Measuring the change in deleterious mutation burden of barley germplasm conserved over one regeneration cycle Authors: Yong-Bi Fu Published: 2026 ABSTRACT AbstractDeleterious mutations will occur and accumulate in germplasm conserved under long-term cold storage. However, little is known about the extent of deleterious mutation accumulation in conserved germplasm. This study represents an attempt to measure the change in deleterious mutation burden of barley germplasm conserved over 1 regeneration cycle. Specifically, seed-based RNA-Seq analysis was performed on paired samples of 90 barley accessions representing germplasm conserved under cold storage since its original acquisition (G0) and since the first regeneration (G1). The analysis identified 1,642 deleterious variants by 2 SNP calling methods. These deleterious variants were widely distributed across 7 chromosomes, were present in only 8 or fewer samples, occurred predominantly in heterozygotes, were associated with diverse biological processes, and were largely predicted to be weakly detrimental. Based on BCFtools-based SNP calls, the change in total mutation burden per deleterious locus per conservation year was estimated at 0.00123, with a standard deviation of 0.00084, for an accession conserved through the first regeneration. This estimate suggests that a conserved barley accession accumulated 28 to 37 deleterious variants over the G0 to G1 period. More deleterious variants were found in G1 than in G0 samples; variants unique to each sample type were also observed; and a high proportion of variants occurred in heterozygotes. These findings clearly demonstrated the technical and practical feasibility of measuring mutation accumulation in barley germplasm conserved in a genebank. Introduction Deleterious mutations will occur in more than 7 million plant germplasm accessions conserved in 2,000 genebanks worldwide (FAO 2010; Engels and Ebert 2024), particularly in those conserved under long-term cold storage, as a result of changes in DNA that disrupt normal gene function (Ashton 1956; Roberts 1978; Dourado and Roberts 1984). These mutations can accumulate in conserved germplasm from the joint actions of selection and genetic drift, both before and during germplasm regeneration (Schoen et al. 1998). Mutation accumulation will alter the original genetic background of conserved germplasm (Charlesworth et al. 1993), potentially increasing its vulnerability to future environmental pressures by reducing reproductive success and survival (Hoffman et al. 2004; Lu et al. 2006; Roles and Conner 2008; Moyers et al. 2018; Kono et al. 2019; Dwivedi et al. 2023). This accumulative genetic change could compromise some objectives of the long-term germplasm conservation mission of minimizing genetic shift and reducing loss of the original genotypes (Schoen and Brown 2001; Fu 2024a). Therefore, there is a clear need to develop effective genebank management procedures to minimize the genetic changes from deleterious mutation accumulation (Fu 2024a). However, the knowledge of deleterious mutation accumulation in conserved germplasm is lacking (Fu et al. 2023; Fu and Horbach 2025). Assessing mutation burden (or the extent of deleterious mutations) was traditionally quite challenging (Mukai 1964; Charlesworth et al. 1990; Kondrashov and Kondrashov 2010) but has become both technically possible and practically feasible. This progress reflects advances in genomics, particularly through genetic load studies in the human genome (eg Cooper et al. 2005; Henn et al. 2015), as well as the development of bioinformatics tools for identifying deleterious genetic variants (eg Ng and Henikoff 2003). Technically, genome-wide genetic variants are screened across a sequenced genome and then identified to be deleterious through gene function prediction of a nonsynonymous site change alone (eg Ng and Henikoff 2003) and/or in combination with the intensity of purifying selection inferred from phylogenetic restraints on the site (Davydov et al. 2010). Over the last decade, numerous studies have been conducted to identify genome-wide deleterious genetic variants in many plant genomes (Mezmouk and Ross-Ibarra 2014; Renaut and Rieseberg 2015; Kono et al. 2016; Liu et al. 2017; Ramu et al. 2017; Valluru et al. 2019; Lozano et al. 2021; Fu et al. 2023; Sun et al. 2023; Wu et al. 2023; Razifard et al. 2025). These predicted deleterious mutations are more likely to be harmful (Kono et al. 2019; Plekhanova et al. 2019; Sun et al. 2023), although their fitness effects are largely unknown (Roles and Conner 2008; Bertorelle et al. 2022). Overall, these studies have successfully demonstrated the effectiveness of identifying deleterious genetic variants across plant genomes and have provided informative estimates of mutation burden in plant germplasm (eg Ramu et al. 2017; Kistler et al. 2018; Sun et al. 2023; Wu et al. 2023). In genebanks, plant germplasm is typically conserved in the form of seeds and managed following Food and Agriculture Organization (FAO) Genebank Standards (FAO 2014, 2022). When seed of an accession is acquired for conservation, it must go through various genebank procedures, including seed cleaning and increase, before being packaged in a sealed envelope at an assumed 20% relative humidity and stored at −18 °C to −20 °C for long-term conservation. Seed viability of conserved germplasm is monitored over time (Hay and Whitehouse 2017). When the seed viability is below the threshold value (commonly 85%), part of the seed sample for the accession will be regenerated in the greenhouse or field. The newly regenerated seed will be packaged as a new sample for the originally acquired accession for long-term cold storage. Regeneration and storage procedures are repeated over time to maintain germplasm viability and availability for future use. This process provides a unique opportunity to assess the changes in the mutation burden of conserved germplasm over regeneration cycles, as seed samples from multiple regenerations of the same accession commonly exist in genebanks. To enhance the long-term management of plant germplasm conserved in worldwide genebanks, an attempt was made to measure the change in deleterious mutation burden of barley (Hordeum vulgare L.) germplasm conserved over 1 regeneration cycle using seed-based RNA-Seq analysis (Wang et al. 2009). Specifically, 90 barley accessions with paired samples (1 conserved since the original acquisition and 1 since the first regeneration cycle) were selected from the barley collection conserved at Plant Gene Resources of Canada (PGRC), Saskatoon, Canada. RNA-Seq analysis was applied to 180 barley samples, and genome-wide deleterious variants were identified based on phylogenetic restraints. Total mutation burden per deleterious locus was estimated for each assayed sample. The pairwise sample differences in total mutation burden per accession were calculated using a novel method that adjusts for variable years of conservation to estimate