Snow crab were captured by pot fishing during Fisheries and Oceans Canada research vessel surveys conducted in NL, Canada (47 4.2 N, 53 39.5 W) in December, 2016. Pots were designed to select for large, terminally-molted, commercial size (> 95 mm carapace width) male snow crab. They were transported in refrigerated (1°C) seawater to a laboratory facility at the Northwest Atlantic Fisheries Centre (Fisheries and Oceans Canada), St. John’s, NL, Canada. In the laboratory, individual snow crab were tagged (plastic tubing tied around the carapace widthwise) with a unique tag number, weighed, and divided amongst four (i.e. two control and two experimental) identical 1150 l tanks, at stocking densities of 6-11 kg m^-3. To ensure a consistent environment for the duration of the experiment, water flow was kept at a constant rate, water temperature was between 1-3°C, and the tank room was maintained at the natural photoperiod for this latitude. Tank functions were inspected daily. They were fed a diet of capelin once per week and allowed to feed for 24 h, after which all remaining food was removed from the tanks. Crab weight, mortality and feeding were monitored throughout the experiment.

Starting on January 13, 2017, on alternating days, the experimental tanks were exposed to impulse sound every 10 sec from a seismic soundtrack played through an underwater speaker (Lubell Labs, Model LL9161) for approximately 22 weeks. In the wild, snow crab in this study experience median daily noise levels (24 h sound exposure level) of 150 dB re 1 µPa² ·s, with considerable variability largely due to wind storm events; during offshore seismic surveying activity, the noise level exceeded 190 dB re 1 µPa² ·s (Morris et al., 2018). In this laboratory study, the daily noise level measured in the control tanks was 136 dB re 1 µPa² ·s, while the noise level in the exposure tanks was greater than 200 dB re 1 µPa² ·s. Therefore, in our experiment, noise exposure conditions in the control tanks were lower and more consistent than open ocean conditions, with those in the experimental tanks consisting of very loud noise impulses, even in comparison to offshore soundscapes directly impacted by seismic surveying. We recognize that the acoustic properties of tanks and the noise generated by speakers are not fully representative of real-world conditions; however, captive animals in this experiment were exposed to very different levels of noise, under controlled conditions, allowing us to directly measure the effects of noise on gene expression.

For both the control and experimental tanks, feeding was stopped 8 days prior to sampling. Snow crab were dissected and sampled on June 14 and 15, 2017 alternating between tanks [i.e. 1 crab from control tank 1 (n=11), then 1 crab from control tank 2 (n=13), then 1 crab from experimental tank 1 (n=17), then 1 crab from experimental tank 2 (n=14)]. Sampling alternated between tanks to control for any unknown effects of time of day on gene expression. Body mass and carapace width were measured, and claw meat density was calculated. For molecular analyses, hemolymph was collected from the right leg of the first walking pair (from the front) by cutting 1 cm above the first joint using clean kitchen shears that had been treated with RNaseZAP (Sigma-Aldrich, Oakville, ON, Canada) and then pouring it into an RNase-free 15 ml tube. Approximately 2 x 1.5 ml of hemolymph were then transferred by pipette into 2 x 1.5 ml RNase-free tubes and centrifuged at 500 x g for 5 min at 4°C. The plasma was completely removed by pipette and the cell pellets (hemocytes) were immediately flash-frozen on dry ice. The carapace was then removed to expose the organs. Approximately 100 mg of hepatopancreas were removed, placed in a 1.5 ml RNase-free tube and immediately flash-frozen on dry ice. Samples were stored at -80°C until preparation of RNA. Additional tissues were collected for assessment of indicators of animal health (see Morris et al., 2021). Multiple people were involved in sampling using a team-based assembly line approach to minimize sampling time and ensure that crabs were fully processed within 5-10 min after removal from the tank.

RNA extraction was performed using TRIzol Reagent (Invitrogen/Thermo Fisher Scientific, Burlington, ON, Canada). For hemocytes, 400 µl of TRIzol were added to each of the duplicate cell pellets, and homogenized using a motorized Kontes RNase-Free Pellet Pestle Grinder (Kimble Chase, Vineland, NJ, United States). The homogenates were then combined, mixed by pipetting, frozen on dry ice and stored at -80°C. Hepatopancreas samples were homogenized in 400 µl of TRIzol as for hemocytes. An additional 400 µl of TRIzol were added, mixed by pipetting, and the homogenates frozen on dry ice and stored at -80°C. Frozen homogenates were further processed by thawing on wet ice and then passing through a QIAshredder (QIAGEN, Mississauga, ON, Canada) spin column following the manufacturer’s instructions. Two-hundred microliters of TRIzol were added to each sample to make a total homogenate volume of approximately 1 ml. The TRIzol total RNA extractions were then completed following the manufacturer’s instructions.

The hepatopancreas samples had low A260/230 ratios following the TRIzol extraction, which interferes with column purification. Therefore, a subsample (120 µg) of each TRIzol-extracted RNA sample was re-extracted using the phenol-chloroform phase separation method as described in Hall et al. (2021).

The TRIzol-extracted hemocytes and the acid phenol:chloroform:isoamyl alcohol extracted hepatopancreas RNA samples (45 µg) were then treated with 6.8 Kunitz units of DNaseI (RNase-Free DNase Set, QIAGEN) with the manufacturer’s buffer (1X final concentration) at room temperature for 10 min to degrade any residual genomic DNA. DNase-treated RNA samples were column-purified using the RNeasy Mini Kit (QIAGEN) following the manufacturer’s instructions. RNA integrity was verified by 1% agarose gel electrophoresis, and RNA purity was assessed by A260/280 and A260/230 NanoDrop UV spectrophotometry for both the pre-cleaned and the column-purified RNA samples.

Five crabs from each of the 2 control and the 2 experimental tanks (n=20) were selected for molecular analyses based on comparable size, health (e.g. no missing limbs or black spots on the hepatopancreas) and RNA quality (i.e. high integrity and with A260/280 and A260/230 ratios > 2). Hemocytes and hepatopancreas samples from each of these crabs were then subjected to RNA-seq and qPCR analyses.

