A genome‐wide association study of the occurrence of genetic variations in Edwardsiella piscicida, Vibrio harveyi, and Streptococcus parauberis under stressed environments

Abstract

Bacterial mutation and genetic diversity in aquaculture have led to increasing phenotypic variances, which can weaken or invalidate strategies for controlling diseases. However, few studies have monitored the degree of mutation in fish bacterial pathogens caused by environmental pressure within a short period. In this study, transcriptomic sequences from Edwardsiella piscicida, Vibrio harveyi and Streptococcus parauberis under stressed environments were used for investigating the emergence of variants. In detail, a sub‐inhibitory concentration of formalin and phenol for E. piscicida, sea water at 30°C for V. harveyi and flounder serum for S. parauberis were used as stressed environments, and significant single‐nucleotide polymorphisms (SNPs) and/or mutation sites were investigated after culture in the ordinary liquid media (control) and the stressed environment through a genome‐wide association study. As results, several SNPs or mutations during incubation were observed under different environments in E. piscicida and/or V. harveyi in the genes relevant to flagella, fimbria type 3 secretion systems, and outer and inner membranes that have been directly exposed to external environments. In particular, given that flagella and fimbriae are considered important factors in differentiating the serotypes in some bacterial pathogens, it can be speculated that different environmental pressures are the source of phenotypic or serotypic differentiation from the same origin. On the other hands, S. parauberis did not exhibit notable changes for 4 h when inoculated in the serum from olive flounder. The results presented in this study provide examples of possible molecular evolution in pathogens relevant to the aquaculture industry as a response to different environmental pressure.

1. INTRODUCTION

Bacterial diseases are the major threats to farmed fish in aquaculture, and multiple studies have been conducted to develop methods for controlling them (e.g., Roh et al., ²⁰¹⁶; Seo et al., ²⁰²¹; Zhang et al., ²⁰²¹). However, the damage caused by bacterial diseases in aquaculture has continued, and the ability to alter their genetic makeup much faster than eukaryotes is regarded as a major obstacle (Figueroa et al., 2019; Kim et al., ²⁰²¹; Roh et al., ²⁰¹⁹). For example, the emergence and spread of multi‐drug‐resistant bacteria in aquaculture where bacteria are required to survive against antibiotics can be originated from the pressure to evolve by modifying genes that help them adapt to the newly encountered situation (Algammal et al., 2022; Nadella et al., ²⁰²²). In addition, because most bacterial pathogens in aquaculture can be horizontally transmitted through water where many hydrophilic pollutants and chemicals are easily dissolved, different water environments can accelerate the occurrence of bacterial mutations. It is therefore very important to understand the process or patterns of genetic changes among genomes differentiated from the same strain. In recent years, the importance of interpretation and analysis of massive data has received an enormous amount of attention as technological advances are made in the field of sequencing technology, and studies investigating the genotypic characteristics between strains at the genome level have been conducted (e.g., Le et al., ²⁰²¹; Roh et al., ^{2019, 2020}; Roh & Kim, ²⁰²¹). Because big data from next‐generation sequencing (NGS) may contain valuable multidimensional results, these can be interpreted from multiple perspectives. However, in numerous studies, massive sequencing results have mainly been interpreted in a manner consistent with their purposes, which means some results might be important but are overlooked or underestimated (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). Hence, making the most of released sequences in an open database such as a sequence read archive (SRA) and Genbank is becoming more important, given the release of a tremendous amount of sequencing results every day.

Among the methods of bacterial genomic study, the pan‐genome method that profiles genomic differences by the existence of certain genes has been widely used and contributes to understanding the relationship between phenotypic and genotypic characteristics of bacterial pathogens to a certain extent (Kim et al., 2020; Roh & Kim, ²⁰²¹; Rouli et al., ²⁰¹⁵). However, because pan‐genome analysis is based on the existence of genes classified as core, accessory, and singleton genes, rather than the sequence differences which are annotated as the same gene (Blom et al., 2016; Zekic et al., ²⁰¹⁸), it is not suitable for analysing the occurrence of different single‐nucleotide polymorphisms (SNPs) or mutations by strain. Even though emerging mutations and/or SNPs in certain genes or locations might be an earlier response than the differences in accessory genes, studies on SNPs and deletion and/or mutations caused by different environmental niches have been underestimated. To overcome these shortcomings, genome‐wide association studies (GWASs) offer a valuable alternative. GWAS is a method for excavating significant SNPs associated with phenotypic differences by profiling all genetic variants between phenotypic groups, not at the level of genes but at each nucleotide (Lees & Bentley, 2016). The advent of GWAS in bacteria has provided insight and understanding into SNPs associated with phenotypic characteristics (Chen & Shapiro, 2015). To estimate the occurrence of SNPs or mutated sequences when the same strain is cultured under different environments through GWAS, simultaneous deep sequencing of raw files from a large number of bacterial genomes is necessary (Wang et al., 2015). Thus far, few studies have examined genomic DNA sequences from the same strain cultured in different environments, although some studies have investigated transcriptomic responses for major bacterial pathogens in aquaculture (Vibrio harveyi, Edwardsiella piscicida and Streptococcus parauberis) (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). However, the information on nucleic acid sequences in transcriptomics that may contain promising pieces of evidence for the emergence of mutated sequences can be valuable themselves. V. harveyi, E. piscicida and S. parauberis are well‐known bacterial pathogens in aquaculture, and both raw sequencing data (FastQ file) and experimental designs have been published (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). In addition, their transcriptome responses have been known to change remarkably under sea water when exposed to high incubation temperature, pollutants or host serum environments, compared with cultures in liquid media at an optimal growth temperature (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). These transcriptomic changes are evidence that bacterial pathogens are strongly influenced by environmental factors. Based on these studies, the purposes of this study were to (1) analyse RNA‐seq raw files from several fish bacterial pathogens in various environments with a focus on mutations and SNPs, (2) identify genes that exhibit significant SNPs and mutations by different groups and (3) speculate on the microbiological meaning of these changes.