the change occurring over 1 germplasm regeneration. Overall, this seed-based RNA-Seq analysis was undertaken with the hope to assess the technical feasibility of measuring the extent of deleterious mutation accumulation in plant germplasm conserved in a genebank. Materials and methods Sample selection and acquisition We selected 90 barley accessions (Table 1) from the PGRC barley collection based primarily on 2 factors: the number of regeneration cycles and the availability of seed for distribution. These accessions were originally acquired by PGRC from 8 countries between 1973 and 1989, conserved in sealed envelopes at an assumed 20% relative humidity under PGRC long-term storage at −18 °C and subjected to only one regeneration cycle between 1990 and 2016, following FAO Genebank Standards (FAO 2014). Specifically, when an original accession was identified with seed viability of 85% or lower, approximately 200 seeds were randomly sampled from the accession and regenerated at the experimental farm of the Saskatoon Research and Development Centre with a field layout of 1 single 3-m row plot per accession with 60 cm between rows, along with strict control of weeds, pests, and diseases. Seeds were harvested from all plants in the row for each accession. Planting 90 to 210 seeds is expected to retain alleles with frequencies between 0.003 and 0.05 across 10 to 150 loci with a probability of 90% to 95% (Crossa et al. 1993). This regeneration procedure was repeated over the years because the selected accessions were not regenerated in the same year. Table 1.List of 90 barley accessions assayed in 2021 with their paired samples representing the conserved germplasm since its original acquisition (G0; from 1973 to 1989) and since the first regeneration cycle (G1; from 1990 to 2016), along with their countries of origin and accession descriptions.AccessionOriginDescriptionYear0Year1AccessionOriginDescriptionYear0Year1CN9448CanadaBeacon19801994CN31558United StatesBetzes19781995CN9450CanadaBetzes19801993CN31673United StatesAtlas 4619781994CN9456CanadaB.T.32219801993CN32160GermanyNackta19781995CN9479CanadaOlli19801993CN32585RussiaWiner19781995CN9487CanadaVantmore19801993CN33302CanadaU48/5419731995CN12930CanadaC18-7319832009CN33306CanadaFjola19731995CN13176CanadaC269-2219831990CN35778AzerbaijanPallidum 59619792003CN13180CanadaC273-319831990CN35779RussiaToomas19791994CN13203CanadaD12-319831990CN35780RussiaZavet19791994CN13224CanadaD29-219831990CN39286CanadaBichita19821996CN13372CanadaF93-819831990CN39309CanadaQB179.9719821994CN14132CanadaS55-G9019832016CN39314CanadaQB722.719821994CN14393CanadaS99-G18419831999CN39339GermanyVolla19821994CN14494CanadaS112-G36419831999CN39379CanadaBT91119821994CN14508CanadaS114-G2019831999CN39408CanadaM7619-433719821994CN14527CanadaS114-G12219831999CN39410CanadaSB8021119821994CN14648CanadaS144-419831999CN39413CanadaSF7920119821995CN14653CanadaS148-4819831999CN40276RussiaPGR 1330219831994CN14690CanadaS193-119832011CN40278RussiaWIR 451419831994CN14913CanadaS309-G7619832011CN40300United StatesPGR 1333419831994CN15040CanadaS379-1719832014CN42483DenmarkBinder19841994CN15059CanadaS382-719832011CN42525United StatesComposite Cross19841994CN15073CanadaS383-1519832014CN42855CanadaSM8206119841994CN15140CanadaS535-5919832011CN42877United StatesAdvance19841994CN16296CanadaW248-7819832013CN42934CanadaJackson19851994CN16530CanadaB17-619842013CN43235CanadaSM8309119851994CN16547CanadaB25-4119842013CN43288CanadaTBM80-1919851994CN16562CanadaC21-2119842013CN43556AustraliaIlineckii 4319861994CN16563CanadaC45-119842013CN43634Canada82-RCBB-1319861994CN16700CanadaC398-1319842016CN43666CanadaOB752-2119861994CN16873CanadaD259-519842016CN43731CanadaSM8427519861994CN17537CanadaS709-5119842014CN43743CanadaH6907219319861994CN17557CanadaS779-4119842014CN43747United States2B81-403819861994CN17580CanadaS783-2219842014CN45881CanadaTR 11519881994CN17600CanadaS791-319842014CN45883CanadaC.D.C. Guardian19881994CN17616CanadaS899-119842014CN45886CanadaTR 22719881994CN17628CanadaS925-4519842014CN45889CanadaTR 31119881994CN17630CanadaS933-919842014CN45892Canada92-4119881994CN17638CanadaS973-519842014CN46226CanadaClark19891994CN17820CanadaL06/L64-119851994CN46747CanadaOB959-719891994CN17822CanadaBT 36419851994CN46748CanadaOB956-1319891994CN31045CanadaPGR 205719761994CN46750CanadaOB758-1719891994CN31054CanadaSaxonia19761994CN46753CanadaOB762-1319891994CN31058Canada12502 C/219761994CN51811CanadaBrier19891994CN31140FranceSandra19771994CN52167FrancePatty19891994Year0 and Year1 show the years of the original accession acquisition and the first completed regeneration cycle. For each accession, Y0 = 2021−Year0; Y1 = 2021−Year1; and YR = Y0−Y1. List of 90 barley accessions assayed in 2021 with their paired samples representing the conserved germplasm since its original acquisition (G0; from 1973 to 1989) and since the first regeneration cycle (G1; from 1990 to 2016), along with their countries of origin and accession descriptions. Year0 and Year1 show the years of the original accession acquisition and the first completed regeneration cycle. For each accession, Y0 = 2021−Year0; Y1 = 2021−Year1; and YR = Y0−Y1. Seed for these 90 accessions, with paired samples representing germplasm conserved from the time of original acquisition (G0) and from the first regeneration cycle (G1), along with their inventory data (Table 1), such as passport information, country of origin, year of acquisition (Year0), and year of the first regeneration (Year1), was acquired in October 2021 from Mr Brian James at PGRC, following the Standard Material Transfer Agreement of the International Treaty on Plant Genetic Resources for Food and Agriculture (https://www.fao.org/plant-treaty/areas-of-work/the-multilateral-system/smta/en/; accessed 15 May 2026). RNA-Seq analysis From each of the 180 barley samples, 10 seeds were randomly selected, crushed between pieces of copy paper with a hammer, and ground into a fine powder using a mortar and pestle with liquid nitrogen. Approximately 50 mg of ground tissue from each sample was transferred into a 2-mL tube containing grinding beads (RNeasy PowerPlant Kit, QIAGEN Inc., Toronto, ON, Canada). Samples were flash-frozen in liquid nitrogen and stored at −80 °C until RNA extraction. Total RNA was extracted using the RNeasy PowerPlant Kit with the following modifications: (i) ground tissue was further disrupted using a TissueLyser II (QIAGEN Inc.) at 20 Hz for 5 min and (ii) RNA was eluted twice with 50 µL RNase-free water and combined prior to quantification. RNA concentration was assessed with a Thermo Scientific NanoDrop 8000 (Thermo Fisher Scientific, Mississauga, ON, Canada) and purity was assessed using the 260/280 and 260/230 absorbance ratios. RNA integrity number was determined using the RNA 6000 Nano Kit on an Agilent 2100 Bioanalyzer (Agilent Technologies, Mississauga, ON, Canada). RNA samples were stored at −80 °C until library preparation. mRNA-Seq libraries were prepared using the SENSE mRNA-Seq Library Prep Kit V2 (Lexogen, Greenland, NH, USA) with 12-nt Unique Dual Index sets A1, A2, and A3, following the manufacturer's protocol. cDNA amplification was performed with 19 to 20 PCR cycles for samples containing ≥1,000 ng of total RNA, and 23 cycles for those <1,000 ng. Final libraries were quantified using the Invitrogen Quant-iT PicoGreen dsDNA assay kit (Thermo Fisher Scientific) and then pooled into a multiplex library and assessed using an Agilent 2100 Bioanalyzer with a High-Sensitivity DNA chip. To improve the library quality, the multiplex library was further cleaned to remove small fragments (<250 bp) using 2 rounds of KAPA Pure Beads (Roche Canada, Mississauga, ON, Canada) at a 0.9× beads-to-sample ratio, following the single-tail clean-up protocol. Sequencing was performed at the Centre d’expertise de la services Génome Québec on a single lane of an Illumina NovaSeq 6000 S4 flow cell, generating 100 bp paired-end reads. Raw sequence data was deposited in June 2025 to the National Center for Biotechnology Information Sequence Read Archive under the BioProject ID PRJNA1274897. SNP calling RNA sequencing generated a pair of demultiplexed forward and reverse FASTQ files for each sample. FASTQ files were assessed with FastQC v0.11.9 (Andrews 2010) and summarized for each read direction using MultiQC v1.10.1 (Ewels et al. 2016). The FASTQ files were processed using Trimmomatic v0.32 (Bolger et al. 2014) to remove any adapter sequences, trim low-quality bases (below a Phred score of 24), and remove any sequences shorter than 60 bp. The following trim settings were used: ILLUMINACLIP: TruSeq3-PE-2.fa; SLIDINGWINDOW:10:24; and MINLEN:60. FastQC was rerun after trimming to verify the removal of the Illumina adapter sequences. A split-chromosome version of the barley reference (Mascher et al. 2017) was obtained from Dr Martin Mascher, IPK Gatersleben, Germany. The split-chromosomes have the same number of bases and are similarly masked as the “160404_barley_pseudomolecules_masked.fasta” (Mascher et al. 2017) and the EnsemblPlants v40 version of the reference. However, the split-chromosome reference was used to circumvent the 239−1 base (≈536 Mb) limit of SAMtools (Li et al. 2009, https://www.htslib.org) binary alignment map (BAM) indexing. A graph Ferragina–Manzini index of the genome reference was generated using Hisat2-build (Kim et al. 2019), and alignments against the split genome sequences were performed using Hisat2 (Kim et al. 2019), a fast, splice-aware aligner. The resulting BAM files were filtered to remove PCR duplicates using the MarkDuplicates tool from the Genome Analysis Toolkit v4.2.6.1 (Van der Auwera et al. 2013). SAMtools sort was applied to produce sorted BAM files. To increase the accuracy of identifying deleterious variants, SNP calling was performed using 2 commonly applied methods: BCFtools v1.9 (Danecek et al. 2021) and ANGSD v0.935 (Korneliussen et al. 2014). BCFtools is one of the most widely used variant calling and SNP genotyping tools, based on estimated genotype likelihoods at each genomic position, and is therefore expected to generate SNP data sets with missing data. ANGSD applies a probabilistic framework to call SNPs based on estimated genotype likelihoods by optimizing allele frequencies and commonly generates SNP data sets without missing data; however, mapping bias (favoring reference over alternative alleles) may occur (Günther et al. 2025). Specifically, BCFtools was used with the following command: bcftools mpileup -Ou -f -b | bcftools call -vmO z -V indels -o. To control SNP quality and the level of missing data, Vcftools v0.1.15 (Danecek et al. 2011) was applied with the following command: vcftools –vcf input.vcf –out output.vcf –recode-INFO-all –max-alleles 2 –min-alleles 2 –minDP 10 –minQ 20 -max-missing 80 –recode. This filtering generated a BCFtools-based data set of SNPs with up to 20% missing values. For the ANGSD application, the following parameters were selected: -dovcf 1 -gl 1 -dopost 1 -domaf 1 -doMajorMinor 4 -doGeno 3 -snp_pval 1e-6, generating an ANGSD-based SNP data set without missing values. These 2 methods of SNP calling were applied to all 180 barley sample BAM files based on the split-chromosome reference, and the generated variant call format (VCF) files were then realigned to full-length chromosomes with a custom Perl script. Identification of deleterious SNPs For both SNP data sets, annotation was performed using the stand-alone Ensembl Variant Effect Predictor (VEP) (McLaren et al. 2016; Naithani et al. 2017) from the generated SNP VCF file. The VEP output also included the score of the sorting intolerant from tolerant (SIFT) algorithm (Vaser et al. 2015) for each annotated SNP. The SIFT score measures the predictive impact of an amino acid substitution and can distinguish between functionally neutral and deleterious amino acid changes. An amino acid substitution with a SIFT score of ≤0.05 is predicted to be deleterious. This study also applied a combination of SIFT scores and GERP++ Rejected Substitution (RS) scores (Davydov et al. 2010) to identify deleterious SNPs (dSNPs). The RS scores for the extremely conserved chromosomal regions of the barley genome were previously generated using GERP++ based on reference genomes of 12 plant species to measure the phylogenetic conservation from the substitution of a locus (Fu et al. 2023). These published RS scores provide a quantification of the conservation of each nucleotide in a multispecies alignment. A positive score (RS >0) at a substitution site means fewer substitutions than expected. Thus, substitutions occurring at a conserved site with RS >0 are predicted to be deleterious, with higher RS scores indicating increasingly deleterious effects. Combining SIFT (<0.05) and GERP++ RS (>0) annotations (or SIFT-RS combination) enabled more accurate identification of dSNPs in constrained portions of the genome. In this study, an additional step was included to identify multiple data sets of dSNPs that were shared between the BCFtools-based and ANGSD-based SNP data sets, depending on variable levels of missing SNP data. The shared dSNPs are not only expected to have an improved accuracy in the identification of deleterious variants, but also can provide more informative comparisons of mutation burden estimations based on different SNP genotyping methods. Characterizing the identified dSNPs For each SNP genotype VCF file, the total detected SNPs were counted for each chromosome and plotted using R hist() (R Core Team 2024). Similarly, for each deleterious SNP genotype VCF file, the shared dSNPs were also counted for each chromosome. The SNP counting was done with a custom shell script. For BCFtools-based dSNP data sets, minor allelic frequencies of dSNPs in the assayed samples were analyzed using Vcftools and plotted using R plot(). Similarly, minor allelic frequencies of ANGSD-based shared dSNPs were also analyzed to compare their frequency distributions. To understand the detrimental severity of the shared dSNPs, their