Prior to library construction, total RNA integrity was further validated using the 2100 Bioanalyzer system (Agilent, Santa Clara, CA, United States). Libraries were generated from 250 ng of total RNA. Library construction and sequencing services were performed at Génome Québec laboratories, Montréal, QC, Canada. As described in Hall et al. (2021), mRNA enrichment was performed using the NEBNext Poly(A) Magnetic Isolation Module (New England BioLabs, Whitby, ON, Canada) following the manufacturer’s instructions. cDNA was synthesized using the NEBNext RNA First Strand Synthesis Module and NEBNext Ultra Directional RNA Second Strand Synthesis Module (New England BioLabs) following the manufacturer’s instructions. Library preparation was then completed using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England BioLabs) following the manufacturer’s instructions. Adapters and PCR primers were purchased from New England BioLabs. Libraries were quantified using the Quant-iT™ PicoGreen^® dsDNA Assay Kit (Life Technologies/Thermo Fisher Scientific) and the Kapa Illumina GA with Revised Primers-SYBR Fast Universal Kit (Kapa Biosystems/Roche Diagnostics, Laval, QC, Canada) following the manufacturer’s instructions. Average fragment size was determined using the LabChip GX (PerkinElmer, Woodbridge, ON, Canada).

The libraries were normalized and pooled, denatured in 0.05 N NaOH, and then neutralized using HT1 buffer (Illumina, San Diego, CA, United States). ExAMP (Illumina) was added to the mix following the manufacturer’s instructions. The pool was loaded at 200 pM on a cBot (Illumina) and the flowcell was run on a HiSeq 4000 (Illumina) for 2 x 100 cycles (paired-end mode). A phiX library was used as a control and mixed with libraries at the 1% level. The Illumina control software was HCS HD 3.4.0.38 and the real-time analysis program was RTA v. 2.7.7. The program bcl2fastq2 v2.18 was used to demultiplex the samples and generate fastq reads.

Individual assemblies were generated for each cell/tissue type (i.e. hemocytes and hepatopancreas; referred to as RNA-seq analysis 1). A hemocytes-hepatopancreas co-assembly was also built (referred to as RNA-seq analysis 2) which allows for cross-cell/tissue type comparisons.

Reads were trimmed from the 3’ end to a minimum Phred score of 30. Illumina sequencing adapters were removed from the reads, and reads < 50 bp were excluded. Trimming and clipping were performed using Trimmomatic (Bolger et al., 2014). Reads were assembled into a de novo transcriptome using Trinity software (v2.4.0) with default parameters (Haas et al., 2013). Assembled contigs shorter than 200 bp were discarded. Kallisto (v. 0.43.1) (Bray et al., 2016) was used to quantify contig abundance [raw and transcripts per million (tpm) normalised counts]. The de novo assembled contigs were annotated using the BLAST (Altschul et al., 1997) program (BLASTx with default parameters was used to search protein databases using a translated nucleotide query) on three major protein databases (nr NCBI non-redundant protein database, SwissProt and UniProt UniRef90 Reference Clusters). An Expect value (i.e. E-value) cutoff of < 10⁻⁶ was used. A Trinotate annotation pipeline (Bryant et al., 2017) was used for functional annotation including homology search to known sequence data (BLAST+/SwissProt/Uniref90), protein domain identification (HMMER/PFAM), protein signal peptide and transmembrane domain prediction (ggnog/tmHMM), and comparison to currently curated annotation databases (EMBL UniProt ggnog/GO Pathways databases).

Differential transcript expression between experimental conditions was estimated using DESeq2 (Love et al., 2014). Within each pairwise comparison, only transcripts with a Benjamini-Hochberg adjusted p-value less than or equal to 0.05 were considered significantly differentially expressed.

For the chronic noise laboratory study, qPCR assays were developed for 61 genes that were identified as differentially expressed in any one of the three pairwise comparisons in the RNA-seq analysis: 1) hemocytes of crab exposed to chronic noise (HCE) compared with hemocytes of control crab (HCC), 2) hepatopancreas of crab exposed to chronic noise (HPE) compared with hepatopancreas of control crab (HPC) and 3) HPC compared with HCC. The pairwise comparison and the assembly (i.e. individual cell/tissue type or co-assembly) in which the transcript was identified as differentially expressed are provided in both Supplemental Table 1 and Supplemental File 1 . These genes were selected based on functional annotations related to biological processes of interest (e.g. immunity, defense, regulation of gene expression, metabolism and stress-response).

In a preliminary qPCR analysis, expression levels of these transcripts of interest (TOIs) were measured in cDNA pools representing the four groups (HCC, HCE, HPC, HPE) in this study. In a second qPCR analysis, levels of TOIs that appeared to be differentially expressed (>1.5-fold) in response to chronic noise exposure in either cell/tissue type in the pooled sample screen, and/or with functional annotations related to the aforementioned biological processes of interest were then measured in individual samples. Expression levels of 30 TOIs were assessed in the HPC and HPE samples (n=10 per group); expression levels of 20 of these TOIs were assessed in the HCC and HCE samples (n=10 per group) due to low/no expression of the other 10 TOIs in hemocytes.

For the field study conducted in 2017 (see Hall et al., 2021), a preliminary qPCR analysis was performed in which levels of the aforementioned 61 transcripts were measured in cDNA pools from hepatopancreas representing the eight groups (i.e. pools) in this study. This analysis was performed to help select the transcripts to be assessed in individual samples. In a second qPCR analysis, expression levels of 25 transcripts (19 of which were assessed in individual hepatopancreas samples from both the chronic noise laboratory and the field 2017 studies) were measured in 80 individual samples (n=10 per each of the eight groups). Eight of these transcripts were selected as they were qPCR confirmed as being significantly differentially expressed or showed a non-significant trend (0.05 < p ≤ 0.1) in the HPE group compared with the HPC group. The application of these candidate noise-responsive biomarkers to the field 2017 samples would determine if their expression levels are reproducible going from the laboratory to the field (i.e. if they are consistent noise-responsive molecular biomarkers). The additional 11 transcripts common to both studies were selected as they have functional annotations associated with the aforementioned biological processes of interest, and thus confirming their response to noise in both the laboratory and to both 2D and 3D seismic noise was important. The final 6 transcripts (totalling to 25) were selected as they appeared to be differentially expressed (at least 1.5-fold range among the 8 groups) in response to seismic noise in the pooled sample screen.

First-strand cDNA templates for qPCR were synthesized for each individual sample (n=40, chronic noise laboratory study; n=80, field 2017 study) in 20 μl reactions from 1 µg of DNaseI-treated, column-purified total RNA using random primers (250 ng; Invitrogen/Thermo Fisher Scientific), dNTPs (0.5 mM final concentration; Invitrogen/Thermo Fisher Scientific) and M-MLV reverse transcriptase (200 U; Invitrogen/Thermo Fisher Scientific) with the manufacturer’s first-strand buffer (1X final concentration) and DTT (10 mM final concentration) at 37°C for 50 min. In all cases where pooled samples were analyzed, a subsample of each individual cDNA template (n=10) from a given group (n=4, chronic noise laboratory study; n=8, field 2017 study) was included in that pool, with each individual contributing equally.

