Recovery of Novel Sequence Variants in Chemically Mutagenized Seed and Vegetatively Propagated Coffea arabica L.

Abstract

The negative effects of climate change impact both farmers and consumers. This is exemplified in coffee, one of the most widely consumed beverages in the world. Yield loss in high-quality Coffea arabica L., due to the spread of coffee leaf rust (Hemileia vastatrix), results in lower income for subsistence farmers and volatile prices in markets and cafes. Genetic improvement of crops is a proven approach to support sustainable production while mitigating the effects of biotic and abiotic stresses and simultaneously maintaining or improving quality. However, the improvement of many species, including coffee, is hindered by low genetic diversity. This can be overcome by inducing novel genetic variation via treatment of seeds or cells with mutagens. To evaluate this approach in coffee, mutant populations created by incubating seed or embryogenic calli with the chemical mutagens ethyl methanesulphonate or sodium azide were subject to reduced-representation DNA sequencing using the ddRADseq approach. More than 10,000 novel variants were recovered. Functional analysis revealed hundreds of sequence changes predicted to be deleterious for gene function. We discuss the challenges of unambiguously assigning these variants as being caused by the mutagenic treatment and describe purpose-built computational tools to facilitate the recovery of novel genetic variation from mutant plant populations.

1. Introduction

It is estimated that more than 2 billion cups of coffee are consumed daily, making it one of the most popular beverages [1,2]. Like wine, the complex flavor profiles of artisanal coffee have attracted connoisseurs [3]. In addition to the satisfaction it gives consumers, it is also an economically important crop in more than 80 countries in the tropics and subtropics [4]. While the genus Coffea represents 124 species, Coffea arabica L. (Arabica) and Coffea canephora (Robusta) are most frequently consumed, with Arabica being considered of higher quality [5]. Terroir in high-value, single-origin arabica coffee is a combination of the environment, including the growing location and altitude, and the genotype [6,7].

C. arabica L. is an autogamous allotetraploid (2n = 4x = 44) with preferential chromosome pairing and disomic inheritance resulting from spontaneous hybridization of diploids C. eugenioides and C. canephora. [8,9]. Arabica is propagated both sexually (seed) and asexually through methods including graftings and cell culture [10,11]. Newly planted coffee trees may take 3–4 years before bearing fruit, and seed germination requires between 1 and 2 months [12]. Like many plant species that humans rely upon, coffee has been subjected to genetic bottlenecks that have resulted in low genetic diversity [8,13]. Genetic diversity is a key component for improving crop resilience to climate change and variation, including resistance to diseases [14]. Coffee leaf rust (CLR) caused by Hemileia vastatrix, for example, poses a major global threat to Coffea arabica L. production and can result in yield reductions of up to 25%, resulting in losses of up to 2 billion dollars annually [15].

Low genetic diversity can be overcome by inducing mutations in DNA. This was first described in plants nearly 100 years ago when treatment of seed with ionizing radiation was shown to create novel phenotypes in insects and plants [16,17]. The use of chemicals to induce mutations was established by the late 1940s [18]. Importantly, mutations created through mutagenesis are unique in that many may not have previously existed in nature, providing completely novel genotypic variation that can be created directly in elite genotypes. In addition, unlike other technologies for altering genetic composition, there are no regulations surrounding growing and transporting plant materials derived from inducing mutations [19]. The approach was quickly adopted as a tool for plant breeding, and today, there are more than 3400 officially released mutant crop varieties in over 170 species registered in the IAEA’s mutant variety database (https://mvd.iaea.org/, accessed on 27 July 2024). Most crops are seed-propagated annuals, such as cereals, that produce seed and are harvested within one year. The predominance of these species is likely due to a combination of their importance for food security and the global economy, as well as the relative ease of inducing and recovering novel mutations. Technological innovations over the past 25 years have enabled a growing body of knowledge on the effects of mutagenesis on plant genome sequence [20,21]. For example, treatment of seed with gamma irradiation or X-rays results in a broad spectrum of mutations, including large insertions and deletions (indels), rearrangements, small indels, and single-nucleotide variants (SNVs) [22,23,24]. Chemical mutagens, on the other hand, produce primarily SNVs [25]. Large data sets on the effects of chemical mutagens, especially the mutagen ethyl methanesulphonate (EMS), on the sequence of genomic DNA have been gathered as the result of reverse genetic screens using the TILLING (Targeting-Induced Local Lesions IN Genomes) approach. These data can be used to establish expectations when creating a new mutant population. For example, a tetraploid plant treated with the optimal dosage of the chemical mutagen EMS or sodium azide is expected to produce one novel mutation approximately every 40 kb, with the majority of changes being G:C to A:T transitions. Chemical mutagenesis in plants has been carried out on seed, pollen, tissue, and cell cultures, with millions of novel genetic variants being reported [26,27,28]. Mutagenesis of multi-cellular seeds is most common and results in the first post-treatment generation (termed M₁) being a genetic mosaic, or chimera, where germinal and somatic cells harbor different induced mutations. This generation is thus unsuitable for genotype evaluations using conventional methods relying on genomic DNA extracted from multiple cells. Thus, seed-mutagenized material must be grown and propagated. The resulting M₂ generation is the first non-chimeric generation suitable for evaluation [29]. On the other hand, mutagenesis of embryogenic calli and generation of a plant from a single cell can result in a plant that is non-chimeric and, thus, suitable for DNA sequence evaluation without the need for meiotic propagation [30].