2. MATERIALS AND METHODS

2.1. Samples and FastQ files information

In total, 41 FastQ files for transcriptomic studies on E. piscicida, V. harveyi and S. parauberis have been downloaded from a SRA using the SRA‐tool kit (Ver. 2.11.3‐ubuntu64). The SRR accessions used in this study are summarized in Table 1 (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). To observe the emergence of genetic variants stimulated by sub‐lethal chemicals in E. piscicida, a study on E. piscicida CK108 transcriptomics that compared a control cultured in ordinary liquid media (brain and heart infusion) and treatments cultured in a media supplemented with either 0.0039% formalin or 0.146% phenol were used in this study (Yoon et al., 2020). In the case of V. harveyi, this study focussed on what kind of genetic variances occurred over time in the sea water with high water temperature compared with V. harveyi grown in sufficient nutrients and at the optimal temperature. In detail, the transcriptomic results of ATCC®14126™ grown in the marine broth at 26°C overnight were used as the control, while groups of V. harveyi transferred in sea water (SW) at 30°C for 12 h, 3 and 6 days were regarded as treatment groups for GWAS (Montánchez et al., 2019). Likewise, the possibilities of genetic variances in S. parauberis under liquid media and serum environment with the elapsed time were investigated in this study. In more detail, transcriptomic results of S. parauberis SPOF3K cultured in liquid media (brain and heart infusion supplemented with 1% NaCl) and serum from olive flounder (Paralichthys olivaceus; ~200 g) for 1, 2 and 4 h were utilized (Lee et al., 2021). The summaries of the experiment designs and comparison groups for GWAS are presented in Table 1.

TABLE 1.

Bacteria species, comparison groups for genome‐wide association studies, experimental conditions, SRR accession numbers, and references used in this study

Species	GWAS comparison	Experimental conditions	SRR accession	Reference
Edwardsiella piscicida	Control vs. formalin group Control vs. phenol group	E. piscicida CK108 was incubated in liquid media (brain heart infusion) for 24 h at 27°C with 200 rpm shaking as the control (n = 3). The formalin (n = 4) and phenol treated groups (n = 4) were cultured under the same conditions but supplemented by formalin (final concentration = 0.0039%; 0.38 mM) and phenol (final concentration = 0.146%; 1. 3 mM)	SRR11652662‐SRR11652676	Yoon et al. (2020)
Vibrio harveyi	Control vs. 12 h seawater incubation at 30°C (12 h) Control vs. 3 days seawater incubation at 30°C (3 days) Control vs. 6 days seawater incubation at 30°C (6 days)	V. harveyi ATCC® 14126™ were aerobically cultured overnight in liquid media (marine broth) at 26°C as the control group (n = 3). The same V. harveyi strain which had been cultured for 12–16 h incubation in the same culturable environment as the control was diluted by sterile seawater (1:20; 20 folds dilution) and incubated at 30°C for 12 h, 3, and 6 days	SRR7058340‐SRR7058351	Montánchez et al. (2019)
Streptococcus parauberis	1 h broth vs. 1 h serum 2 h broth vs. 2 h serum 4 h broth vs. 4 h serum	S. parauberis SPOF3K grown in liquid media (brain and heart infusion with 1% NaCl) at 26°C for 18 h with 180 rpm shaking was incubated in fresh liquid media or serum achieved from Olive flounder (~200 g). The tri‐replicated cultures in both broth (control) and serum (treatment) were sampled at 1, 2, and 4 h post‐incubation	SRR13349544‐SRR13349561	Lee et al. (2021)

2.2. Sequence analysis

The adapter and raw quality of reads in all FastQ files used in this study were trimmed using Trim Galore! (Ver. 0.6.7+ galaxy0) under the conditions of a Phred quality score threshold = 20, maximum allowed error rate = 0.1, and discarding reads that became shorter than 20 bp (Krueger, 2021). The genome sequences of E. piscicida CK41 (Genbank accession: CP047671.1), V. harveyi ATCC 14126 (Genbank accession: NZ_BCUF01000001.1) and S. parauberis SPOF3K (CP025420.1) downloaded from GenBank were indexed, and the trimmed FastQ files were mapped onto each indexed genome using Bowtie2 (Ver. 2.4.1) (Langmead & Salzberg, 2012). The SAM or BAM file was converted to a sorted BAM file using the sort option of samtools (Ver. 1.9) (Li et al., 2009). Based on the comparison groups of GWAS in E. piscicida, V. harveyi and S. parauberis described in Table 1, multiple sorted BAM files were combined to generate a bcf file in the conditions of ‐Ob, ‐m and‐‐ploidy 1 using the mpileup option of bcftools (Danecek et al., 2016). Likewise, all sorted BAM files for each species were merged, and its bcf file was used to observe correlation missing genotypes based on the identity‐by‐missingness (IBM) between samples. All bcf files were converted to vcf files for further analyses using bcftools, and variant sequence frequency (VSF) was calculated for minor variant sequences from each comparison of vcf file.

2.3. SNP calling, genome‐wide association analysis and bioinformatics

All genome‐wide association analyses used in this study were conducted according to the guidelines of second‐generation PLINK (Ver. 1.9) (Chang et al., 2015). In brief, SNPs and/or mutations were identified with a threshold of 0.01 minor allele frequency (MAF), and ped and map files were generated using vcftools (Danecek et al., 2011). The label information and ped and/or map files were then merged, and bed, fam and bim files were used to locate missing SNPs and SNPs that exhibited significant differences by treatment with‐‐allow‐no‐sex mode. To analyse the correlation of missing genotypes in each bacterial species, the bcf file containing all sorted BAM files of the same species was used. Conversely, for GWAS, the merging file containing only the sorted BAM files relevant to a GWAS comparison was used. A p‐value less than .05 for each GWAS comparison was considered a significant SNP.