All PCR amplifications [i.e. for primer quality control (QC) testing, endogenous control (normalizer) testing and experimental qPCR analyses] were performed in 13 μl reactions using 1X Power SYBR Green PCR Master Mix (Applied Biosystems/Thermo Fisher Scientific), 50 nM of both the forward and reverse primers, and the indicated cDNA quantity (see below). For all primer pairs in this study, an annealing temperature of 60°C was used. Amplifications were performed using the ViiA 7 Real Time PCR system (384-well format) (Applied Biosystems/Thermo Fisher Scientific). The real-time analysis program consisted of 1 cycle of 50°C for 2 min, 1 cycle of 95°C for 10 min and 40 cycles of 95°C for 15 sec and 60°C for 1 min (i.e. primer annealing and extension were combined into one step), with fluorescence detection at the end of each 60°C step, and was followed by dissociation curve analysis.

To confirm the identities of all transcripts subjected to qPCR analyses and to identify the coding sequences (CDS), a BLASTx 2.8.0+ search of the non-redundant (nr) protein sequences database using a translated nucleotide query was performed between March and July, 2018 ( Supplemental File 1 ). All primers are located in the CDS and in an area which overlapped with that of the best BLASTx-identified sequence. The amplicon size range was 87-150 bp. All primers were designed with a melting temperature (T_m) of 60°C, mostly using Primer3 (Koressaar and Remm, 2007; Untergasser et al., 2012); however, in the case of gene paralogues/isoforms [e.g. spastic ataxia of Charlevoix-Saguenay (sacsin) (sacs), toll-like receptor (tlr)] they were hand-designed in paralogue/isoform-specific areas (i.e. in an area with ≥ 3 bp difference between them) to ensure specificity. Each primer pair was quality control (QC)-tested to ensure that a single product was amplified (dissociation curve analysis) and that there was no primer-dimer present in the no-template control. Amplicons were electrophoretically separated on 2% agarose gels and compared with a 1 kb plus ladder (Invitrogen/Thermo Fisher Scientific) to verify that the correct size fragment was being amplified. Finally, amplification efficiencies (Pfaffl, 2001) were calculated for cDNA pools representing the two groups in which the TOI had been identified as differentially expressed in RNA-seq studies; however, with the exception of period circadian regulator isoform 1 (per1), the TOIs identified in HCE compared with HCC analysis 2 were QC tested in HPE and HPC cDNA pools due to low/no expression levels in hemocytes. Standard curves were generated using a 5-point 1:3 dilution series starting with cDNA representing 10 ng of input total RNA. The reported efficiencies ( Supplemental Table 1 ) are an average of the two values with the exception of cytochrome P450 CYP330A1 (cyp330a1), gamma-interferon induced thiol reductase 1 (gilt1), gamma-interferon induced thiol reductase 3 (gilt3), leucine-rich repeat-containing protein 24 (lrrc24), myoferlin (myof) and per1 (HPC only), and macrophage mannose receptor 1 (mrc1) and clotting factor B (cfb) (HCC only) due to low/no expression levels in the other group. These same efficiencies were used in the 2017 field study. The sequences, amplicon sizes and efficiencies for all primer pairs used in the qPCR analyses are presented in Supplemental Table 1 .

Expression levels of the TOIs analyzed herein were normalized to expression levels of two endogenous control transcripts, namely 60S ribosomal protein L32 (alias ribosomal protein 49) (rpl32) and elongation factor 1-alpha (ef1a). These transcripts had been previously shown to be stably expressed in the hepatopancreas of field-collected crab that had been subjected to seismic surveying in 2016 and in 2017 (also analyzed in the current study) using geNorm (Vandesompele et al., 2002) analyses (Hall et al., 2021). In the chronic noise laboratory study, they were also stably expressed in hepatopancreas samples with average fluorescence threshold cycle (C_T) values between control and chronic noise exposed samples that were nearly identical, and in hemocyte samples with average C_T values between control and chronic noise exposed samples < 0.2 cycle different. However, average C_T values between hepatopancreas and hemocytes samples were ~0.7 cycle different (data not shown). Therefore, expression levels of the TOIs were compared within a given cell/tissue type only (i.e. HCE compared with HCC only; HPE compared with HPC only).

In all four experimental qPCR analyses, cDNA representing 4 ng of input RNA was used as template in the PCRs. On each plate, for every sample, the TOIs and endogenous controls were tested in triplicate and a no-template control was included. In the two individual sample qPCR analyses, as expression levels of a given TOI were measured across multiple plates, a plate linker sample (i.e. a sample that was run on all plates in a given study) was also included to ensure there was no plate-to-plate variability. The relative quantity (RQ) of each transcript was determined using the ViiA 7 Software Relative Quantification Study Application (Version 1.2.3) (Applied Biosystems/Life Technologies), with normalization to both rpl32 and ef1a transcript levels, and with amplification efficiencies incorporated. For each TOI, for the given cell/tissue type, the sample with the lowest normalized expression (mRNA) level was set as the calibrator sample (i.e. assigned an RQ value = 1.0).

For the chronic noise laboratory study, transcript expression data were analyzed with non-parametric statistics as there were some cases where values did not exhibit homogeneity of variance (Levene’s Test for Equality of Variances). The Mann-Whitney test was used to assess if there were any significant differences in expression levels of a given TOI in response to chronic noise exposure within a given cell/tissue type. In all cases, p < 0.05 was considered to be statistically significant. All data are expressed as mean ± standard error (S.E.).

For the field study conducted in 2017, transcript expression data were log₁₀ transformed and one-way ANOVA followed by Tukey’s B post-hoc test were used to assess expression levels of a given TOI across the eight groups. In all cases, p < 0.05 was considered to be statistically significant. All data are expressed as mean ± S.E.