To evaluate the efficacy of inducing mutations in Coffee arabica L., three different mutant populations were generated from the variety Catuaí: seed mutagenesis with sodium azide [31], mutagenesis of embryogenic calli with sodium azide and mutagenesis of embryogenic calli with EMS [30]. Plants from each treatment type were previously phenotypically evaluated and shown to exhibit variation in morphological traits and response to CLR [32,33]. While phenotypic variation suggests underlying genetic variation, one objective of the present study was to investigate this further. In addition, an understanding of the effect of different mutagenic treatments on the spectrum and density of accumulated sequence variation was sought. To achieve this, reduced-representation ddRADseq and purpose-built bioinformatics tools were used to recover sequence variants unique to individual mutant lines. In total, 10,831 novel variants were recovered in mutagenized material mapping to all 22 haploid chromosomes. Functional analysis using SIFT4G revealed 356 predicted deleterious nonsynonymous changes.

2. Materials and Methods

2.1. Mutagenesis

Mutant populations were previously described [30,31]. For seed mutagenesis, genetically uniform, disease-free seeds of the Catuaí coffee variety were disinfected and then immersed for 8 h in 200 mL of distilled water containing 50 mM sodium azide in a 0.1 M potassium phosphate buffer (pH 3.0). Untreated seeds served as controls. During the incubation period, the seeds were kept in the dark at 27 ± 2 °C, with continuous shaking at 100 rpm. Afterward, the seeds were rinsed three times with distilled water for 10 min under constant agitation. The treated seeds were then sown in plastic containers filled with sterilized peat substrate and placed in a germination chamber with 100% humidity at 30 °C for eight weeks in darkness. Subsequently, the seedlings were transferred to pots with autoclaved peat substrate and grown under greenhouse conditions at 27 ± 2 °C with a 12 h light cycle until they were ready for field transplantation (Figure 1, Supplementary Table S1).

For cell cultures, two-week-old embryogenic suspension cultures (1 mL) were incubated in the dark at 26 ± 2 °C in 10 mL of TEX medium [35]. The medium was supplemented either with 5 mM sodium azide in a 0.1 M potassium phosphate buffer (pH 3.0) for 15 min or with 185.2 mM ethyl methanesulfonate (EMS) for 120 min, both under constant shaking at 100 rpm. After treatment, the cultures were washed three times with 10 mL of TEX medium (pH 5.6). Following the washes, 10 mL of R medium (pH 5.6) was added, and the cultures were incubated for 48 h in the dark on a rotary shaker at 100 rpm. For both sodium azide and EMS treatments, somatic embryos were induced, and plant regeneration was achieved by culturing the treated suspension cultures on 20 mL of semisolid R medium [36].