2.4. Data analysis, visualization and statistics

The results of the correlation of missing genotypes between samples based on IBM were employed to identify the differences among all aligned sequences in each bacterial species, and hierarchical clustering with the complete method was performed using the ape package in R (Paradis & Schliep, 2019). The location and annotation of coding DNA sequences in each reference genome were predicted using RastTk pipeline (Brettin et al., 2015). All nucleic acid sequences that exceeded 0.01 MAF and SNPs belonging to genes that possessed the first to third highest numbers of significant SNPs were visualized in a Manhattan plot using qqman and ggplot2 library in R (Ver. 4.0.5) (Turner, 2014; Wickham, ²⁰¹⁶). The gene with at least one significant SNP was mapped onto KEGG pathways using the KEGG automatic annotation server to trace the biological pathways that harboured genes with SNPs identified in each GWAS comparison (Moriya et al., 2007). For this, a single‐directional best hit mode was used, and several data sets of reference bacterial species (eic and etr; E. piscicida, the same or closest species to E. piscicida (eic and etr), V. harveyi (vch, vcj, vco, vcm, vvu, vvy, vvm, vpa, vha, vex, vsp and van), and S. parauberis (spy, spn, sag, smu, stc, ssa, ssb, sgo, sez, sub, sds, sga, smb and std) were selected as the reference databases for the KEGG annotation. The ratio for the number of genes in featured pathways where more than three genes have been harboured was visualized using the Enrichment map in Cytoscape (Ver. 3.7.2) (Merico et al., 2010; Shannon et al., ²⁰⁰³). The significant difference in the number of genes that have at least one SNP in each KEGG pathway has been identified by the chi‐square analysis using a gmodels package in R (Warnes et al., 2015).

3. RESULTS

3.1. SNPs and GWAS for E. piscicida

Although one control sample of E. piscicida (E. piscicida WT_3) was located relatively far from the other two samples, most samples were well‐clustered in accordance with the phenol and formalin treatments (Figure S1a). The number of SNPs that exceeded 0.01 MAF was 1721 and 2731 in the formalin and phenol groups, respectively, and 514 and 512 were significant SNPs (p < .05) (Figure S1b). The number of significant SNPs belonging to each gene was then counted based on the location of significant SNPs and genes in E. piscicida, and 15 genes (malS, TcfC, MrdA, ClC, nrfE, SDM, L‐Asc family, dppA, fucP, tolC, TREM2, LcrD, DgcZ, MetC and FimD; all abbreviations are shown in Table 2) were found to contain more than 3 significant SNPs in either the formalin or the phenol comparison (Table 2). Among these genes, MrdA, TcfC and fucP had 6, 4 and 4 significant SNPs, which were the first to third highest genes between the control and formalin groups (Figure 1A). For the control and phenol comparison, both genes (MrdA and LcrD) contained highest number of significant SNPs (n = 6), and fucP was the third highest (n = 5) (Figure 1A). Similar or identical numbers of SNPs were observed in 8 of 15 genes in both formalin and phenol groups; however, seven genes (L‐Asc family, dppA, fucP, tolC, DgcZ, nrfE and FimD) were primarily observed in the formalin comparison group (Figure 1B). On the other hand, a smaller number of SNPs from these 15 genes was identified in the comparison between formalin and phenol groups (Table S1). To estimate the biological meaning of the genes where at least one SNPs or mutation was observed, all genes with the significant SNP(s) were annotated using the KEGG database; the numbers of genes mapped onto each pathway are depicted in Table S2. The proportion of the number of genes in each pathway out of all genes confirmed in both formalin and phenol comparison was calculated, and the pathways where more than three genes were assigned to at least one group are visualized in Figure 2. Five of 28 KEGG pathways (metabolic, galactose metabolism, bacterial chemotaxis, flagellar assembly and two‐component systems) contained a significantly higher number of genes in phenol groups than in formalin groups (Figure 2).

TABLE 2.

Location, a reference code, A2 (major allele sequence)/A1 (minor allele sequence), chi‐square and p‐value of significant SNPs from GWAS for Edwardsiella piscicida in the formalin and phenol group, and mapping information (MI) including depth (DP) and variant sequence frequency (VSF) of each SNP