RS values were also extracted and plotted using R plot() to assess their frequency distribution. Gene ontology (GO) enrichment analysis was also performed on the dSNPs that were shared between BCFtools- and ANGSD-based SNP data sets. The first step was to identify and extract genes associated with the shared dSNPs from the barley gene annotation file using a custom shell script. The associated gene list was analyzed using the online tool agriGO v2.0 (Tian et al. 2017) to generate enriched GO terms. The resulting significant GO term sets were further analyzed and visualized using the online tool REVIGO (Supek et al. 2011) with treemaps to assist in the interpretation of gene enrichments and functions. Method for estimating mutation burden change A new method was specifically developed to estimate mutation burden change over 1 regeneration cycle using paired G0 and G1 samples with varying durations of cold storage conservation. G0 and G1 samples could differ in the number of conservation years since original acquisition (Y0) and since the first germplasm regeneration (Y1), respectively. Consequently, the number of conservation years between acquisition and the first regeneration, YR (=Y0−Y1), also varies among the assayed accessions. These differences in Y0, Y1, and YR reflect the variations in the timing of accession acquisition, regeneration, and storage. For example, 2 accessions, A and B, were acquired 40 and 30 yr ago, respectively, and underwent their first regenerations 15 and 20 yr ago, resulting in differing conservation year values (Y0A = 40 vs Y0B = 30; Y1A = 15 vs Y1B = 20; and YRA = 25 vs YRB = 10). Following the commonly used measure of mutation burden (eg Ramu et al. 2017), the total mutation burden per deleterious locus Mti for a sample representing accession i is estimated as the proportion of the predicted deleterious variants over all the identified variants in the sample as: where Chet,iandChom,i are the counts of deleterious variants in heterozygous and homozygous loci in the G0 or G1 sample of the accession i (i = 1 to n) for all diallelic deleterious loci L. The change in total mutation burden per deleterious locus per conservation year between the paired G0 and G1 samples of 1 accession i is measured as DMtyi: Note that Mt1i and Mt0i are the total mutation burdens per deleterious locus for the G1 and G0 sample of the accession i, respectively. As DMtyi may be correlated with the years of conservation from G0 to G1 (YRi) for the accession, the change should be adjusted for YRi. Thus, the DMtyi that adjusts for variable years of conservation from G0 to G1 per accession is redefined as: Note that YR¯=1n∑inYRi, the mean of variable conservation years YRi for all accessions. The average change across all assayed accessions with variable years of conservation (Y0, Y1, and YR) over 1 regeneration cycle is calculated as: DMtyRi from Equation (3) is scaled to a deleterious locus and a conservation year and has the unique property of being independent of the number of deleterious mutations L being identified and the number of years under conservation (Y0, Y1, and YR) for each sample. Thus, DMtyR from Equation (4) not only allows for an informative comparison of the mutation burden changes estimated from different studies but also makes the projection of total mutation burden in conserved germplasm feasible with the assumption of variable levels of deleterious mutations and different conservation scenarios. A positive or negative DMtyR value indicates the accumulation or disappearance of deleterious alleles in a conserved accession after regeneration, respectively. For this study, the 3 conservation years (Y0, Y1, and YR) for each accession were calculated from Table 1. Counting deleterious variants in heterozygotes and homozygotes across all deleterious loci for each sample was performed using a custom shell script from each dSNP data set, so Mti was obtained using Equation (1) for each G0 or G1 sample of the accession i. For BCFtools-based dSNP data sets, Mti was calculated based only on the effective deleterious loci without missing SNP data. DMtyi and DMtyRi were estimated using Equations (2 and 3), respectively, for the accession i from each dSNP data set. The average change DMtyR was estimated using Equation (4) across all 90 accessions. To validate the need for adjusting for YR for DMtyR, a linear regression analysis of DMtyi over YRi was made in each dSNP data set to determine the significance and strength of their correlation. The acquired estimates of DMtyRi were further analyzed for their distributions in each dSNP data set using R barplot() and tested for (i) their correlations between BCFtools-based and ANGSD-based dSNP data sets using R lm(), (ii) significance of their median larger than zero using Wilcoxon signed-rank test implemented with R wilcox.test(), (iii) their differences between 2 dSNP data sets using Wilcoxon signed-rank test implemented with Wilcoxsign_test() in R package coin, and (iv) their overall differences among different dSNP data sets using nonparametric Kruskal–Wallis test with R kruskal.test(). Comparative analyses of DMtyR estimates were also made with respect to the extent of missing data and SNP calling method. Assessing allelic changes from G0 to G1 samples To understand the changes of deleterious alleles over 1 regeneration cycle, we conducted several comparative counts for each dSNP data set. Specifically, we quantified (i) the numbers of deleterious loci and alleles in G0, G1, and G0+G1 samples; (ii) deleterious alleles and loci shared between G0 and G1; (iii) deleterious alleles detected in G1 but undetected in G0 and their associated loci; and (iv) deleterious alleles detected in G0 but undetected in G1 and their associated loci. We also counted deleterious alleles in homozygotes and heterozygotes and estimated the contributions of each to total mutation burden without considering the conservation years. As only one single G0–G1 sample pair for a heterogenous accession was assayed and the paired G0–G1 sample did not originate from the same initial plant, these comparative counts are not informative for determining whether an observed allelic turnover resulted from de novo mutation, standing variation, or a sampling effect. However, these analyses enable an assessment of the extent of allelic maintenance and turnover from G0 to G1, thus enhancing our understanding of mutation dynamics in conserved germplasm. Results SNP identification and annotation RNA-Seq produced 2.29 billion sequence reads for the 180 barley samples with an average of 12.7 million sequence reads per sample (Supplementary Table 1). Specifically, there were an average of 8.4 and 9.0 million mapped sequence reads for each G0 and G1 sample, respectively. SNP calling by BCFtools identified 513,997 SNPs with up to 99.4% missing data across the 180 samples. After filtering based on SNP quality and missing data, there were 122,463, 72,431, 42,763, and 4,768 SNPs with up to 20%, 10%, 5%, and 