To test the utility of selected transcripts to act as biomarkers of seismic noise in the field, we also examined transcript expression in multivariate space using principal component analyses (PCAs). Transcript expression data from field collected samples in 2017 were scaled and centered using the scale function in R prior to analyses. PCAs were performed using prcomp function in R and plotted using the ggbiplot R package (Vu, 2011). Individual PCA scores on PC axes 1 and 2 were compared among sample groups using one-way ANOVA or t-test (depending on comparison) in R, and for any significant comparisons using ANOVA, Tukey’s post-hoc test was used to test pairwise differences between groups using the TukeyHSD function in R. The contributions of different transcripts (loadings) on PC1 and PC2 were examined and the top 5 transcripts that loaded the highest on each PC axis were identified. A PCA was performed on all field samples from 2017 using 25 transcripts for which expression was quantified in the hepatopancreas tissues. Correlations in transcript expression patterns among all transcripts were examined using cor.test function and plotted using corrplot package in R. The same methods for the PCA were applied to subsets of these data, where we ran a separate PCA for: 1) only samples from Carson Canyon, 2) only samples exposed to 3D seismic noise north of Carson Canyon, and 3) only samples exposed to no seismic noise from two geographically separate locations. Transcripts loading highly on PC 1 and PC 2 in each of the four PCAs were compared using a Venn diagram in the R package VennDiagram.

RNA-seq libraries were generated for hemocytes samples (10 each of HCC and HCE) and for hepatopancreas samples (10 each of HPC and HPE) from the same 20 individual snow crab (see Supplemental File 2 ; Sequencing tab). The samples have been deposited in the NCBI BioSample database under accession numbers SAMN10833272-SAMN10833281 (HCC), SAMN10833282- SAMN10833291 (HCE), SAMN10833292-SAMN10833301 (HPC) and SAMN10833302-SAMN10833311 (HPE). All libraries have been deposited in the NCBI Sequence Read Archive (SRA) under accession number PRJNA517457. Individual assemblies for hemocytes and hepatopancreas sample reads produced 431,293 and 422,579 putative transcripts, and 260,702 and 249,325 putative genes, respectively. A hemocytes-hepatopancreas co-assembly produced a total of 440,481 putative transcripts and 263,826 putative genes.

The DET lists, including log₂ fold-change values, transcripts per million (tpm) counts, BLAST results and functional annotations for the following five comparisons are provided in Supplemental File 3 . For the individual cell/tissue type assemblies (i.e. RNA-seq analysis 1), a comparison of the transcriptomes for HCE samples with HCC samples, and for HPE samples with HPC samples using DESeq2 identified 34 and 45 DETs, respectively. The identity of only 11 and 17 of these DETs, respectively, could be determined using BLAST. Nucleotide sequences (in FASTA format) for the DETs are provided Supplemental File 4 and Supplemental File 5 , respectively. For the hemocytes-hepatopancreas co-assembly (i.e. RNA-seq analysis 2), each of the aforementioned pairwise comparisons identified over 300 DETs; however, less than 100 were of known identity (i.e. associated gene or protein name due to significant BLAST hit). In addition, a comparison of the transcriptomes for HPC samples with HCC samples identified over 100,000 DETs, including many with immune-relevant functions. Nucleotide sequences for the DETs from the three pairwise comparisons from RNA-seq analysis 2 are provided in Supplemental File 6 . Summary information for the co-assembly is provided in Supplemental File 2 .

A subset (n=40) of candidate noise-responsive snow crab transcripts with BLASTx hits (and potential functional annotation) were selected for qPCR analyses. Of these, 9 and 13 were identified in the HCE compared with HCC RNA-seq analyses 1 and 2, respectively; 11 and 7 were identified in the HPE compared with HPC RNA-seq analyses 1 and 2, respectively. In addition, 21 snow crab transcripts primarily with functional annotations related to the immune response (HPC compared with HCC RNA-seq analysis 2) were selected for qPCR analyses (n=61 total) ( Supplemental Table 1 , Supplemental File 3 ). A summary of the results of the BLASTx (performed for primer design), RNA-seq and qPCR analyses for these transcripts is provided in Supplemental File 1 .

Expression levels of these 61 transcripts were initially measured in a pooled HCC sample and in a pooled HCE sample ( Supplemental Figure 1 ), and in a pooled HPC sample and in a pooled HPE sample ( Supplemental Figure 2 ). Transcripts are grouped by the specific pairwise comparison in which they were identified as differentially expressed in the RNA-seq analyses and are sorted by Gene ID number (see Supplemental File 1 ). Of the 9 transcripts selected for qPCR from the HCE and HCC comparison (RNA-seq analysis 1), fold-change values for 7/9 in the pooled hemocytes sample qPCR analysis [ Supplemental Figures 1A (A–I) ] agreed with the direction of change in the RNA-seq analysis. Of the 13 transcripts selected for qPCR from the HCE and HCC comparison (RNA-seq analysis 2), only per1 was expressed at levels that could be detected in the pooled hemocytes sample qPCR analysis [ Supplemental Figure 1A (J) ]; fold-change values for per1 did agree with the RNA-seq direction of change. Of the 11 and 7 transcripts selected for qPCR from the HPE and HPC comparison (RNA-seq analyses 1 and 2), fold-change values for 8/11 and 7/7 in the pooled hepatopancreas sample qPCR analysis ( Supplemental Figures 2C, D ) agreed with the RNA-seq direction of change, respectively (see Supplemental File 1 for a summary of the RNA-seq and qPCR results).

Expression levels of transcripts that appeared to be differentially expressed in response to noise in the pooled sample qPCR analyses (see above), and/or with functional annotations related to immunity, regulation of gene expression, metabolism and stress-response were then measured in individual samples from the HCC and the HCE groups (n=20) ( Figure 1 ). Levels of these 20 transcripts and an additional 10 transcripts that were not measured in hemocytes (due to low/no expression levels) were measured in individual samples from the HPC and HPE groups ( Figure 2 ). GO annotations for the putative human orthologue [or for crab spp. for cfb, cuticle protein CBM (cbm) and hemocyanin C chain (hcyc)] of these transcripts were collected from QuickGO (Binns et al., 2009) ( Supplemental Table 2 ).

In the HCE group compared with the HCC group, six transcripts were either significantly differentially expressed or showed a non-significant trend (0.05 < p < 0.10) ( Figures 1A–F ). Expression levels of two transcripts were induced by chronic noise; regulator of G-protein signaling 2 (rgs2) was 2-fold significantly higher and cfb was 1.5-fold higher (p=0.052) in the HCE group. Expression levels of four transcripts were suppressed by chronic noise; one sacs paralogue (DN162290_c0_g1_i1) was 1.3-fold significantly lower and a second paralogue (DN143706_c0_g1_i1) was 1.3-fold lower (p=0.063) in the HCE group. Myof (18-fold) and per1 (360-fold) also showed a non-significant trend (p=0.052) towards lower expression; however, expression levels of both transcripts were highly variable. There were no significant differences in expression levels of the other transcripts ( Figures 1G–T ).