Gene (abbreviation)	SNP site	Ref	WT vs. formalin						WT vs. phenol
			GWAS				MI		GWAS				MI
			A2/A1	CHISQ	P	FDR	DP	VSF (%)	A2/A1	CHISQ	P	FDR	DP	VSF (%)
Periplasmic alpha‐amylase (malS)	370,815	T	T/C	8	0.005	0.040	13	10	T/C	8	0.005	0.086	11	11
	370,888	T	T/C	6	0.014	0.100	9	17	T/C	10	0.002	0.086	14	8
	371,452	G	G/A	8	0.005	0.040	13	8	G/A	8	0.005	0.086	10	11
TcfC E‐set like domain‐containing protein a (TcfC)	1,238,362	A	A/G	8	0.005	0.040	6	17	G/A	4	0.046	0.185	2	50
	1,239,622	G	G/A	4.8	0.028	0.103	47	6	–				14	18
	1,240,084	A	A/G	6	0.014	0.100	4	25	G/A	4	0.046	0.185	2	50
	1,240,241	A	A/G	10	0.002	0.017	22	5	G/A	4	0.046	0.185	2	50
	1,238,381	G	–				8	0	G/A	4	0.046	0.185	2	50
	1,239,178	T	–				44	3	T/C	6	0.014	0.152	8	14
Peptidoglycan D,D‐transpeptidase (MrdA)	1,394,009	A	A/C	4.8	0.028	0.103	142	2	A/C	4.8	0.028	0.185	29	8
	1,394,010	A	A/T	4.8	0.028	0.103	144	2	A/T	4.8	0.028	0.185	29	8
	1,394,012	A	A/G	4.8	0.028	0.103	145	2	A/G	4.8	0.028	0.185	29	8
	1,394,013	C	C/G	4.8	0.028	0.103	146	2	C/G	4.8	0.028	0.185	29	8
	1,394,014	C	C/A	4.8	0.028	0.103	147	2	C/A	4.8	0.028	0.185	29	9
	1,394,016	T	T/A	4.8	0.028	0.103	147	2	T/A	4.8	0.028	0.185	30	8
Chloride channel protein (ClC)	1,437,777	G	G/A	4.8	0.028	0.103	11	11	G/A	4.8	0.028	0.185	14	9
	1,437,628	C	–						C/T	4.8	0.028	0.185	27	5
	1,437,717	G	–						G/A	4.8	0.028	0.185	28	5
Cytochrome c‐type heme lyase subunit (nrfE)	1,539,560	A	A/G	10	0.002	0.017	43	4	G/A	4	0.046	0.185	5	33
	1,539,936	G	G/A	6	0.014	0.100	6	50	–				4	100
	1,540,797	T	T/C	10	0.002	0.017	20	6	T/C	6	0.014	0.152	4	25
SAM‐dependent methyltransferase (SDM)	1,601,113	A	A/G	10	0.002	0.017	24	14	A/G	8	0.005	0.086	12	30
	1,601,399	T	T/C	10	0.002	0.017	33	4	T/C	10	0.002	0.086	12	14
	1,601,414	A	A/G	10	0.002	0.017	26	5	A/G	10	0.002	0.086	12	13
PTS system, ascorbate‐specific IIC component (L‐Asc family)	2,073,595	T	T/C	4.8	0.028	0.103	90	6	–				22	33
	2,074,422	T	T/C	4.8	0.028	0.103	24	5	–				4	33
	2,074,672	C	C/T	4.8	0.028	0.103	21	10	–				7	29
	2,074,142	T	–				41	0	T/C	6	0.014	0.152	5	40
	2,074,671	T	–				21	0	T/C	6	0.014	0.152	7	14
Dipeptide ABC transporter (dppA)	2,194,891	T	T/C	4.8	0.028	0.103	30	4	T/C	4.8	0.028	0.185	20	6
	2,194,894	T	T/C	4.8	0.028	0.103	30	4	T/C	4.8	0.028	0.185	21	6
	2,195,101	C	C/T	4.8	0.028	0.103	17	13	–				8	33
Fucose permease (fucP)	2,690,807	A	A/T	10	0.002	0.017	19	7	T/A	4	0.046	0.185	4	50
	2,690,834	G	G/A	10	0.002	0.017	20	7	–				2	100
	2,691,021	T	T/C	4	0.046	0.126	3	33	–				2	0
	2,691,140	T	A/T	4	0.046	0.126	4	25	–				1	100
TolC family protein (tolC) a	2,726,025	T	T/C	10	0.002	0.017	15	13	–				2	100
	2,726,173	T	T/C	4.8	0.028	0.103	11	20	–				5	50
	2,726,409	A	A/G	4.8	0.028	0.103	28	6	–				6	25
Ig‐like domain‐containing protein (TREM2) a	2,735,523	G	G/A	8	0.005	0.040	39	3	A/G	4	0.046	0.185	2	50
	2,738,408	A	A/G	6	0.014	0.100	4	33	–				2	100
	2,738,491	T	T/C	6	0.014	0.100	5	25	–				1	100
	2,734,265	A	–				24	9	A/G	4.8	0.028	0.185	9	29
	2,736,314	A	–				15	0	A/G	6	0.014	0.152	7	17
	2,736,408	T	–				10	0	T/C	6	0.014	0.152	8	25
Type III secretion inner membrane channel protein (LcrD)	3,326,526	T	T/C	10	0.002	0.017	230	1	C/T	4	0.046	0.185	3	33
	3,327,336	A	A/G	4.8	0.028	0.103	196	2	A/G	4.8	0.028	0.185	15	17
	3,327,402	A	A/G	4.8	0.028	0.103	107	2	A/G	4.8	0.028	0.185	21	25
	3,326,852	T	–				52	0	T/C	4	0.046	0.185	2	50
	3,327,486	C	–				125	0	C/A	4	0.046	0.185	2	50
	3,328,029	T	–				82	0	T/C	4	0.046	0.185	2	50
Diguanylate cyclase (DgcZ)	3,401,083	T	T/C	4.8	0.028	0.103	106	2	T/C	4.8	0.028	0.185	23	5
	3,402,611	T	T/C	4.8	0.028	0.103	175	1	‐				29	4
	3,402,641	T	T/C	4.8	0.028	0.103	197	1	–				33	4
Cystathionine beta‐lyase (MetC)	3,518,551	T	T/G	10	0.002	0.017	7	14	T/G	8	0.005	0.086	5	25
	3,518,851	C	T/C	4	0.046	0.126	3	67	T/C	4	0.046	0.185	4	67
	3,518,926	T	T/C	4	0.046	0.126	6	33	–				4	0
	3,519,395	G	–				17	0	G/A	4	0.046	0.185	3	50
Type 1 fimbriae anchoring protein (FimD)	3,733,406	T	T/C	4.8	0.028	0.103	24	5	–				12	10
	3,734,033	T	T/C	6	0.014	0.100	5	25	–				1	100
	3,734,698	T	T/C	4	0.046	0.126	2	50	–				1	0

Note: Bold letters indicated SNPs with FDR <0.25 and DP ≥30.

^a Gene was annotated by non‐redundant (NR) protein sequence database (blastp).

FIGURE 1.

Manhattan plot of genome‐wide association analysis (GWAS) between WT and formalin ((a)‐a) or phenol group ((a)‐b). Circos plot for the genes with more than 3 significant SNPs in one group. The thickness of green and pink coloured lines denotes the number of Edwardsiella piscicida SNPs in phenol and formalin groups, the abbreviation of each gene was written in brackets (b). The brown and green dots (a) indicate the genes that contained the first to the third highest number of significant SNPs (p < .05)

FIGURE 2.

The ratio for the number of genes with at least one significant SNP between formalin and phenol groups belonging to KEGG pathways in Edwardsiella piscicida. The thickness of edges between nodes indicates similarity based on the number of shared genes. The yellow background with the asterisk in the node indicates significantly different counts based on chi‐square analysis (*p < .05; **p < .01; ***p < .001)

3.2. SNPs and GWAS for V. harveyi

In general, the clustering result of missing genotype correlations based on the IBM did not show any specific patterns among the samples, although the samples from V. harveyi incubated in the SW at 30°C for 6 d tended to be closer than other groups (Figure S2a). The numbers of SNPs higher than 0.01 MAF from the three comparisons (between Con and SW at 30°C for 12 h, 3 and 6 days) were 2955, 6678 and 2015, respectively, and 143, 574 and 143 sites in 12 h, 3 and 6 days groups, respectively, were significant sites (p < .05) (Figure S2b). The upper genes with the first to the third‐largest number of significant SNPs were identified for each sampling time point, and six genes (fliN, FlaD, RcpC, DECR, PrkC and Eis) were selected in the 12‐h group. Likewise, seven genes (COX1, RMD‐MFP, RND‐IMT, mcpB, FliN, lapD and PrkC) and six genes (FliN, RcpC, OMP, PrkC, DUF4150 and TKT) were regarded as those that had the largest number of significant SNPs in the 3 and 6 days groups, respectively (Figure 3A). Among these genes, seven (RND‐MFP, RND‐IMT, mcpB, COX1, lapD, Eis and TKT) exhibited 3 or more significant SNPs in a gene from at least one group (Table 3). In particular, five genes (RND‐MFP, RND‐IMT, mcpB, COX1 and lapD) had SNPs in only the 3 days group, and GNAT family N‐acetyltransferase and Transketolase contained SNPs in the 12 h or 6 days groups (Figure 3B). Because numerous genes were dominantly enriched into KEGG pathways in the 3 days group, compared with other groups, 20 of 28 KEGG pathways had a statistically different number of genes that harboured significant SNPs between the groups (Figure 4; Table S2). In particular, KEGG pathways such as flagella assembly, ABC transporters, two‐component system, biosynthesis of cofactors, microbial metabolism in diverse environments, metabolic pathways, biosynthesis of secondary metabolites, and biofilm formation—Vibrio cholera had a higher number of genes in the 3 days group (p < .001).

