediction accuracy, was measured on the testing set (Akbarzadeh et al., 2024). The RF‐based classifiers were built using the Random Forest R package, based on the algorithm of Breiman (2001).

RESULTS

Experimental parameters (CO₂ and temperature) were highly stable in this experiment and oxygen and ammonia were continuously maintained at optimum levels (>89% DO and <0.01 mg/L ammonia). Despite being wild fish, juvenile salmon in the fed groups consumed the offered krill well and quickly. In the high CO₂ treatment, fish occasionally behaved skittishly or swam more erratically than in the control treatment; however, this behavior was not quantified. After 2–3 weeks in the experiment, mortality rapidly increased with high mortality observed in the CO₂ treatment. Mortality was significantly affected by both CO₂ exposure and days in the experiment (CO₂: p = 0.002, F ₁ = 14.00; Day: p < 0.001, F ₁ = 34.3), while food did not significantly affect mortality (p = 0.62, F ₂ = 0.48). Continuous exposure to high CO₂ caused a mean 235% increase in mortality in juvenile chum salmon by 25 days compared to the control CO₂, regardless of food ration (Figure 1).

FIGURE 1.

Cumulative mortality (mean ± SD) of juvenile chum salmon during exposure to control (530 μatm CO₂; green) and high CO₂ (3200 μatm CO₂; maroon) over the course of 25 days of exposure.

While mortality was consistently higher in smaller sized fish (Figure 2A), survival was unrelated to condition factor (Figure 2B). Weight was not significantly affected by feeding (p = 0.184, F ₂ = 1.91) or CO₂ (p = 0.968, F ₁ = 0.002), but was significantly affected by days in experiment (p < 0.001, F ₁ = 22.94) (Figure 2A). The average length of fish was 81.1 ± 7.4 mm and did not differ significantly by treatment, tank, or days.

FIGURE 2.

(A) Weight and (B) condition factor (Fulton's K) for juvenile chum salmon (boxplots) exposed to control (530 μatm CO₂; green) and high CO₂ (3200 μatm CO₂; maroon) for 12 and 25 days at three different food rations (panels: ad libitum, ½ ration, and no food). Superimposed are mortalities (colored dots) for the respective treatment and binned by day of exposure.

Food and days in experiment, and the interaction between day and food, significantly affected juvenile condition but CO₂ did not (Fulton's K, Food: p < 0.001, F ₂ = 25.2; CO₂: p = 0.723, F ₁ = 0.13; Day: <0.001, F ₁ = 312.47; Day × Food: p < 0.001, F ₂ = 17.32). Food deprivation led to a 20% (±12%) decrease in Fulton's K by 25 days in the high CO₂ treatment compared to the ad libitum fed fish in both the high and control CO₂ treatments (Figure 2B; see Appendix S1: Table S2 for effect size analysis). Unfed fish in the control CO₂ treatment showed a 15% (±13%) reduction in Fulton's K compared to the ad libitum fed fish after 25 days. The ½ ration diet did not lead to significantly different Fulton's K. Hematocrit was only affected by days in treatment (p < 0.001, F ₁ = 15.45) but was not affected by CO₂ (p = 0.149, F ₂ = 2.19), food (p = 0.91, F ₂ = 0.10), or the interaction of day and food (p = 0.79, F ₂ = 0.24) (see Appendix S1: Table S2 for effect size analysis).

Gill NKA activity was significantly affected by days of exposure (p = 0.005, F ₁ = 8.33), as well as the interaction between day and food ration (p = 0.001, F ₂ = 7.56) (Figure 3). CO₂ did not have a significant effect on gill NKA activity (p = 0.217, F ₁ = 1.70).

FIGURE 3.

Gill Na+, K+‐ATPase (NKA) activity in juvenile chum salmon exposed to control (530 μatm CO₂) and high CO₂ (3200 μatm CO₂) for 12 and 25 days on three different food rations (gold: ad libitum; green: ½ ration; gray: food deprived).

Of the investigated genes, 10 were significantly influenced by CO₂ exposure and 37 were significantly affected by food deprivation. The CO₂ biomarkers were predominantly genes associated with ion regulation and generally clustered into two groups, separating the CO₂‐exposed fish (MRPL40, NKA1a, and CFTR‐1) and control fish (CCL19, HLA21‐G1, H2EB1, NAMPT, CCL4, and IFIT5‐2) (Figure 4A, Table 2). Of the 37 food biomarkers, 18 contributed to a PCA, mainly associated with immune response, general stress, and imminent mortality, as well as two associated with hypoxia stress and three with ion regulation (Figure 4B, Table 2). One gene associated with elevated stress, immune response, and morbidity, ODC1, had 30% outliers in the fed treatment (10‐fold higher expression compared to the other samples) and was removed from further analysis. Genes classifying food deprivation clustered into two distinct categories: (1) genes associated with ad libitum fed fish coded for proteins involved in hypoxia response and immune function and (2) food‐deprived fish which clustered around genes associated with imminent mortality and general stress (Figure 4B, Table 2).