0% missing data, respectively. In contrast, SNP calling by ANGSD revealed 537,776 SNPs without missing data. The identified SNPs were widely distributed across all 7 barley chromosomes in both BCFtools- and ANGSD-based SNP data sets and displayed similar chromosomal distributions, as illustrated in Fig. 1a and b. The chromosomal distributions of all identified SNPs from BCFtools with up to 20% missing data (a) and from ANGSD (b) and predicted dSNPs (c), frequency distributions of shared dSNPs (d and e), and frequency distributions of GERP++ RS values for the shared dSNPs (f). Each panel is labeled with its specific data set. a and b) Show the chromosomal distributions of all SNPs from BCFtools and ANGSD, respectively. c) Illustrates the chromosomal distributions of dSNPs shared between BCFtools- and ANGSD-based data sets. d and e) Show the frequency distribution of shared dSNPs in m20-b and m20s-a data sets, respectively. VEP-based annotation analyses of the identified SNPs were performed on BCFtools-based data sets of SNPs with up to 20% missing data and ANGSD-based SNP data set. These VEP analyses allowed for the classification of SNPs into up to 17 different classes with the most severe consequences (Table 2). The amount of SNPs per class varied substantially among the 17 classes and 2 SNP data sets representing the 2 SNP calling methods. For example, there was a higher proportion of missense_variant over total SNPs identified in the BCFtools-based SNP data set (0.302) than in the ANGSD-based SNP data set (0.257). Similarly, there was a lower proportion of loss-of-function variants over total SNPs identified in the BCFtools-based SNP data set (0.027) than in the ANGSD-based SNP data set (0.040). Table 2.Annotations of SNPs generated by BCFtools and ANGSD for 180 assayed samples representing paired G0 (original acquisition) and G1 (first regeneration) samples from 90 barley accessions.VariantSNPs from BCFtoolsSNPs from ANGSDSNP calling Total SNPs122,463537,776SNP annotation with VEP (most severe consequences) Missense_variant (MV)36,926138,248 Proportion of MV in total SNPs0.3020.257 Synonymous_variant (SV)27,22298,213 Proportion of SV in total SNPs0.2220.183 Splice_acceptor_variant1001,988 Splice_donor_variant1311,434 Stop_gained2,64610,341 Stop_lost173525 Start_lost98729 Splice_region_variant486,347 Stop_retained_variant127126 Coding_sequence_variant24 5_prime_UTR_variant2,96928,585 3_prime_UTR_variant34,70094,573 Non_coding_transcript_exon_variant4,97518,024 Intron_variant75633,494 Upstream_gene_variant1,53316,211 Downstream_gene_variant2,34722,817 Intergenic_variant6,86966,117Loss-of-function (LOF) varianta Total count3,32321,490 Proportion0.0270.040SIFT analysis with CTb SIFT-deleterious SNPs (SDS)14,99749,976 Proportion of SDS in total SNPs0.1220.093Deleterious SNPs by SIFT+RS SDS+RS-filtered SNPs (RSD)2,0947,508 Proportion of RSD in total SNPs0.01710.0140Shared RSD between 2 SNP calls SNPs with up to 20% missing values1,6421,642 SNPs with up to 10% missing values750750 SNPs with up to 5% missing values325325 SNPs without missing values99The SNP data set from BCFtools had up to 20% level of missing data in 180 samples, and the ANGSD data set had no missing values. aLOF variants consist of those from the annotation classes of 3 “Stop_,” 3 “Splice_,” and 1 “Start_lost.” bSIFT-filtered with canonical transcripts (CTs). Annotations of SNPs generated by BCFtools and ANGSD for 180 assayed samples representing paired G0 (original acquisition) and G1 (first regeneration) samples from 90 barley accessions. The SNP data set from BCFtools had up to 20% level of missing data in 180 samples, and the ANGSD data set had no missing values. aLOF variants consist of those from the annotation classes of 3 “Stop_,” 3 “Splice_,” and 1 “Start_lost.” bSIFT-filtered with canonical transcripts (CTs). Deleterious mutation Analyzing SIFT scores for SNPs associated with canonical transcripts revealed 14,997 and 49,976 predicted dSNPs for the BCFtools- and ANGSD-based SNP data sets, respectively, as illustrated in Table 2. Combining SIFT scores with GERP++ RS scores identified much fewer, but more accurately predicted, 2,094 and 7,508 dSNPs than those from the SIFT scoring alone for 2 SNP data sets (Table 2). Extra efforts were made to identify dSNPs that were shared between the BCFtools- and ANGSD-based SNP data sets. A total of 1,642, 750, 325, and 9 dSNPs that represented the BCFtools-based SNP data set with the missing levels of 20%, 10%, 5%, and 0%, respectively, were found to be shared with the ANGSD-based SNP data set (Table 2). To make the comparative analysis of mutation burden easier, the first 3 pairs of shared BCFtools- and ANGSD-based dSNP data sets were named as m20-b vs m20s-a, m10-b vs m10s-a, and m5-b vs m5s-a, respectively. The shared 9 dSNPs for 0% missing data were too small to be further considered. Analyzing the chromosomal distributions of dSNPs revealed their wide distributions across 7 barley chromosomes in m20-b with shared dSNPs (Fig. 1c). Further analysis of their frequency distributions showed similar L-shaped allelic frequency distributions with slightly different means of 0.044 and 0.052 for the m20-b vs m20s-a data set, respectively (Fig. 1d and e). These results meant that most dSNPs were infrequent and harbored in an average of 8 to 9 (out of 180) samples or fewer. Further assessment on the frequency distribution of RS values for shared m20-b revealed that these deleterious alleles had a mean RS value of 0.64 in a range of 0.067 to 1.64 (Fig. 1f). These distributions of RS values indicated that these deleterious alleles were predicted to be weakly detrimental. Only 194 (9.3%) and 126 (6%) dSNPs had RS values greater than 0.64 and 1.00, respectively. The 1,642 shared dSNPs were found to be associated with 1,025 barley genes across 7 chromosomes. The ontology analyses of the associated genes using agriGO identified 103 significant (P < 0.05) GO terms. Analysis of these significant GO terms using REVIGO revealed 56 biological processes (Fig. 2a), 11 cellular components (Fig. 2b), and 23 molecular functions (Fig. 2c) for the associated genes. The major biological processes included protein catabolic process, purine-containing compound metabolic process, cellular localization, iron–sulfur cluster assembly, and response to abiotic stimulus (Fig. 2a). The major cellular components were organelle envelope, membrane, endomembrane system, and cytoplasm (Fig. 2b). The major molecular functions were GTP binding, ATP hydrolysis activity, and oxidoreductase activity acting on the CH-OH group of donors, NAD or NADP as acceptor (Fig. 2c). As expected for stored seeds, these associated genes were mainly involved with protein catabolic process, purine-containing compound metabolic process, response to abiotic stimulus, organelle envelope, membrane, and different molecular bindings. REVIGO gene ontology treemaps showing the cluster representatives of 56 biological processes (a), 11 cellular components (b), and 23 