FIGURE 3.

Manhattan plot of genome‐wide association analysis (GWAS) in Vibrio harveyi with 12 h ((a)‐a), 3 days ((a)‐b) and 6 days ((a)‐c) seawater at 30°C incubation. Circos plot for the genes that have the first to third highest SNPs in one group. The thickness of the green and pink coloured lines denotes the number of V. harveyi SNPs at each time point (12 h, 3 and 6 days) (b). The coloured dots (brown, green and blue; a) indicate the genes that contained the first to the third highest number of significant SNPs in each time point (p < .05)

TABLE 3.

Location, a reference sequence, A2 (major allele sequence)/A1 (minor allele sequence), chi‐square, and p‐value of significant SNPs from GWAS for Vibrio harveyi incubated in seawater at 30°C for 12 h, 3 and 6 days, and mapping information (MI) including depth (DP) and variant sequence frequency (VSF) of each SNP

Gene (abbreviation)	SNP site contig)	Ref	Con vs. 12 h						Con vs. 3 days						Con vs. 6 days
			GWAS				MI		GWAS				MI		GWAS				MI
			A2/A1	CHISQ	P	FDR	DP	VSF (%)	A2/A1	CHISQ	P	FDR	DP	VSF (%)	A2/A1	CHISQ	P	FDR	DP	VSF (%)
RND efflux system, membrane fusion protein (RND‐MFP)	218,971 (2)	T	–				66	0	T/C	8	0.005	0.092	51	2	–				64	0
	219,094 (2)	A	–				35	3	A/T	8	0.005	0.092	32	7	–				37	3
	219,098 (2)	T	–				35	0	T/C	8	0.005	0.092	32	3	–				36	0
RND efflux system, inner membrane transporter (RND‐IMT)	51,394 (8)	C	–				25	0	C/T	8	0.005	0.092	5	20	–				15	7
	51,397 (8)	A	–				25	4	A/T	8	0.005	0.092	6	20	‐				16	0
	51,789 (8)	T	–				21	0	T/G	8	0.005	0.092	10	11	–				37	0
Methyl‐accepting chemotaxis sensor/transducer protein (mcpB)	83,161 (15)	T/A	–				31	0	A/G	8	0.005	0.092	87	20	–				135	0
	83,178 (15)	T	–				24	0	T/A	6	0.014	0.158	68	22	–				110	0
	83,798 (15)	G/A	–				37	0	A/G	8	0.005	0.092	44	6	–				91	0
putative flagellar motor switch protein (fliN)	7823 (63)	T	T/C	8	0.005	0.162	30	10	T/C	6	0.014	0.158	9	22	T/C	8	0.005	0.111	86	21& INDEL; TG (6)
	7824 (63)	G	G/T	8	0.005	0.162	30	3	G/T	6	0.014	0.158	9	14	G/T	8	0.005	0.111	85	7
	8309 (63)	A	A/G	6	0.014	0.162	12	17	G/A	4	0.046	0.158	5	50	A/G	8	0.005	0.111	18	12
Cytochrome O ubiquinol oxidase subunit I (COX1)	5731 (71)	A	–				70	2	A/G	8	0.005	0.092	42	5	–				82	1
	5967 (71)	T	–				47	2	T/C	8	0.005	0.092	31	7	–				62	13
	6420 (71)	A	–				80	5	A/C	8	0.005	0.092	47	11	–				81	8
	6731 (71)	A	–				62	0	A/G	8	0.005	0.092	32	3	–				60	2
diguanylate cyclase/phosphodiesterase (GGDEF & EAL domains) with PAS/PAC sensor(s) (lapD)	11,269 (77)	A	–				29	0	A/C	6	0.014	0.158	11	14	–				36	0
	11,718 (77)	G	–				21	0	G/A	6	0.014	0.158	5	25	–				26	0
	12,403 (77)	G	–				22	0	G/A	8	0.005	0.092	9	14	–				42	0
Serine/threonine protein kinase, regulator of stationary phase (PrkC)	21 (128)	T	A/T	4	0.046	0.162	10	20	–				2	100	T/A	6	0.014	0.144	5	60
	930 (128)	A	A/G	6	0.014	0.162	14	17	–				5	50	A/G	6	0.014	0.144	42	11
	53 (128)	C	–				15	0	C/T	8	0.005	0.092	6	17	–				20	10
	54 (128)	T	–				15	0	T/C	8	0.005	0.092	6	17	–				20	0
	57 (128)	T	–				15	0	T/G	8	0.005	0.092	6	17	–				20	0
Flagellin protein (FlaD)	325 (132)	A	G/A	6	0.014	0.162	81	59	–				52	47	–				127	56
	328 (132)	G	A/G	6	0.014	0.162	80	58	–				51	46	–				127	54
	688 (132)	G	A/G	6	0.014	0.162	59	63	A/G	4.444	0.035	0.158	35	50	–				101	51

Note: Bold letters indicated SNPs with FDR <0.25 and DP ≥30.

FIGURE 4.

The ratio for the number of genes with at least one significant SNP in each time point of sea water at 30°C incubation (12 h, 3 and 6 days) belonging to KEGG pathways in Vibrio harveyi. The thickness of edges between nodes indicates similarity based on the number of shared genes. The yellow background with the asterisk in the node indicates significantly different counts based on chi‐square analysis (*p < .05; **p < .01; ***p < .001)

3.3. SNPs and GWAS for S. parauberis

In the case of S. parauberis, the dendrogram analysis based on missing genotype correlation revealed that the culture conditions and incubation time, especially the difference between broth and serum environments, were greatly influenced by their clustering. Although one sample (broth 4 h rep3) was abnormally closer than serum samples, most could be primarily divided by the culturable environments (broth and serum). The samples from broth environments tended to cluster well by incubation time, but not in the serum environment (Figure S3a). The number of SNPs exceeding 0.01 MAF in the comparison with each time point of broth and serum groups were 54, 43 and 65 at 1, 2 and 4 h, respectively, which was much smaller for E. piscicida and V. harveyi (Figures S3b; Figure 5). Similarly, no gene with more than two significant SNPs in all sampling time points was observed (Table 4).