FIGURE 4.

Principal components analysis (PCA) for genetic biomarkers of juvenile chum salmon affected by 25 days of exposure to (A) CO₂: Control CO₂ (530 μatm, green circles) and high CO₂ (3200 μatm, maroon triangles); (B) food treatment: Ad libitum fed (gold closed circles) and food deprived (gray open circles). Lighter shading of arrows indicates weaker contribution to PCA. List of gene name and functions in Table 2.

TABLE 2.

Candidate genes for genetic biomarkers for chum salmon: genes significantly affected by either CO₂ exposure or food deprivation by t‐test (p‐value <0.05) with area under curve (AUC) values obtained from receiver operating characteristic (ROC) analysis. Primer sequences are shown in Appendix S1: Table S1.

Biomarker	Function	Gene code	Gene name	p‐value	AUC
CO2 exposure	Immune response	IFIT5‐2	Interferon induced protein with tetratricopeptide repeats 5	0.024	0.658
CO2 exposure	Ion regulation	CFTR‐I	Cystic fibrosis transmembrane conductance regulator I	<0.001	0.742
CO2 exposure	Ion regulation	MRPL40	Mitochondrial ribosomal protein L40	0.013	0.674
CO2 exposure	Ion regulation	NKA1a	Na/K ATPase α‐1a	0.017	0.667
CO2 exposure	Ion regulation	CCL19	C‐C chemokine 19	0.018	0.666
CO2 exposure	Ion regulation	HLA21‐G1	Human leukocyte antigen G1	0.022	0.372
CO2 exposure	Ion regulation	H2EB1	Histocompatibility 2, class II antigen E beta	0.026	0.656
CO2 exposure	Ion regulation	NAMPT	Nicotinamide phosphoribosyltransferase	0.029	0.653
CO2 exposure	Ion regulation	CCL4	C‐C chemokine 4	0.042	0.642
CO2 exposure	Mortality	ARRDC2	Arrestin domain containing 2	0.035	0.648
Food deprivation	General stress	HSP90a‐15	Heat shock protein	0.002	0.741
Food deprivation	General stress	FKBP10‐4	FKBP prolyl isomerase 10	<0.001	0.862
Food deprivation	General stress	HSP70	Heat shock protein	0.003	0.730
Food deprivation	General stress	HSP90b	Heat shock protein	0.009	0.702
Food deprivation	Immune response/Mortality	MMP13	Matrix metallopeptidase 13	<0.001	0.788
Food deprivation	Hypoxia stress	ALDOA1	Aldolase, fructose‐bisphosphate A	<0.001	0.889
Food deprivation	Hypoxia stress	MFHAS1	Multifunctional ROCO family signaling regulator	0.006	0.713
Food deprivation	Immune response	GAL3	Galectin‐3	<0.001	0.808
Food deprivation	Immune response	IFI44A	Interferon‐induced protein	<0.001	0.805
Food deprivation	Immune response	NFX	Nuclear transcription factor	<0.001	0.755
Food deprivation	Immune response	MX	MX dynamin‐like GTPase	0.001	0.755
Food deprivation	Immune response	HERC6	Ubiquitin protein ligase	0.002	0.741
Food deprivation	Ion regulation	HLA21‐G1	Human leukocyte antigen	<0.001	0.770
Food deprivation	Ion regulation/Mortality	FKBP5	FKBP prolyl isomerase 5	<0.001	0.783
Food deprivation	Ion regulation	H2EB1	Histocompatibility 2, class II antigen E beta	0.005	0.730
Food deprivation	Imminent mortality	ATP5G3	Mitochondrial ATP synthase	<0.001	0.776
Food deprivation	Imminent mortality	GLUL	Glutamate‐ammonia ligase	<0.001	0.756
Food deprivation	Imminent mortality	CLEC4E	C‐type lectin domain family 4 member E	0.006	0.714