molecular functions (c) associated with 103 significant GO terms that were extracted from 1,025 genes associated with the 1,642 shared dSNPs. Mutation burden change Linear regression analyses revealed significant correlations between the estimates of the change in total mutation burden per deleterious locus per conservation year in an accession (DMtyi) and its conservation years from G0 to G1 in 3 BCFtools-based and 3 ANGSD-based dSNP data sets (Fig. 3). The significance levels obtained were at least 2.0 × 10−16, and the estimates of correlation coefficient were positively high, ranging from 0.791 (Fig. 3d) to 0.857 (Fig. 3b). Interestingly, the changes in total mutation burden were largely flat before the conservation years of 25 and started to increase significantly up to the assayed maximum years of 33 (Fig. 3). These results indicated that the number of conservation years from G0 to G1 was a strong indicator for the change in total mutation burden for the assayed barley accessions. Significant correlations between the estimates of the change in total mutation burden per deleterious locus per conservation year in 1 accession (DMtyi) and its conservation years from G0 to G1 in 3 BCFtools-based (a to c) and 3 ANGSD-based (d to f) dSNP data sets. The change in total mutation burden per deleterious locus per conservation year between the paired samples of 1 accession with the YR adjustment was estimated as DMtyRi using Equation (3). Accession-wise DMtyRi estimates were obtained for 3 paired dSNP data sets (m20-b vs m20s-a, m10-b vs m10s-a, m5-b vs m5s-a). Analyzing these DMtyRi estimates revealed that these estimates had skewed normal distributions in all 6 dSNP data sets (Fig. 4a and b). Wilcoxon signed-rank tests revealed that accession-wise DMtyRi estimates in each dSNP data set had a median that was significantly (P < 0.00001) larger than zero. Regression analyses showed highly significant correlations of these DMtyRi estimates between each pair of dSNP data sets and their estimated correlation coefficients ranged from 0.486 to 0.958 (Fig. 4c). The estimates of the change in total mutation burden per deleterious locus per conservation year over 1 regeneration in an accession (DMtyRi) in 3 BCFtools-based dSNP data sets (m20-b, m10-b, and m5-b) (a1, a2, and a3) and their corresponding ANGSD-based dSNP data sets (m20s-a, m10s-a, and m5s-a) (b1, b2, and b3), along with the correlation between BCFtools-based and ANGSD-based DMtyRi estimates (c1, c2, and c3). The comparative statistics of accession-wise DMtyRi estimates were summarized for 3 paired BCFtools-based and ANGSD-based dSNP data sets in Table 3. The average estimates of the accession-wise DMtyRi estimates as DMtyR values were significantly smaller in the BCFtools-based, than the corresponding ANGSD-based, dSNP data sets, except for the pair m10-b vs m10s-a (Table 3). For example, the DMtyR value was 0.00123 and 0.00130 for m20-b and m20s-a, respectively. These positive DMtyR values indicated an accumulation of deleterious variants in conserved barley germplasm after the first regeneration. Interestingly, the DMtyR values ranged from 0.00123 to 0.00140 with a mean of 0.00133 for 3 BCFtools-based dSNP data sets and from 0.00130 to 0.00148 with a mean of 0.00137 for 3 ANGSD-based dSNP data sets, but these DMtyR variations were not significantly (P > 0.05) different among 3 BCFtools-based or ANGSD-based dSNP data sets, based on the Kruskal–Wallis tests. Table 3.The comparative statistics of estimating the change in total mutation burden per deleterious locus per conservation year over 1 regeneration in each barley accession (DMtyRi) in 6 paired dSNP data sets generated from BCFtools (m20-b, m10-b, and m5-b) and from ANGSD (m20s-a, m10s-a, and m5s-a), along with the results of the Wilcoxon signed-rank tests for differences in DMtyRi between paired dSNP data sets.Paired dSNP data setsdSNPsMissing data (%)Mean (DMtyR)Standard deviationMinimumMaximumRank test (P)m20-b1642200.001230.000840.000020.003570.00550m20s-a164200.001300.00092−0.000270.00377m10-b750100.001370.00132−0.001250.005820.59950m10s-a75000.001340.00096−0.000480.00383m5-b32550.001400.00113−0.000340.004620.03214m5s-a32500.001480.00127−0.001490.00532 The comparative statistics of estimating the change in total mutation burden per deleterious locus per conservation year over 1 regeneration in each barley accession (DMtyRi) in 6 paired dSNP data sets generated from BCFtools (m20-b, m10-b, and m5-b) and from ANGSD (m20s-a, m10s-a, and m5s-a), along with the results of the Wilcoxon signed-rank tests for differences in DMtyRi between paired dSNP data sets. Allelic changes from G0 to G1 samples Analyzing the changes of deleterious variants between 90 G0 and 90 G1 samples revealed consistently more deleterious variants predicted in the G1 samples than the G0 samples, as illustrated with the paired dSNP data sets (m20-b vs m20s-a) (Table 4). For example, there were 11,376 and 12,319 dSNPs for the G0 and G1 samples in m20-b, showing a burden change of 0.00319 per deleterious locus without adjusting for variable conservation years among accessions. In contrast, m20s-a had 12,375 and 13,381 dSNPs for the G0 and G1 samples, respectively, having a burden change of 0.00340 per deleterious locus. Table 4.Comparative counting of the deleterious alleles between the paired samples of 90 barley accessions representing the original generation (G0) and the first regeneration (G1) in m20-b and m20s-a data sets sharing 1,642 dSNPs.Data/sampleLocus/alleleTotalShared alleles between G0 and G1New alleles detected in G1G0 alleles not detected in G1Estimate of total mutation burdenEstimate of the burden changem20-b G0+G1Loci1,6421,5911734 G0+G1Alleles23,69523,5356595 G011,37611,281950.03849(11,082 + 294)(10987 + 294)(95 + 0) G112,31912,254650.041680.00319(11,957 + 362)(11892 + 362)(65 + 0)m20s-a G0+G1Loci1,6421,5921535 G0+G1Alleles25,75625,6105096 G012,37512,279960.04187(12,049 + 326)(11953 + 326)(96 + 0) G113,38013,331500.045270.00340(12,985 + 396)(12,935 + 396)(50 + 0)Two deleterious allelic counts in parentheses represent the numbers of deleterious alleles in heterozygotes and homozygotes in the 90 samples, respectively. Comparative counting of the deleterious alleles between the paired samples of 90 barley accessions representing the original generation (G0) and the first regeneration (G1) in m20-b and m20s-a data sets sharing 1,642 dSNPs. Two deleterious allelic counts in parentheses represent the numbers of deleterious alleles in heterozygotes and homozygotes in the 90 samples, respectively. Comparative counting of deleterious variants that were newly detected in the 90 G1 samples and undetected in the 90 G0 samples for these paired dSNP data sets revealed 65 new alleles at 17 loci vs 50 alleles at 15 loci for m20-b vs m20s-a (Table 4). Similar counting of deleterious variants that were present in the 90 G0 samples but undetected in the 90 G1 samples for the paired dSNP data sets revealed 95 undetected alleles at 34 loci