FIGURE 5.

Manhattan plot of genome‐wide association analysis (GWAS) in Streptococcus parauberis for 1 h (a), 2 h (b) and 4 h (c) incubation

TABLE 4.

Location, a reference sequence, A2 (major allele sequence)/A1 (minor allele sequence), chi‐square, and p‐value of significant SNPs from GWAS for Streptococcus parauberis incubated in broth and serum for 1, 2, and 4 h, and mapping information (MI) including depth (DP) and variant sequence frequence (VSF) of each SNP

Gene (abbreviation)	SNP site	Ref	1 h						2 h						4 h
			GWAS				MI		GWAS				MI		GWAS				MI
			A2/A1	CHISQ	P	FDR	DP	VSF (%)	A2/A1	CHISQ	P	FDR	DP	VSF (%)	A2/A1	CHISQ	P	FDR	DP	VSF (%)
–	11,965	T	–				4	0	A/T	4.00	0.046	0.186	2	50	–				3	0
–	34,946	T	–				13	0	–				7	0	T/C	6.00	0.014	0.172	4	25
–	37,322	T	T/A	6.00	0.014	0.11	6	17	–				1	0	–				4	0
tRNA‐Cys‐GCA (NSUN6)	116,816	A	–				58	48	G/A	8.00	0.005	0.101	23	70	–				36	50
–	135,377	T	T/C	8.00	0.005	0.11	8	13	–				1	0	–				3	0
–	138,394	A	–				3	0	–				4	0	T/A	4.00	0.046	0.172	2	50
–	603,666	A	–				15	0	A/T	4.00	0.046	0.186	2	50	–				9	0
–	677,993	T	–				13	0	–				5	0	C/T	4.00	0.046	0.172	3	33
Replication initiator protein A (DnaA)	1,099,212	TAA	TAA/TAAA	6.00	0.014	0.11	1	20	–				6	0	–				3	0
Hypothetical protein	1,111,230	C	–				5	0	C/G	6.00	0.014	0.186	4	25	–				5	0
6‐phospho‐beta glucosidase (lacG)	1,288,372	A	–				4	0	–				6	0	A/G	8.00	0.005	0.152	10	10
Hypothetical protein	1,309,621	C	–				43	0	C/T	8.00	0.005	0.101	10	10	–				15	0
Phage tail length tape‐measure protein T (yomI)	1,460,848	T	T/C	6.00	0.014	0.11	4	25	–				3	0	–				2	0
Phage terminase arge subunit (A)	1,471,143	T	–				3	0	T/C	4.00	0.046	0.186	2	50	T/C	6.00	0.014	0.172	3	33
–	1,526,293	T	A/T	4.00	0.046	0.168	2	50	–				1	0	–				2	0
–	1,526,442	A	G/A	4.00	0.046	0.168	2	50	–				2	0	–
Hypothetical protein	1,662,165	G	–				6	0	–				4	0	G/A	8.00	0.005	0.152	4	25
Hypothetical protein	1,706,945	CTTTTT	CTTTTT/CTTT	8.00	0.005	0.11	1	4	–				9	0	–				5	0
Tn5252, relaxase (Tn5252)	1,885,284	T	T/C	6.00	0.014	0.11	6	17	–				4	0	–				2	0
Hypothetical protein	2,037,162	T	T/C	6.00	0.014	0.11	3	33	–				1	0	–				1	0
PTS system, cellobiose‐specific IIC component (celB)	2,040,252	T	–				7	0	–				3	0	T/C	4.00	0.046	0.172	2	50
Phosphoribosylformylglycinamidine synthase (purL)	2,089,390	ATTT	–				4	0	–				3	0	ATT/ATTT	4.00	0.046	0.172	2	50

Note: Bold letters indicated SNPs with FDR <0.25 and DP ≥30.

4. DISCUSSION

Although GWAS for fish bacterial pathogens is relatively underrated compared with fish GWAS for the purpose of quantitative trait locus (QTL) marker identification, a few studies have applied GWAS for fish bacterial pathogens, which has contributed to identifying genotypic characteristics in the same species (Correa et al., 2015; Holborn et al., ²⁰¹⁸; Le et al., ²⁰²¹; Rasmussen et al., ²⁰¹⁶). Different phenotypic characteristics between strains in the same species can be created by a genetic pressure to change nucleic sequences and genes to adapt to specific environments, which has been widely observed in both prokaryotes and eukaryotes (e.g., Kim et al., ²⁰²¹; Kjærner‐Semb et al., ²⁰²¹). This is one of the main reasons why the first step in most GWAS is to classify the phenotypic characteristics or measurements (which can be discrete units [e.g., high, intermediate and low virulence] and/or continuous values) (Chen & Shapiro, 2015). Given that bacterial genomes tend to have a much stronger linkage disequilibrium than eukaryotes, bacteria are more flexible in relation to altering their genetic information under different environments, and this can accompany mutation and sequence alternation within short time periods when they are in different culturable environments (Chen & Shapiro, 2015). It is not surprising that bacteria mutation rate can be influenced by death, population dynamics and stressful environments, and generally speaking 1–100 nucleotides in 10 million to 1 billion bases can be substituted per generation depending on environments and conditions (Chevallereau et al., 2019; Frenoy & Bonhoeffer, ²⁰¹⁸; Westra et al., ²⁰¹⁷). Given that bacterial doubling time is approximately 20 min in favourable environments, although the time span for bacterial replication differs owing to multiple factors (e.g., species, temperature and nutrients) (Jaruszewicz‐Błońska & Lipniacki, 2017), more than one‐day incubation in different conditions can be enough time to make a mutation in different sites in the genome and change the major nucleic acid sequences in their population group. In addition, the fact that relatively more frequent mutated, missed or substituted genes for a certain gene compared with others indirectly suggests the importance of the gene for adapting to the surrounding environment. The advance in transcriptomic sequencing technologies can provide valuable information for profiling the sequence dynamics from the survived bacteria even in the same cultures; however, most studies have focussed on mRNA expression in different environments using transcriptomic results (Le et al., 2021; Lee et al., ²⁰²¹; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰). Hence, this study attempted to observe SNPs and/or mutations in the same strain with different environments and exposure times through raw sequencing files that had been used for transcriptomic studies (Lee et al., 2021; Montánchez et al., ²⁰¹⁹; Yoon et al., ²⁰²⁰).