CO₂ exposure caused a downregulation in one gene related to immune response (IFIT5‐2), five related to ion regulation (CCL19, HLA1G1, H2EB1, NAMPT, and CCL4), and one related to imminent mortality (AARDC2) (Figure 5A). Three ionoregulatory genes were upregulated in response to CO₂ exposure (CFTR‐I, MRPL40, and NKA1a) (Figure 5A). In response to food deprivation, two genes related to general stress were upregulated (HSP90a‐15 and HSP90b), while two were downregulated (FKBP10‐4 and HSP70) (Figure 5B). The gene most affected by food deprivation was ALDOA‐1, which was downregulated 3.5‐fold in the unfed group relative to the fed group. Another gene related to hypoxia tolerance was also downregulated (MFHAS1). All but one gene related to the immune response were downregulated in the unfed treatment (GAL3, IFI44a, HERC6, MX, and NFX); the exception was MMP13, which can also be associated with morbidity (Houde et al., 2019; Jeffries et al., 2014), including in chum salmon (Akbarzadeh, unpublished data). Two ionoregulatory genes were downregulated (HLA21‐G1 and H2EB1), while one also associated with morbidity was upregulated (FKBP5). Four of the five genes typically associated with morbidity (imminent mortality) were upregulated in food‐deprived fish on Day 25 (GLUL, CLEC4E, FKBP5, and MMP13).

FIGURE 5.

Gene expression patterns (relative to housekeeping gene) of candidate genes for CO₂ and food biomarkers grouped by physiological category for juvenile chum salmon following 25 days of exposure: (A) CO₂ classifiers (control CO₂ = green; high CO₂ = maroon) and (B) food classifiers (ad libitum = gold; food deprived = gray).

Using the significant genetic biomarkers, the RF model was able to classify the food deprivation with 90% prediction accuracy, and CO₂ exposure with 74% prediction accuracy in test fish (Table 3). The importance of biomarkers in the model based on the Gini scores for CO₂ exposure and food deprivation is shown in Appendix S1: Figure S1. To further examine factors associated with the classification sensitivity of CO₂‐exposed fish, we investigated whether size was a singificant factor. Smaller fish, which comprised the bulk of fish that died over the 25‐day study, showed a higher correlation with positive classification and responded more strongly to CO₂ (p < 0.001).

TABLE 3.

Performance analysis of the Random Forest classification model testing one third of the dataset for exposure to 25 days of food deprivation and CO₂ using a threshold of 0.5 in juvenile chum salmon.

Biomarker	Treatment	Sensitivity %	Specificity %	Prediction accuracy %
Food deprivation	Ad libitum	88.9	90.0	89.5
	No food	90.0	88.9
CO2 treatment	Control	100.0	50.0	73.7
	High	50.0	100.0

All data are available in the data repository PANGEA through the Ocean Acidification International Coordination Centre (OA‐ICC) at https://doi.pangaea.de/10.1594/PANGAEA.981363.

DISCUSSION

The objective of this study was to determine the impact of naturally occurring food limitation on juvenile salmon resilience to ocean acidification and to develop biomarkers for CO₂ and starvation stress for wild salmon. On the Pacific West Coast, juvenile salmon experience naturally high CO₂ levels during their physiologically challenging migration from fresh‐ to saltwater, along with a strongly reduced prey availability and feeding success (Evans et al., 2019; James et al., 2020). Previous studies have shown mixed levels of vulnerability in juvenile salmon to ocean acidification, likely dependent on species and life stage, as well as level of CO₂ tested and duration of exposure (Frommel et al., 2020, 2024; Ou et al., 2015; Williams et al., 2019). While acid–base disturbances were seen in hatchery‐raised Chinook salmon 3–6 days post transfer to seawater at high (1400 μatm) and fluctuating (between 400 and 1400 μatm) CO₂ levels, they were able to overcome these disturbances by 18 days of exposure with no adverse impacts to survival, growth or condition (Frommel et al., 2024). In hatchery‐raised ocean‐phase coho salmon exposed to two weeks of 2700 μatm CO₂, olfactory‐mediated behaviors were impaired, but no negative effects of CO₂ on condition or survival were reported (Williams et al., 2019). This study aimed to test the impacts of CO₂ on another species of Pacific salmon, while challenging them with a level high enough to determine a potential threshold of resilience.

Additionally, studies have found food to play an important role in mitigating CO₂ stress (Sswat et al., 2018; Stiasny et al., 2019), likely by supplying the energy needed for energetically costly regulation against acid–base stress. It therefore stands to reason that the effects of CO₂ stress may be exacerbated in juvenile fish with poor feeding opportunities. To test this theory, we reared juvenile chum salmon in a controlled laboratory study under the combined effects of CO₂ exposure and food limitation. To assess transcriptional responses to food deprivation and CO₂, we conducted high throughput qPCR of gill cDNA on biomarkers previously developed to identify transcriptional responses associated with smoltification, saltwater adaptation, hypoxia, thermal stress, imminent mortality, and viral disease in various species of salmon (e.g., Akbarzadeh et al., 2024; Miller et al., 2017). Many of these genes were involved in ionoregulation, predicted to be affected by elevated environmental CO₂.