In the HPE group compared with the HPC group, eight transcripts were either significantly differentially expressed or showed a non-significant trend (0.05 < p ≤ 0.1) ( Figures 2A–H ). Two transcripts, rgs2 (1.9-fold) and ATP synthase F1 subunit beta (atp5f1b; 1.8-fold) were significantly higher expressed in the HPE group. Five transcripts, glutathione peroxidase 3 (gpx3; 2.5-fold), activating transcription factor 3 (atf3; 1.8-fold), cathepsin A (ctsa; 1.5-fold), tetraspanin-like protein CD9 (cd9; 1.5-fold), and cbm (1.4-fold) showed a non-significant trend towards higher expression in the HPE group. Sulfotransferase 1C4 (sult1c4) was 29-fold significantly lower expressed in the HPE group. There were no significant differences in expression levels of the other transcripts ( Figures 2I-AD ).

To provide insight into how the 61 transcripts from the chronic noise laboratory study may respond to 2D and 3D seismic surveying in the field, their expression levels were initially measured in hepatopancreas cDNA pools representing the eight groups (i.e. BSC, 1DASC, 2DASC, 6WASC, BSL, 6WASL, DSNC and 4WASNC) from the field study conducted in 2017. These data are presented in the same order as Supplemental Figure 2 ( Supplemental Figure 3 ).

In a second qPCR analysis, levels of a subset (n=25) of these transcripts were assessed in individual samples from the 8 groups in the 2017 study ( Figure 3 ). Of the 7 transcripts that were either significantly higher expressed or showed a non-significant trend (0.05 < p ≤ 0.1) towards higher expression in the HPE group compared with the HPC group ( Figures 2A–G ), only gpx3 was significantly higher expressed with seismic surveying in the field 2017 study (i.e. trending higher in the 2DASC group and significantly higher in the DSNC group) ( Figure 3C ). Levels of two (rgs2 and atf3) of these transcripts showed the opposite effect with rgs2 levels significantly lower in the 1DASC, 6WASC and 4WASNC groups compared with the BSC group ( Figure 3A ), and with atf3 showing a non-significant trend (0.05 < p < 0.10) of lower expression in the 1DASC, DSNC and 4WASNC groups ( Figure 3D ). Sult1c4, which was significantly lower expressed in the HPE group compared with the HPC group ( Figure 2H ), was not significantly differentially expressed in the field study ( Figure 3H ).

Of the transcripts that were not significantly differentially expressed in the HPE group compared with the HPC group, signal transducer and activator of transcription (stat) was significantly higher expressed in the 4WASNC group compared with the 2DASC and 6WASC groups ( Figure 3J ), pyruvate dehydrogenase kinase, mitochondrial (pdk2) was significantly lower expressed in the 2DASC compared with the 6WASL and the 6WASC groups ( Figure 3P ), histone-lysine N-methyltransferase 2D (kmt2d) was significantly lower expressed in the DSNC compared with the 6WASL group ( Figure 3Q ), and myof was significantly higher expressed in the 4WASNC group compared with the DSNC group ( Figure 3S ); gilt1 showed a non-significant trend (0.05 < p < 0.10) of higher expression in the 2DASC, DSNC and 4WASNC groups ( Figure 3M ).

Six additional transcripts that appeared to be differentially expressed in response to seismic surveying in the pooled sample screen were also assessed. GO annotations for the putative human orthologue of these transcripts were also collected from QuickGO (Binns et al., 2009) ( Supplemental Table 2 ). Broad complex protein (brc) was significantly lower expressed in the BSC, 2DASC and DSNC compared with the 6WASL, 6WASC and 4WASNC groups ( Figure 3T ). Cationic amino acid transporter (cat) showed a non-significant trend (0.05 < p < 0.10) of lower expression in the 2DASC, DSNC and 4WASNC groups ( Figure 3U ). The four additional transcripts [cyp330a1, chromodomain helicase DNA binding protein 1 (chd1), spectrin alpha chain-like protein (sptan1) and lrrc24] were not significantly differentially expressed ( Figures 3V–Y ).

While rgs2 was shown to be significantly upregulated by chronic noise in both hepatopancreas and hemocytes in our tank-based study, it did not appear to be upregulated by seismic exposure in our field-based study. Rather, hepatopancreas rgs2 transcript was significantly downregulated 1 day and 6 weeks after seismic exposure at the Carson Canyon site compared with pre-exposure controls. In addition, rgs2 expression was significantly higher in hepatopancreas of crabs from Carson Canyon compared with those from another geographic location (Lilly Canyon) when seismic exposure was absent at both sites. In mice, rgs2 is involved in regulating T cell activation and antiviral immunity (Oliveira-dos-Santos et al., 2000). In addition to its immune-related functions, mouse rgs2 is involved in regulating pancreatic β-cell apoptosis and function (Dong et al., 2017). Mammalian RGS2 protein is induced by several forms of stress including ischemia, and is thought to play a key role in modulating the stress response (Endale et al., 2010; Wang and Chidiac, 2019). Supporting this hypothesis, functional annotations associated with rgs2 include “response to amphetamine” and “response to ethanol” ( Supplemental Table 2 ). Therefore, it is not surprising that we found rgs2 to be induced in chronic noise exposed snow crab hepatopancreas and hemocytes, and suggests that rgs2 will be a valuable biomarker in future research on crab stress responses. While our field trial had unexpected rgs2 expression results, it is important to note that rgs2 was among the top five qPCR genes driving the significant differences in multivariate analyses (i.e., on PC axis 2 in PCA) between Carson Canyon pre- and post-seismic exposure groups. Therefore, this gene may be a valuable biomarker in future crab environmental transcriptomic studies.

RNA-seq identified cfb as significantly higher expressed in control hemocytes compared with control hepatopancreas and was shown by qPCR to be upregulated by chronic noise (p = 0.052) in hemocytes in the current study. Cfb is an intracellular hemocyte-expressed serine protease involved in the hemolymph coagulation response of horseshoe crab (Limulus) to bacterial endotoxin [i.e. lipopolysaccharide (LPS)]; this is thought to be important in the animal’s hemostasis and antimicrobial defense (Nakamura et al., 1986; Iwanaga et al., 1992; Muta et al., 1993) ( Supplemental Table 2 ). Our results suggest that chronic noise may disrupt snow crab hemocyte function in hemolymph coagulation and antibacterial defense. Münzel et al. (2017) reported that aircraft noise (administered to mice from loudspeakers over four days) significantly increased macrophages within the vasculature. Based on these collective results, we speculate that noise may impact phagocytic immune cells (e.g. hemocytes, macrophages) of diverse animal taxa.