In general, GWAS have been performed using genomic information rather than the sequencing for transcriptomes. However, sequences from RNA‐seq can provide an alternative source of host genetic information. For example, Berthouly‐Salazar et al. (2016) have used transcriptomes easier to obtain from a non‐model plant for investigating SNPs and genotyping as an alternative strategy, and Chandhini and Rejish Kumar (2019) also found that transcriptomes obtained from NGSs, such as genomic results, can serve as resources to excavate SNPs and other molecular makers. Moreover, as transcripts have been produced from only living bacteria, the result of transcriptomes has the advantage that it mainly comes from survived bacteria adapted to the changed environment regardless of the elapsed incubation time. On the other hand, one drawback of utilizing raw RNA‐seq files for GWAS is that the sequencing depth can fluctuate because of the differences in bacterial mRNA expression between groups, which can cause different depth levels and negatively impact the results. Nevertheless, strategies for deep sequencing of bacterial cultures can identify the occurrence of mutations even in a minority of the major population, and this provides potential insight into bacterial evolution and adaptation with the respect to genomes. For example, the VSF and DP scores in Tables 2, 3, 4 indicate the diversity of sequence and depth rate at each position. The higher VSF with deeper DP can be interpreted as the more reliable SNPs or mutated position because it meant the more different nucleotides and sequences were mapped onto a reference genome (Danecek et al., 2011). Also, as a small number of samples used in this study can lead to the risk of false‐negative (type 2 error) if Bonferroni correction would be applied, the p‐value less than .05 without Bonferroni correction, which thresholds also used in some studies, has been used in this study (Asif et al., 2021; Kap et al., ²⁰¹⁶; Lu et al., ²⁰¹⁵; Zeng et al., ²⁰¹⁵). Instead, the FDR value has been described in Tables 2, 3, 4 for evaluating the reliability of each SNP. Hence, if we keep this in mind for the interpretation of results, RNA‐seq can demonstrate the overall genomic changes under different environments in pathogenic bacteria in aquaculture.

When E. piscicida cultured in the liquid media (broth) and a minimal sub‐inhibitory concentration of formalin and phenol were compared (Yoon et al., 2020), TcfC and MrdA were the top three genes that harboured multiple SNPs and/or mutated sequences (Figure S4). In general, Tcf (typhi colonization factor) has been known to exhibit chaperone‐usher fimbriae and is classified in α fimbrial clade, and its regions are composed of six operons including chaperone (TcfA), major fimbriae subunit (TcfB), usher (TcfC) and adhesion (TcfD) (Azriel et al., 2017; Leclerc et al., ²⁰¹⁶; Yue et al., ²⁰¹²). Given that Tcf has been reported as being part of the type six secretion systems and atypical fimbriae in some species, particularly Salmonella, it can be an important virulence factor and supporting bacterial colonization in E. piscicida (Azriel et al., 2017; Folkesson et al., ¹⁹⁹⁹). Former studies (e.g., Azriel et al., ²⁰¹⁷; Leclerc et al., ²⁰¹⁶) have reported that Tcf is the factor involved in host specificity, such as serotypes in Salmonella enterica. Switching the gene sequences involved in serotypes is noteworthy because the mutation in a Tcf region of E. piscicida suggested one strain can be differentiated into several serotypes under different culturable environments. Similarly, type 3 secretion system (T3SS) is one of the well‐known virulence factors that contribute to invading and surviving in host cells through flagellar‐like needles with injecting effector proteins, and T3SS in E. piscicida has been found to be significantly up‐regulated under the sub‐inhibitory concentration of formalin (0.38 mM) (Hou et al., 2017; Yoon et al., ²⁰²⁰). Notably, the group that had mutated or variant nucleic acid sequences in LcrD was broth cultured E. piscicida rather than formalin and phenol groups. Specifically, there was little change in the nucleotide sequence in an environment that stimulate the expression of T3SS, but a higher percentage of sequence changes was observed in broth media that does not require the high expression of T3SS. Given that numerous studies (e.g., Almaguer‐Chávez et al., ²⁰¹¹; Barrick et al., ²⁰⁰⁹) have observed the decreasing virulence of bacteria when pathogenic bacteria are sub‐cultured continuously, the sequence change of T3SS might affect the virulence of E. piscicida. Barrick et al. (2009) noted that Escherichia coli has continuously evolved through long‐term subcultures with deletions, mutations, inversions and duplications in its genome that resulted in the loss of 1.2% of the original chromosome. This implies that the observation of mutations and/or SNPs in T3SS that has been known to the important virulence factor and factors exposed to the external environment in E. piscicida from the broth group can be the process of evolution to be more adaptable in the liquid media. Further study about the influence of T3SS mutation in a nutrient‐enriched environment is necessary from the perspective of bacterial pathogenicity and environmental adaptation. In this context, the plethora of mutations in MrdA can be interpreted similarly to T3SS. Peptidoglycan is one of the major components of bacterial cell walls, and D‐,D‐transpeptidase is well‐known for its strong involvement in crossing‐linking peptidoglycan strands (Hugonnet et al., 2016; Triboulet et al., ²⁰¹³). The peptidoglycan layer can help to maintain bacteria structures in the face of internal and outer pressures, and physiological impacts can exert on peptidoglycan cross‐linking (Pidgeon et al., 2019). In particular, because D‐,D‐transpeptidase as penicillin‐binding proteins can be easily inhibited by β‐lactam antibiotics, several studies have reported that L‐,D‐transpeptidase, which is highly resistant to β‐lactams, can be replaced when bacteria need to adapt β‐lactam antibacterial agents (e.g., Pidgeon et al., ²⁰¹⁹; Triboulet et al., ²⁰¹³). Based on all these studies and the results of this study, it is speculated that D‐,D‐transpeptidase in E. piscicida is an important gene that exhibits multiple alternations and/or changes depending on environmental changes and differences. In addition, the number of genes with significant SNPs or mutations belonging to certain KEGG pathways (e.g., bacterial chemotaxis, flagellar assembly and galactose metabolisms) differed between the sub‐inhibitory concentration of formalin and phenol groups. All these results support the claim that the E. piscicida genome can be flexibly changed by the external environment and mutations or SNPs do not occur with the same probability among all genes but instead are concentrated on specific genes.