Surprisingly, juvenile chum salmon were highly vulnerable to high CO₂, but much more tolerant to food deprivation than expected. Interestingly, the combination of CO₂ and food deprivation did not have synergistic effects on survival or body condition. There was a strong direct effect of CO₂ on the survival of juvenile chum, with mortality nearly three times higher in fish exposed to high CO₂ for three weeks and up, regardless of feeding. This is in stark contrast to previous findings for juvenile pink and Chinook salmon from the same area, which were able to cope with high CO₂ levels for 2–3 weeks without reductions in survival or condition (Frommel et al., 2020, 2024). Rather than condition, size was a significant determinant for mortality, with the smallest fish being most affected. While the level of CO₂ and duration of exposure may have played a role in the differences in susceptibility observed between studies, another explanation may be related to the small size of chum salmon at seawater entry, making them more susceptible to environmental challenges such as CO₂ than Pacific salmon that enter seawater at a larger size, such as Chinook or coho salmon (Frommel et al., 2024; Williams et al., 2019). It could also suggest that chum salmon are most vulnerable to high CO₂ during the early saltwater acclimation phase soon after ocean entry. This theory would align with negative effects of high CO₂ found in pink salmon physiology and behavior, which also enter seawater early in development and at a small size (Ou et al., 2015).

While direct mortality in response to food deprivation was not observed during the 25 days of exposure in our study, it did lead to a decrease in condition factor and the fish were likely moribund. In the wild, food availability and nutritional quality have been found to significantly affect body condition of juvenile sockeye salmon migrating through the Tidal Mixing Zone (Garzke et al., 2022). Our study suggests that food availability by itself, not the prevailing high CO₂ levels, affects the condition of fish in the wild. NKA activity was also affected by food deprivation, with higher activity after 12 days than after 25 days, potentially indicating an overshoot of activity followed by a reduced ability to regulate with lower energy availability as the duration of food deprivation increased. But there was no interaction of food deprivation with CO₂ exposure. Taken together, this suggests that CO₂ did not increase the energy expenditure to maintain acid–base homeostasis (Brauner et al., 2019), as was previously shown in other species (Sswat et al., 2018; Stiasny et al., 2019). Hematocrit, a measure of stress in fish, was also not affected either by CO₂ or food availability. Rather, our data indicate that CO₂ may directly impair ionoregulatory processes resulting in death, particularly in the smallest fish. This theory is worth pursuing and a future study should focus on testing size‐specific ionoregulatory disturbances with shorter sampling intervals and targeted sample processing of small and large fish.

We identified a number of genes that were significantly affected by our treatments, and were able to establish biomarker panels to detect CO₂ stress and food deprivation. Co‐expression of a panel of 10 CO₂ responsive biomarkers showed high specificity (100%), but low sensitivity (50%) to detect exposed test fish, with an overall accuracy of 74%. This means that the CO₂ classifier was able to recognize all control fish, but only half of the CO₂‐exposed fish responded. This is an important distinction and one that has been seen in other classifiers (e.g., hypoxia and thermal exposure), and may be linked with fish size, as seen in our mortality data. While we did not observe elevated imminent mortality markers in food‐deprived fish, combined with the other markers (condition index) it does signal that these fish were highly stressed by lack of food, and mortality may have been affected over a longer term duration of food deprivation.

Most of the biomarkers differentiating CO₂‐stressed fish were involved in ion regulation, many associated in previous studies with saltwater adaptation and smoltification (e.g. Akbarzadeh et al., 2024; Houde et al., 2019). CFTR‐1, which is typically upregulated in smolts and as fish acclimate to saltwater, was the strongest affected gene, expressed at higher levels under elevated CO₂. CFTR is a low conductance anion channel in the apical membrane in the gills of saltwater teleosts and is responsible for Cl⁻ secretion, which facilitates HCO₃ ⁻ uptake to compensate for a respiratory acidosis (Marshall & Singer, 2002). Upregulation of CFTR could thus enhance acclimation to high CO₂ and contribute to further adaptation to saline water. However, the majority of ionoregulatory biomarkers responding to CO₂ were downregulated, opposing their typical upregulation in smolts (e.g., CCL4, HLA21, NAMPT, H2EB1), while NKA1a, the freshwater form of NKA, typically downregulated upon saltwater entry, was upregulated. This pattern is consistent with osmotic imbalance of CO₂‐exposed fish, particularly in smaller sized fish. While we did not measure shifts in ion balance in the blood, we found no significant impact of CO₂ exposure on gill NKA enzyme activity, a well‐established marker for ionic stress in smolting juvenile salmon (Kennedy et al., 2007; Stich et al., 2015, 2016).