As previously stated, myof was identified herein as downregulated by chronic noise. In our qPCR study with field-collected hepatopancreas samples, while myof was not differentially expressed before and after 2D seismic exposure at the Carson Canyon site, this gene was significantly downregulated during 3D seismic exposure compared with four weeks after 3D seismic exposure at the North of Carson Canyon site. Collectively, our results suggest that myof is a noise-responsive biomarker in snow crab immune cells and tissues that may be useful in studying effects of 3D seismic exposure on crab immune physiology. At the Carson Canyon site, there was not a hepatopancreas sampling time point between pre-seismic and one day post-seismic exposure. It is possible that myof expression may have been suppressed during 2D seismic exposure (as observed in the 3D seismic exposure study), and that future research would benefit from additional sampling time points (e.g. during and shortly after 2D seismic exposure).

While Myoferlin has primarily been studied in mammalian muscle, where it plays roles in myoblast fusion and endocytosis by endothelial cells (Doherty et al., 2005), it is now known to also have important functions in many mammalian cell types including phagocytes (e.g. macrophages and neutrophils), adipocytes and cancer cells (Miyatake et al., 2018; Nozato et al., 2020; Gupta et al., 2021). Functional annotations associated with myof include “plasma membrane repair” and “extracellular exosome” ( Supplemental Table 2 ); extracellular exosomes (also called extracellular microvesicles) are important components of several biological processes including tissue homeostasis, organ remodeling, and cell-cell communication (Ratajczak and Ratajczak, 2020 and references therein). For example, Myoferlin plays a key role in mouse phagocyte function by regulating lysosomal exocytosis (Miyatake et al., 2018). In addition, myof knockout mice on a high-fat diet had attenuated obesity, altered glucose metabolism, and lower adipose inflammation associated with fewer inflammatory macrophages compared with those on a normal diet (Nozato et al., 2020). In human pancreatic cancer cells, Myoferlin is a membrane repair factor that protects against membrane damage and regulates lysosome function, with lower Myoferlin expression alleviating malignant phenotypes (Wang et al., 2013; Zhu et al., 2019; Gupta et al., 2021). If Myoferlin function is conserved between mammals and crustaceans, then the suppressed myof expression that we observed in chronic noise exposed crab hepatopancreas and hemocytes (in the tank-based study) and during 3D seismic exposure (in the field-based study) could influence the repair of membranes damaged by mechanical stress injury from loud sounds, as well as other biological processes such as glucose metabolism and inflammation. Further research is needed to understand the physiological and fitness-related consequences of suppressed myof in noise-exposed crab hepatopancreas and hemocytes.

Some snow crab hepatopancreas and hemocyte genes identified in the current study were little studied in marine invertebrates previously, but known to function in vertebrate immune and/or stress responses. For example, in the current study, snow crab atf3 was upregulated (p = 0.052) in hepatopancreas following chronic noise exposure (in the tank-based study) and also differentially expressed (p = 0.079) between groups in the field-collected hepatopancreas samples from various seismic-exposed and non-exposed sites; as well, atf3 was one of the five genes driving the non-significant (p > 0.05) differences on axis 1 of the PCA including samples from all eight field sampling sites. While there is a paucity of information available on marine invertebrate atf3, Atlantic cod (Gadus morhua) atf3 mRNA was shown to be significantly upregulated (over 60-fold) in spleen 24 h after stimulation with the double-stranded RNA analog polyriboinosinic polyribocytidylic acid [(pIC), a viral mimic] (Feng and Rise, 2011). In mammals, ATF3 is stress-induced and plays critical roles in the regulation of immunity, glucose metabolism, and cancer (Ku and Cheng, 2020). Functional annotations associated with atf3 include GO terms involving response to endoplasmic reticulum stress and regulation of apoptosis ( Supplemental Table 2 ). Mouse atf3 and atf4 transcripts were shown to be strongly induced in the inner ear neurons after acoustic trauma (i.e. noise exposure) (Milon et al., 2021). The transcription factors ATF3 and ATF4 are key parts of the integrated stress response pathway that may, after trauma, lead to either the restoration of homeostasis or cell death (Pakos-Zebrucka et al., 2016; Milon et al., 2021). Based on these results, we hypothesize that the integrated stress response pathway is activated in the crab hepatopancreas by chronic noise, and that transcripts in this pathway (e.g. atf3) may be suitable molecular biomarkers for studying the effects of seismic and other noise exposure on wild crabs. As well, since mammalian ATF3 influences the expression of interferon gamma (Rosenberger et al., 2008), we propose that interferon signaling may be part of a mechanism by which snow crab hepatopancreas responds to loud sounds/noise. The presence of gilt1 among the top five genes driving the significant difference between post-seismic time points at the Carson Canyon sampling site ( Figure 5E ) seems to support this idea.

Although kmt2d was RNA-seq identified (in analysis 1) as downregulated in hemocytes of crabs exposed to chronic noise, this gene could not be confirmed as noise-responsive in either hemocytes or hepatopancreas using qPCR. GO terms associated with kmt2d include “histone H3-K4 methylation”, “regulation of DNA-templated transcription”, “oogenesis”, and “response to estrogen” ( Supplemental Table 2 ). In the field-based component of the current study, kmt2d was shown to be significantly lower expressed during 3D seismic exposure (at the North of Carson Canyon site) compared with one of the control non-seismic sample groups (6WASL). Recently, RNA-seq was used to identify the transcript encoding a different Histone-lysine N-methyltransferase, (H3 lysine-79 specific) as being upregulated in horseshoe crab (Tachypleus gigas) hemocytes exposed to lipopolysaccharide [(LPS), a bacterial pathogen-associated molecular pattern (PAMP)] (Sarmiento et al., 2021). While it is known that human KMT2D has cancer-epigenetic function (e.g. in tumorigenesis and hematological malignancies such as non-Hodgkin lymphoma) (Dhar and Lee, 2021; Yamagishi, 2022), the function of crustacean kmt2d is not known. There is an extensive literature on the influence of noise stress on mammalian immune function and association with various diseases including non-Hodgkin lymphoma (reviewed by Zhang et al., 2021). Assuming KMT2D function is conserved in mammals and crustaceans, it is possible that altered histone modification associated chromatin remodeling may be part of an epigenetic mechanism by which 3D seismic stress could influence the snow crab hepatopancreas in the wild.