vs 96 alleles at 35 loci for m20-b vs m20s-a data sets (Table 4). Clearly, there were fewer newly detected, than undetected, alleles for a smaller share of 1,642 dSNPs. Interestingly, more than 97% of the predicted deleterious variants occurred in heterozygotes in m20-b and m20s-a data sets, and all newly detected and undetected alleles in the G1 samples were present in heterozygotes (Table 4). Clearly, deleterious allelic maintenance and turnover occurred from G0 to G1 samples and an extremely high proportion of the identified deleterious variants harbored in heterozygotes of the assayed samples. Discussion This study represented the first attempt to measure the change in total deleterious mutation burden of barley germplasm conserved under long-term cold storage over 1 regeneration cycle using seed-based RNA-Seq analysis. The analysis identified 1,642 deleterious variants by 2 SNP calling methods (Table 2). These deleterious variants had many genetic features such as wide chromosomal distribution, low allelic frequency, predominantly heterozygous form, association with diverse biological processes, and weak detrimental effect. Based on BCFtools-based SNP calls, the change in total mutation burden per deleterious locus per conservation year was estimated at 0.00123 with a standard deviation of 0.00084 for an accession conserved through the first regeneration (Table 3). More deleterious variants were found in G1 than in G0 samples; variants unique to each sample type were also observed; and a high proportion of variants occurred in heterozygotes (Table 4). These findings not only demonstrate the technical and practical feasibility of measuring mutation accumulation in conserved barley germplasm but also are valuable for enhancing long-term plant germplasm conservation. The estimated changes in total mutation burden of conserved germplasm are novel and significant, as such estimations have never been made, and they differ in biological meaning and scale from the genome-wide mutation rates commonly estimated for human and other species (eg see The 1000 Genome Project 2011; Sharp et al. 2018). The estimates of the averaged change in total mutation burden per deleterious locus per conservation year were significantly larger than zero. This finding alone has provided the first genomic evidence on the accumulation of deleterious variants in conserved barley germplasm after the first regeneration. Interestingly, the predicted deleterious variants also displayed several informative genetic features, including their presence in only 8 or fewer samples and the occurrence of 97% variants in heterozygotes, their associations with diverse biological processes, and the prediction of weakly deleterious effects. These features are consistent with previous findings of accumulating heterozygous mutation burden in conserved soybean germplasm after up to 10 regenerations (Fu and Horbach 2025) and in other plant genomes (Zhu et al. 2022). Together, these findings confirm that the accumulation of heterozygous deleterious mutations is the major mode of mutation accumulation in conserved germplasm. Based on the genetic theories on deleterious mutations (Charlesworth et al. 1990), they also imply the technical difficulty of eliminating these genome-wide, weakly detrimental variants from conserved germplasm (Fu 1999), thereby increasing the genetic cost associated with maintaining conserved germplasm. The most significant finding was the estimated extent of deleterious variant accumulation in an accession conserved over 1 regeneration: 0.00123 per deleterious locus per conservation year with a standard deviation of 0.00084. Based on such an extent of mutation burden change with 1,642 loci and averaged 16 yr of conservation for the first regeneration, a barley accession is predicted to accumulate 28 to 37 deleterious variants with a mean of 32 (=0.00123 × 1,642 × 16) over the first regeneration. This predicted range of accumulated deleterious variants was lower than the observed counts of new deleterious variants identified in G1 samples (65 in m20-b and 50 in m20s-a), which did not account for the variable conservation years among the assayed accessions (Table 3). As the estimate of the burden change (DMtyR) was scaled to a deleterious locus and a conservation year, it is possible for a genebank manager to make a projection of total mutation burden in germplasm conserved over a period. For example, if the number of averaged conservation years for the first regeneration is 32, not 16 as in this study, there would be 64 deleterious variants projected to be accumulated in the conserved barley germplasm. To make the projection more accurate and useful, however, more measurements of mutation burden change with variable years of conservation are needed to establish a baseline data set of deleterious mutation accumulation for a germplasm collection. Two methods of SNP calling based on BCFtools and ANGSD were applied with the goal to identify their shared dSNPs and to compare the differences in the estimates of mutation burden change. The use of the shared dSNPs acquired from the BCFtools-based and the ANGSD-based dSNP data sets should improve the accuracy of identifying dSNPs with the SIFT-RS combination alone. The 1,642 shared dSNPs in this study were smaller than the 3,855 dSNPs identified by Kono et al. (2019) from exome capture resequencing of 21 barley lines, but larger than the 1,148 dSNPs identified by Fu et al. (2023) from 70 barley samples and the average of 1,000 deleterious variants per barley accession identified by Kono et al. (2016). As expected, the 1,642 dSNPs were widely spread over 7 chromosomes and associated with 1,025 expressed genes conditioning diverse biological processes, cellular components, and molecular functions. These GO results also revealed a genetic link of some deleterious variants conditioning the response to abiotic stress (probably the cold condition) (Fig. 2a). A comparison of the estimated changes in total mutation burden per deleterious locus per conservation year between 3 paired BCFtools-based and ANGSD-based dSNP data sets revealed some significantly higher ANGSD-based estimates than BCFtools-based estimates, although the differences were relatively small (Table 3). These higher ANGSD-based estimates are consistent with the larger total number of deleterious alleles in m20s-a (25,756) compared with m20-b (23,695), and with the greater G0 to G1 change in deleterious alleles for m20s-a (1,006) than for m20-b (943) (Table 4). It is worth noting, however, that some allelic changes from G0 to G1 may be false positives caused by sampling effects, as the G0–G1 samples originated from different initial plants and only one pair was sampled for each heterogeneous accession. Also, the higher estimates may reflect mapping biases introduced by allele-frequency optimization procedures designed in ANGSD to genotype low-frequency alleles (Günther et al. 2025), which can enable the detection of rare deleterious