A panel of 18 food‐responsive biomarkers showed 90% accuracy in categorizing fish as fed or food deprived and consisted of a suite of genes involved in immune response, stress, imminent mortality, hypoxia stress, and ion regulation. The gene most affected by food deprivation was ALDOA‐1, which was downregulated 3.5‐fold. This is a gene commonly associated with hypoxia stress in teleosts but mainly involved in glycogen storage in salmonids (Furukawa et al., 2018). Immune responsive genes, particularly those associated with anti‐viral immunity (Miller et al., 2017), were downregulated in food‐deprived fish, consistent with previous findings of immunosuppression in salmon sampled in areas with low prey availability (Deeg et al., 2022). While food deprivation had no effect on observed mortality up to 25 days, several genes previously associated with morbidity in salmon (Akbarzadeh et al., 2024; Houde et al., 2019) were indeed upregulated in food‐deprived chum salmon, including GLUL, FKBP5, CLEC, and MMP13. These findings underscore the potential that longer term food deprivation could very well affect survival and surprisingly, we are able to detect these signals in the gills, an organ not generally considered to be metabolically involved. Overall, the immunosuppressive patterns of expression associated with food deprivation indicate that food‐deprived fish may be more vulnerable to infection, that elevations in general stress could make fish more vulnerable to environmental stressors other than CO₂, and that survival over longer timeframes than measured herein may be negatively impacted. There was no overlap of differential gene expression between CO₂ and food deprivation markers.

Transcriptomic biomarkers for salmon have previously been developed for salinity and temperature stress (Houde et al., 2019), hypoxia (Akbarzadeh et al., 2020), viral disease (Miller et al., 2017), general stress and imminent mortality (Evans et al., 2011; Jeffries et al., 2014). The current work adds transcriptional biomarkers for CO₂ and food stress in salmon to the Fit‐Chips used for ongoing salmon health monitoring efforts by DFO and nongovernmental conservation groups. By combining these biomarker panels into a single tool that can be applied in wild‐caught salmon, the recent history of environmental stressors can be determined through non‐lethal gill clip sampling. This can greatly advance our understanding of cumulative and synergistic responses to physiological stressors contributing to variation in survival and potentially involved in the decline of wild salmon populations. This is particularly relevant in the context of the continued increase in atmospheric pCO₂ due to anthropogenic emissions, and the associated climate‐driven changes in ocean near‐shore conditions that are impacting the food webs, which support salmon and other fishes. Locally in British Columbia, the stress experienced by migrating salmon through tidally mixed zones and estuaries may make them more susceptible to cumulative stressors, for example due to pathogen and parasite loading. Identifying areas where migrating salmon experience increased stress can directly inform focused management decisions aimed at minimizing risks to wild salmon for improved conservation outcomes.

FUNDING INFORMATION

This research was supported by a Mitacs Grant and the Tula Foundation IT13677 to B. Hunt and A. Frommel, and by Natural Sciences and Engineering Research Council (NSERC) Discovery grants RGPIN‐2023‐03456 to C. Brauner and RGPIN‐2022‐03092 to A. Frommel. Funding for Fit‐Chip analysis was provided by Fisheries and Oceans Canada.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

ETHICS STATEMENT

This study was approved by the University of British Columbia Animal Care Committee (AUP number A19‐0284) according to the Canadian Council of Animal Care Standards. Collection and transfer of fish was approved by Fisheries and Oceans Canada (numbers XR 1222021 and 124630).

Supporting information

Appendix S1.

Frommel, Andrea Y. , Akbarzadeh Arash, Chalifoux Virginie, Ming Tobi J., Collicutt Brenna, Rolheiser Kate, Opie Rumer, Miller Kristina M., Brauner Colin J., and Hunt Brian P. V.. 2025. “High Sensitivity to Ocean Acidification in Wild Out‐Migrating Juvenile Pacific Salmon is Not Impacted by Feeding Success.” Ecological Applications 35(5): e70058. 10.1002/eap.70058

DATA AVAILABILITY STATEMENT

Data (Frommel, 2025) are archived in PANGEA through the Ocean Acidification International Coordination Centre (OA‐ICC) at https://doi.pangaea.de/10.1594/PANGAEA.981363.