The snow crab transcript identified herein as brc (contig hepatopancreas_DN112978_c0_g1_i3) has significant similarity (BLASTx against nr database) to proteins including Zinc finger protein 169-like (Homarus americanus), Longitudinals lacking protein (Portunus trituberculatus), Zinc finger protein Helios (C. opilio), and Broad-complex protein isoform 4 (Penaeus vannamei) (data not shown), with associated GO terms including “metal ion binding” and “retinoic acid-responsive element binding” ( Supplemental Table 2 ). This brc-like transcript was RNA-seq identified (in analysis 1) as down-regulated in chronic noise exposed hepatopancreas compared with control hepatopancreas. In our qPCR study with field-collected hepatopancreas samples, brc was among the most informative biomarkers, being significantly lower expressed during 3D seismic compared with 4 weeks after 3D seismic, and also significantly lower expressed before and 2 days after 2D seismic compared with 6 weeks after 2D seismic. As well, brc was one of the top five qPCR genes driving the significant differences in multivariate analyses (i.e., in PCA on PC 2) between Carson Canyon seismic exposure groups. In Drosophila, brc and lola (encoding Longitudinals lacking protein) are members of the BTB-zinc finger protein family of transcription factors that play roles in embryonic and ovarian development, e.g. repressing apoptosis (Ohsako et al., 2003; Zhao et al., 2022); as well, lola is involved in the formation of a sexually dimorphic courtship circuit in Drosophila (Sato et al., 2019). In ridgetail white prawn (Exopalaemon carinucauda), lola was identified as a gonad development associated gene that mapped to a sex-related quantitative trait locus (Lv et al., 2020). In addition to roles in sex and development, members of this protein family also have immune-related functions; for example, the human zinc finger transcription factor Helios is a marker used to identify thymically-derived regulatory T-cells (T_reg) and plays a critical role in the maintenance of a stable T_reg phenotype in tumors (Tanoue et al., 2016; Wang et al., 2021). If the brc-like gene involved in our study also functions in development and/or immunity, then it may be a useful molecular biomarker for future studies of seismic and other environmental stress on snow crab physiology.

In addition to the previously discussed genes associated with the immune response and inflammation, many other putative defense-relevant transcripts were RNA-seq-identified and qPCR-assayed in the current study. Immune-related functional annotations (GO terms) associated with genes represented in Figures 1 – 6 include “adaptive immune response” [cathepsin L (ctsl)], “immune response” [ctsl, interferon regulatory factor (irf), il16, and tlr], “innate immune response” [E3 ubiquitin-protein ligase TRIM38 (trim38), toll-interacting protein (tollip), and tlr], “lipopolysaccharide immune receptor activity” (tlr), and several related to B cell differentiation [stat, interferon regulatory factor binding protein (irf2bp), irf] ( Supplemental Table 2 ). Of these genes, stat was differentially expressed between groups in the field study, and il16 and tlr were among the top five highest loading genes in multivariate analyses. Immune-relevant gene sequences and qPCR assays generated in this study represent important new resources for future studies of snow crab defense responses to seismic and other environmental stressors.

Both gilt1 and gilt3 were RNA-seq identified (in analysis 2) as significantly much higher expressed in control hepatopancreas compared with control hemocytes. This agrees with Huang et al. (2015), who showed that mud crab (Scylla paramamosain) gilt1 and gilt3 constitutive transcript expression levels were highest in hepatopancreas (of 11 tissues tested), with low and moderate expression of gilt1 and gilt3, respectively, in hemocytes. While our qPCR study did not detect gilt3 in hemocytes, gilt1 was non-significantly (p > 0.05) 2.5-fold upregulated in noise-exposed crab hemocytes compared with control hemocytes; qPCR showed that neither gilt1 nor gilt3 appeared to respond to noise in hepatopancreas. In our field study, gilt1 appeared to be upregulated (p = 0.087) in hepatopancreas two days after 2D seismic exposure compared with the three other Carson Canyon sample groups (i.e. before seismic, and one day and six weeks after seismic); gilt1 showed no change in expression between the two sampling times at the non-seismic control site (Lilly Canyon), and no response to 3D seismic exposure at the North of Carson Canyon site. In our multivariate analysis of Carson Canyon hepatopancreas transcript expression data, gilt1 was among the top five genes driving the significant difference on PC axis 2 between samples collected two days versus six weeks after 2D seismic exposure.

GILT, which catalyzes the reduction of protein disulfide bonds, plays roles in numerous defense-relevant processes in mammals including antigen presentation, regulation of the cellular redox state, response to bacterial infection, and viral entry into host cells (Rausch and Hastings, 2015; Majdoul and Compton, 2022) ( Supplemental Table 2 ). Indeed, GILT is known to function in innate immunity (e.g. response to bacterial infection) and regulation of oxidative stress in diverse animals including invertebrates such as planarians (Gao et al., 2021) and lower vertebrates such as sturgeon (e.g. You et al., 2018). In Atlantic salmon (Salmo salar) fed diets with partial replacement of fish meal with solvent-extracted camelina meal for 16 weeks, gilt transcript expression in the distal intestine was significantly correlated with some histopathological signs of inflammation (Brown et al., 2016). In recent years, gilt-like gene expression has been studied in crustaceans. For example, Chinese mitten crab gilt mRNA expression in hemocytes was significantly induced after infection with bacterial pathogens Vibrio parahaemolyticus and Staphylococcus aureus (Nan et al., 2022). As well, red swamp crayfish (Procambarus clarkii) gilt mRNA expression was shown to be over 2-fold higher in hepatopancreas compared with hemocytes (agreeing with our RNA-seq results), and also strongly upregulated in hepatopancreas following challenge with PAMPs, namely pIC and peptidoglycan (Liu et al., 2019). In shrimp (Penaeus monodon), the GILT catalyzed reduction of viral protein disulfide bonds restricted white spot syndrome virus infection (Thipwong et al., 2019). As well, mud crab gilt1 and gilt3 transcript expression in hepatopancreas was shown to be significantly upregulated by Vibrio parahaemolyticus infection (Huang et al., 2015). Given GILT’s conserved functions across animal taxa, and the relatively high expression of gilt1 and gilt3 in snow crab hepatopancreas observed herein, it seems likely that these gilt family members function in defense (e.g. inflammation and/or innate immune responses to pathogens). The gilt1 expression change after 2D seismic exposure at the Carson Canyon site, then, could potentially influence the crabs’ immune function and health. The regulation and function of snow crab gilt1 and gilt3, and their potential use as biomarkers of exposure to environmental stressors such as seismic surveys, warrant further investigation.