2.2. DNA Extraction and ddRADseq

Coffee leaves from the mutagenized and non-mutagenized plants listed in Supplementary Table S1 were collected, and DNA was extracted according to Gatica-Arias et al. [37]. Genomic DNA was isolated following a modified cetyltrimethylammonium bromide (CTAB) protocol based on Doyle (1990) [38]. A total of 100 mg of leaf material was ground in 750 µL of extraction buffer containing 2% (w/v) CTAB, 2% (w/v) polyvinylpyrrolidone (PVP), 100 mM Tris-HCl, 1.4 M NaCl, 20 mM ethylenediamine tetraacetic acid (EDTA), and 0.2% (v/v) β-mercaptoethanol at pH 8. The samples were incubated at 65 °C for 20 min with intermittent inversion. Next, 750 µL of chloroform:isoamyl alcohol (24:1) was added, and after thorough mixing, the samples were centrifuged at 137.9× g for 5 min in an Eppendorf 5415-D centrifuge (Eppendorf, Hamburg, Germany). The supernatant (300 µL) was transferred to a fresh tube, mixed with 300 µL of cold isopropanol, and centrifuged again at 137.9× g for 5 min. After discarding the supernatant, 500 µL of 70% cold ethanol was added to wash the pellet, followed by another centrifugation at 137.9× g for 5 min. The ethanol was removed and the pellet was dried at 40 °C for 30 min, then resuspended in 50 µL of 1 × TE buffer (10 mM Tris-HCl, 1 mM EDTA, pH 8.0). The DNA samples were treated with 1 µL of RNase (10 mg/mL; Thermo Fisher Scientific^®, Waltham, MA, USA) and incubated at 37 °C for 5 min. All chemicals were obtained from Sigma-Aldrich^® unless specified otherwise. DNA concentration and purity were determined by measuring the absorbance ratio (260/280 nm), and concentrations were adjusted to 100 ng. The quality of the extracted DNA was confirmed via electrophoresis on 0.8% agarose gels (Phytotechnology Laboratories^®, Lenexa, KS, USA) prepared with 1 × Tris/borate/EDTA (TBE) buffer.

ddRADseq was performed at Cinvestav (Centro de Investigación y de Estudios Avanzados del Instituto Politécnico Nacional) using the modified ddRADSeq essentially as described by Cooke et al. [39]. Briefly, sequencing adapters with Bgl II or Dde I 5′ overhangs were prepared by annealing single-stranded oligos (45 mM each) in 10 mM Tris-Cl pH 8.0 with 50 mM NaCl. Annealing was carried out in a thermal cycler by a stepwise reduction in temperature (30 s per cycle with 0.5 °C steps between cycles) from 97 °C to 12 °C. Genomic DNA (200 ng) was digested with 10 U BglII and 10 U DdeI in a 40 ul reaction mix for 4 h at 37 °C. The resulting DNA fragments were then purified using Agencourt AMPure XP beads (Beckman Coulter, Mexico City, Mexico) according to the manufacturer’s instructions. The quality and size of the DNA fragments was then evaluated using an Agilent bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA). An 11-fold molar excess of BglII sequencing adaptor, relative to the amount of BglII 5′ overhangs, and an 11-fold excess of DdeI sequencing adaptor were added to the samples and ligated with T4 DNA ligase in 40 ul for 2 h at room temperature. The ligation reaction was stopped by a 10 min incubation at 65 °C, followed by cleanup. To avoid PCR amplification of inserts flanked on both ends by DdeI adapters, the nick-containing ligation product was extended with ddTTP using the Klenow fragment of DNApol 3′–5′ exo and incubated for 30 min at 37 °C. The reactions were cleaned using Agencourt AMPure XP beads and amplified by 8 PCR cycles (initial denaturation 98 °C for 30 s; denaturation 98 °C for 10 s; annealing 62 °C for 30 s; extension 72 °C for 30 s; final extension 72 °C for 5 min). Samples were purified with Agencourt AMPure XP beads. Samples containing DNA in a size range of 300 to 400 bps was pooled in equal amounts according to the concentration of the PCR product as measured using a Qubit instrument (ThermoFisher Scientific, Waltham, MA, USA). Paired-end sequencing was performed using an illumina NovaSeq6000 (Illumina, Inc., San Diego, CA, USA).

2.3. Mutation Discovery and Analysis

Paired-end fastq data from 115 mutagenized samples, plus 5 non-mutagenized controls (2 from embryogenic calli and three from plants grown from seed), were mapped to the Coffea arabica L. genome (Cara_1.0, NCBI RefSeq GCF_003713225.1) using the Burrows–Wheeler Aligner with the mem command and the -M option to mark shorter split hits as secondary [40]. GATK version 4.2.5.0 was used to generate a multi-sample VCF [41]. The HaplotypeCaller tool was used to produce gVCFs, followed by the GATK tools GenomicsDBImport and GenotypeGVCFs to generate a multi-sample VCF, and bcftools view was used to filter for biallelic SNPs as previously described [42,43]. Quality filtering for variants was performed using GATK VariantFiltration with the following filters: QD < 2.0, QUAL < 30.0, SOR > 3.0, FS > 60.0, MQ < 40.0, MQRankSum < −12.5, and ReadPosRankSum < −8.0. The resulting VCF was then filtered to retain variants where at least 61 samples had a coverage of 20× or higher for that variant using the purpose-built program MVFF (Multi-Sample VCF Filtering, https://github.com/bjtill/MVFF-GUI, accessed on 27 July 2024)). Putative induced mutations were identified using the purpose-built tool Mutation-Finder-and-Annotator with the Relatives tool (MFAR, [43], and https://github.com/bjtill/Mutation-Finder-Annotator-with-Relatives-GUI, accessed on 27 July 2024)). The program works on the principle that chemically induced mutations accumulate randomly, and thus, unique variants are expected in every mutant line. Filtering of the VCF for 61 samples that have at least 20× coverage provided a minimum of 60 controls for every mutant line to remove natural variants segregating in the Catuaí genotype. The MFAR program identifies candidate induced mutations that are unique to a mutant line and not found in any of the other samples. The program annotates identified putative mutations using SnpEff and collects information on the types of nucleotide changes using bcftools stats.