alleles that BCFtools-based SNP calling would otherwise classify as missing. In contrast, the lower BCFtools-based estimates were conservative because they were based on the effective deleterious loci defined in Equation (1), which excluded loci with missing data rather than using all 1,642 loci. For example, for a sample with a total number of deleterious variants (for both heterozygotes and homozygotes) counted as 1,500 effective loci and with 142 missing data loci, the total mutation burden per deleterious locus is calculated with the total variant count divided by 3,000 (=1,500 loci × 2), rather than 3,284 (=1,642 loci × 2). Thus, missing data were expected to have less impact on the estimations of the burden change, as supported with nonsignificant differences in DMtyR among m20-b, m10-b, and m5-b data sets. This feature alone tends to favor the choice of BCFtools-based SNP calling for mutation burden analysis, in spite of the strong correlation (R = 0.958) between the estimates of the burden changes from m20-b and m20s-a data sets (Fig. 4c). Germplasm conservation in a genebank is unique with the regular (although not necessarily mandated) maintenance of multiple samples of an accession after germplasm regenerations. This unique feature provides the opportunity to measure the changes of deleterious variants in conserved germplasm. As illustrated in this study, seed-based RNA-Seq analysis of the paired samples of 90 barley accessions representing the germplasm before and after 1 regeneration cycle can allow for estimating mutation burden changes over 1 regeneration cycle (Table 3). However, several issues are worth mentioning for future improvements. First, our study did not consider the variabilities in the number of accessions assayed and in mutation burden within accessions (Fu 2024b). These variabilities could affect the estimation accuracy and/or precision, as each barley accession was genetically heterogenous (due to outcrossing or sample contamination) and represented only by a paired G0–G1 sample that originated from different initial plants. Such a sample pair cannot determine whether an observed allelic turnover was resulted from de novo mutation, standing variation, or simply a sampling effect. Thus, future informative analysis should utilize a minimum of 30 accessions, each containing 15 to 30 G0–G1 sample pairs. Second, the variabilities in years of conservation (Y0, Y1, and YR) for a conserved accession could also affect the estimation of mutation burden change if these conservation years are not linearly correlated with deleterious allelic changes (Fig. 3). More empirical and theoretical research is needed to analyze their impacts of nonlinear correlation on the estimation accuracy. Third, the GERP++ RS scores generated based on reference genomes of 12 plant species (Fu et al. 2023) for 1,642 deleterious variants ranged from 0.07 to 1.64 with a mean of 0.64, suggesting that some predicted deleterious variants, even based on 2 methods of SNP calling, may be false positives (Kistler et al. 2018; Wu et al. 2023). Fourth, representation bias in seed-based RNA-Seq analysis also existed (eg see Shi et al. 2021), even with the sequence alignments by a splice-aware aligner Hisat2 (Kim et al. 2019), and false positives may have also occurred for some predicted deleterious variants. The reported findings are valuable for enhancing long-term plant germplasm conservation in a genebank. First, the approach developed here for measuring mutation burden changes is technically achievable and practically feasible for its applications to seed germplasm collections of different species conserved in genebanks. There are RNA extraction protocols available for seeds of many plant species (see eg Tetreault et al. 2025), genome sequences for more than 1,800 plant species (Schwacke et al. 2025), and many advanced genomic tools to identify deleterious variants. In this study, the RNA-Seq analysis of 180 barley samples had an experimental cost in 2022 of CAD 18,000 (including RNA extraction kit, library preparation kit, and sequencing), and the whole mutation screening can be completed within a few months from seed RNA extraction, sequencing, to mutation analysis. Second, more research is needed to acquire baseline information on mutation burden changes in seed germplasm of different species conserved under long-term cold storage, as the extent and mechanism of their mutation accumulation after regenerations may differ. Linking mutation burden changes to storage conditions and/or conservation practices should be prioritized, if possible, to enable the development of specific mitigation strategies. Collaborative research among worldwide genebanks is also needed to expand the scope of measuring mutation burden changes with germplasm collections of different species under variable conservation conditions. Third, with the baseline information on mutation burden changes, genebank managers could project the extent of deleterious mutation accumulation in conserved germplasm under variable years of cold storage. Accordingly, effective genebank managements for long-term conservation can be explored to minimize the genetic changes in conserved germplasm (Fu 2024a). For example, for those germplasm collections with a high level of mutation accumulation, existing cold storage conditions may need to be optimized along with long-term regeneration frequency (Richards et al. 2010; Redden and Partington 2018). For long-term conservation involving cycles of germplasm regeneration, genebank managers may utilize estimated mutation burden changes to extrapolate overall genetic changes in conserved germplasm resulting from mutation accumulation alone, expressed as the projected number of accumulative deleterious variants divided by genome size. Conclusion This seed-based RNA-Seq analysis of 90 barley accessions with paired samples representing germplasm conserved under cold storage since its original acquisition (G0) and since the first regeneration (G1) identified 1,642 deleterious variants. These variants were widely distributed across 7 chromosomes, were present in only 8 or fewer samples, occurred predominantly in heterozygotes, were associated with diverse biological processes, and were largely predicted to be weakly detrimental. Based on BCFtools-based SNP calls, the change in total mutation burden per deleterious locus per conservation year was estimated at 0.00123, with a standard deviation of 0.00084, for an accession conserved through the first regeneration. This estimate suggests that a conserved barley accession accumulated 28 to 37 deleterious variants over the G0 to G1 period. More deleterious variants were found in G1 than in G0 samples; variants unique to each sample type were also observed; and a high proportion of variants occurred in heterozygotes. These findings clearly demonstrate the technical and practical feasibility of measuring mutation accumulation in barley germplasm conserved in a genebank. Supplementary Material