In the current study, sacs was shown by RNA-seq and qPCR to be significantly downregulated in snow crab hemocytes exposed to chronic noise; however this gene was not differentially expressed in our lab-based and field-based hepatopancreas qPCR experiments. The human sacs gene encodes a molecular chaperone (Anderson et al., 2011) associated with early-onset spastic ataxia (Engert et al., 2000), with functional annotations including “Hsp70 protein binding”, “proteasome binding”, and “protein-folding chaperone binding” ( Supplemental Table 2 ). Recently, a CRISPR-Cas9-generated zebrafish sacs knockout strain exhibited characteristics similar to mammalian models of sacs-associated spastic ataxia (e.g. increased oxidative stress, mitochondrial dysfunction, motor impairment; Naef et al., 2021). In some teleost fish species, sacs has also been shown to be responsive to viral or viral-like stimuli; for example, sacs transcript was induced in Atlantic cod brain following stimulation with pIC (Rise et al., 2010), and in Atlantic salmon macrophage-like cells after infection with the viral pathogen that causes Infectious Salmon Anemia (ISA) (Workenhe et al., 2009). Based on these results, Rise et al. (2012) speculated that sacs could be part of the core set of antiviral fish genes discussed by Verrier et al. (2011). Importantly, a proteomic study of the reef coral Pocillopora acuta identified Sacsin (an HSP40 family member) to be associated with elevated temperature, and a sacsin-like gene may be involved in the responses of Acropora corals to environmental stressors (Mayfield et al., 2018; Fuller et al., 2020; Takahashi-Kariyanzono and Terai, 2021). Collectively, our results alongside literature associated with noise-responsive genes identified in the current study that are known to be involved in both stress and antiviral responses (i.e. sacs and atf3) suggest that these genes may be valuable molecular biomarkers for assessing immune-related impacts of chronic noise and other environmental stressors on snow crab and other marine animals.

GPx3, a selenoprotein that catalyzes the reduction of H₂O₂ [hydrogen peroxide, a reactive oxygen species (ROS)] to H₂O and O₂, is a key component of the redox system that protects cells from oxidative damage (Hernández-Aguirre et al., 2022). Functional annotations associated with gpx3 include “selenium binding”, “oxidoreductase activity”, “cellular oxidant detoxification”, response to lipid hydroperoxide”, and “response to oxidative stress” ( Supplemental Table 2 ). While gpx3 was RNA-seq identified as downregulated in noise-exposed crab hemocytes, this transcript was not detectable by qPCR in hemocytes. However, we found that gpx3 was upregulated (p = 0.052) in noise exposed crab hepatopancreas. Further, in the field-based qPCR studies, gpx3 was significantly upregulated during 3D seismic exposure compared with four weeks later (at the North of Carson Canyon site) and also one of the top five genes driving differences in multivariate analysis (i.e., on PC 2 in PCA) between 2D seismic exposure groups (at the Carson Canyon site). Previous studies have shown that marine invertebrate Glutathione peroxidase responds to environmental stressors. For example, cadmium exposure of the freshwater crab Sinopotamon henanense significantly increased ROS levels and activity of Glutathione peroxidase and other antioxidant enzymes [Superoxide dismutase (SOD) and Catalase (CAT)] in hemocytes (Zhou et al., 2016). At the transcript expression level, Cheng et al. (2020a) showed that mud crab (Scylla paramamosain) gpx3 had relatively high constitutive expression in hepatopancreas and hemocytes, and was significantly upregulated (along with sod and cat) in hepatopancreas of crabs exposed to acute nitrite stress. A mud crab hepatopancreas RNA-seq study found that gpx3 and other oxidative stress relevant transcripts (thioredoxin, thioredoxin reductase, cytochrome p450) were upregulated after air exposure, an important environmental stressor for crab aquaculture and transport (Cheng et al., 2020b). As well, crayfish (Cherax quadricarinatus) gpx3 transcripts were shown to be upregulated in nerve cord by elevated temperature and hypoxia (Hernández-Aguirre et al., 2022), and Zhikong scallop (Chlamys farreri) gpx3 upregulation in kidney after exposure to paralytic shellfish toxin (PST)-producing dinoflagellates may protect this organ from PST-associated oxidative stress (Hlaing et al., 2021). While the aforementioned crustacean gpx3 expression studies showed induction following exposure to various environmental stressors (in agreement with our observation of snow crab gpx3 upregulation by chronic noise in hepatopancreas), gpx3 was found to be significantly downregulated in Chinese mitten crab (Eriocheir sinensis) gill after salinity stress (Li et al., 2014). Collectively, these results support gpx3 as a robust environmental stress responsive biomarker in crustaceans, and suggest that gpx3 induction may be part of a mechanism by which snow crab hepatopancreas is protected from oxidative stress associated with chronic noise exposure. In light of these gpx3 results, and the many immune-relevant snow crab genes identified as noise-responsive in our study, it is notable that Zhang et al. (2021) hypothesized that noise may alter the human immune system through modulation of oxidative stress.

In the current study, atp5f1b was significantly upregulated in snow crab hepatopancreas after chronic noise exposure, and was also one of the five genes driving the non-significant (p > 0.05) differences on axis 2 of the PCA including samples from all eight field sampling sites. Chronic low pH (i.e. acidic) stress significantly influenced the expression of many Chinese mitten crab hepatopancreas transcripts involved in oxidative phosphorylation (e.g. atp5f1) and in glycolysis/gluconeogenesis and pyruvate metabolism (e.g. pdhA, pdhD) (Wang et al., 2020). While our tank-based chronic noise qPCR study did not confirm pdk2 as noise-responsive in either hemocytes or hepatopancreas, in the field qPCR study, pdk2 was differentially expressed between groups [i.e. significantly lower expressed two days after 2D seismic exposure compared with six weeks after 2D seismic at the Carson Canyon site, and also compared with one of the control non-seismic sample groups (6WASL)]. PDK2 plays a key role in regulating cellular energy metabolism and homeostasis (Shi and McQuibban, 2017), and associated GO terms include “mitochondrial pyruvate dehydrogenase complex” and “cellular response to reactive oxygen species” ( Supplemental Table 2 ). A recent RNA-seq study followed by KEGG pathway enrichment found that hypoxia-responsive Chinese mitten crab hepatopancreas genes were significantly enriched for the “Pyruvate metabolism” pathway (Xu et al., 2021). Collectively, these results suggest that pdk2 may be a valuable biomarker for assessing the impact of seismic and other environmental stressors on crab mitochondrial metabolism.