The comparison of mutation or SNP sites in V. harveyi, although under the same environment (sea water at 30°C), indicated the different incubation times significantly affected the patterns of mutation. The number of SNPs that exceeded 0.01 MAF identified in the 3 days group was 6678, whereas 12 h and 6 days groups had less than half that number. This difference would have resulted in the 3 days group in most KEGG pathways having genes with at least one significant SNP or mutation site compared to others. The genes involved in flagella and flagellin components exhibited several mutations during the samplings, and these patterns were more notable as time elapsed (Figure S5). Namely, the sequence of the control group did not necessarily mean that the sequence of mutations or SNPs in the regions where mutations frequently occurred was not observed. For instance, although putative flagellar motor switch protein and flagellin protein in the control group of V. harveyi also have contained a piece of minor variants, the number of mutated sequences is much higher enough to switch the major sequence after incubation in high‐temperature sea water (Figure S5). In addition, the variances of sequences were not identical in each group, taking into account that several ‘T’ were identified in the 325 position (contig 63) from only one sample belonging to the 6 days group. The mutation and differences of flagellin and flagella relevant genes may be the reason for the alternation of phenotypic characteristics. The flagellar motor switch proteins are known to promote motility and biosynthesis in flagella, and their inactivation can cause decreasing velocity (swimming ability) in highly viscous media (1% methylcellulose) and infectivity in Borrelia burgdorferi (Li et al., 2010). In addition, Kim et al. (2014) reported that Vibrio vulnificus which has a few nucleotide deletions of flagellin(s) affected the significant reduction in cytotoxicity, motility and adhesion. All these results suggest that although there may be genes and regions where mutations in bacteria occur frequently and randomly, environmental differences can create greater diversity and phenotypic characteristics. In the case of a resistance nodulation division (RND) where the system is responsible for transporting dyes, antibiotics, host molecules and detergents located between the inner and outer membrane in gram‐negative bacteria, some mutations were identified in the 3 days group (Blair and Piddock, 2009). However, given that these mutations were not consistent until the 6 days group, some sequences in certain genes exhibited greater variation depending on sampling time points, although not all mutated sequences are necessarily meant to become major populations continuously.

Regarding the comparison of S. parauberis, the correlation of missing genotypes between samples was well‐clustered in broth and serum groups. However, no gene contained several significant mutations or SNPs, as observed in E. piscicida and V. harveyi. Because S. parauberis cultured in broth and serum for 4 h did not exhibit less than two times the high number of bacteria compared to 0 h (Lee et al., 2021), it is suggested that the incubation times were not sufficient to initiate significant SNPs or mutation. However, missing genotypes might be affected by gene expression because RNA‐seq, files not genome sequencing results, were used in this study. According to the results of principal investigation analysis based on transcriptomic results, despite the big difference in transcriptomic responses as incubation time elapsed, closer clustering distance by incubation time rather than a different environment suggested that environmental pressure can be a much stronger factor in initiating the missing sequence or deletions.

In conclusion, this study first investigated SNPs or occurring mutations in major fish bacterial pathogens from the same strain, and successfully identified genes (especially, flagella, fimbriae and flagellin relevant genes) that have a much higher number of mutated sequences than other genes under different environments. Given that multiple studies have already demonstrated fimbriae that are directly exposed to the external environment are the important gene for differentiating bacterial serotypes, these genes can be more flexible in changing their structures which can be linked to alternation in their phenotypic characteristics. In addition, other genes that contained several significant mutations and SNPs based on GWAS can be potential biomarkers for tracking the origin and sources of the same bacterial species. Although further studies are necessary to investigate the meaning of mutations in each gene and species, the data presented in this study provide valuable information and insight into the evolution of bacterial pathogens in aquaculture.

CONFLICT OF INTEREST

The author declares no competing interest exists.

Supporting information

Figure S1 Dendrogram for the correlation of missing genotypes in each E. piscicida based on identity‐by‐missingness (IBM) (a). Venn diagram for the number of SNPs with more than 0.01 minor allele frequency (MAF) in the comparison between wild type (WT) and treatments (formalin and phenol) (b).

Figure S2. Dendrogram for the correlation of missing genotypes in each group of V. harveyi based on identity‐by‐missingness (IBM) (a). Venn diagram for the number of SNPs with more than 0.01 minor allele frequency (MAF) in the comparison between control and sea water at 30°C incubation for 12 h, 3, and 6 days (b).

Figure S3. Dendrogram for the correlation of missing genotypes in each group of S. parauberis based on identity‐by‐missingness (IBM) (a). Venn diagram for the number of SNPs with more than 0.01 minor allele frequency (MAF) in the comparison between each time point of broth (control) and serum (treatment) incubation for 1, 2 and 4 h (b).

Figure S4. The reference and portion of mapped sequences where significant SNPs were identified through GWAS from E. piscicida cultured in control (Wild‐type; broth), formalin, and phenol.

Figure S5. The reference and portion of mapped sequences where significant SNPs were identified through GWAS from V. harveyi incubated in control (broth at 26°C) and sea water at 30°C for 12 h, 3 and 6 days, respectively.

Table S1 Location, a reference code, A2 (major allele sequence)/A1 (minor allele sequence), chi‐square, and p‐value of significant SNPs from GWAS for E. piscicida between formalin and phenol group, and mapping information (MI) including depth (DP) and variant sequence frequency (VSF) of each SNP.

Table S2 The number of gene(s) that have at least one significant SNP(s) in each KEGG pathway in E. piscicida or V. harveyi group comparison

Roh, H. (2022). A genome‐wide association study of the occurrence of genetic variations in Edwardsiella piscicida, Vibrio harveyi, and Streptococcus parauberis under stressed environments. Journal of Fish Diseases, 45, 1373–1388. 10.1111/jfd.13668

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.