2.4. Functional Analysis of Missense Changes and Evaluation of Allele Ratios

The ratio of reads supporting the alternative allele in the multi-sample VCF was collected using the purpose-built program VARP (VCF Allele Ratio Plotter, https://github.com/bjtill/VARP-GUI, accessed on 27 July 2024). Predictions on the effect of missense changes on protein function were performed using SIFT4G. To accomplish this, a SIFT database was created using the Cara_1.0 genome, the corresponding gff annotation file that was converted to gtf using gffread, and the UniRef90 protein database from UniProt by using SIFT4G_Create_Genomic_DB as described [44] (https://github.com/pauline-ng/SIFT4G_Create_Genomic_DB, accessed on 27 July 2024)). Variants were annotated using this database and the SIFT4G Annotator tool (https://github.com/pauline-ng/SIFT4G_Annotator, accessed on 27 July 2024)). Peptide sequences were collected from SIFT annotated loci and aligned to the reviewed Swiss-Prot sequences from UniProt using local BLAST [45,46]. The top alignment was retained and used to collect Gene Ontology (GO) information using the UniProt online tool Retrieve/ID mapping (https://www.uniprot.org/id-mapping, accessed on 27 July 2024). A Circos plot was created from the resulting data [47].

3. Results

3.1. Novel Single-Nucleotide Variants Recovered in All Samples

DNA from plants representing 115 mutant lines, plus 5 non-mutagenized controls, was subjected to ddRADseq (Figure 1). Chemical mutagenesis has been shown to induce novel mutations randomly with regard to the chromosome and position of altered bases, with some local biases being reported [48]. Thus, DNA sequence variants caused by chemical mutagens are expected to be different, in terms of genomic location, in different lines. The criteria that a DNA variant must be unique to a single mutant line for it to be considered a putative induced mutation has been used in many plant mutation reports. Unique single-nucleotide variants, compared to the reference genome, were recovered in all tested material (Figure 2 and Supplementary Table S2).

Excluding sample AH1CC3SS102_S102, which had fewer than 3000 mapped base pairs at 20× or higher coverage, the average number of bases per sample at or above 20× coverage in all samples was 4,061,983. The novel variant rate ranged between 1 mutation per 24,660 bps to 1 mutation per 83,429 bps. Heterozygous changes predominated, ranging from 81.6 to 100% of changes per sample. No clear trend was observed in treatment type or mutation rate. EMS treatment of plant seeds has previously been reported to cause primarily GC to AT transition changes. The percentage of GC to AT changes in the total variants ranged from 6.74% to 25.33% for rates of 1 variant per 149,856 bps to 1 variant per 692,923 bps (Supplementary Table S2 and Figure 3).

3.2. Ratio of Reads Supporting Reference and Alternative Alleles

Recovered novel variants were further evaluated for the ratio of reads supporting the reference and alternative alleles. While a perfect heterozygous call would have 50% of reads supporting the reference and 50% supporting the alternative allele, a range of allele ratios is typically observed in sequencing data. Deviations in allele ratios can be caused by stochastic events during the sequencing process or from changes in ploidy or errors during the mapping process. Allele ratios were calculated from the multi-sample coffee VCF prior to filtering for unique variants. A VCF from previously reported mutagenized diploid rice generated using similar methods was included for comparison (Figure 4) [22]. Homozygous calls supporting the alternative allele represent the minority of calls in coffee, whereas the proportion of homozygous alternative calls in rice having undergone 6 rounds of self-fertilization is similar to the number of homozygous calls supporting the reference allele. Heterozygous calls in coffee with an allele ratio between 40 and 60% (considered unambiguous heterozygous calls) represent approximately 50% of all potentially heterozygous variants. In rice, approximately 33% of variants fall into this category (Figure 4).

3.3. Predicted Effect of Novel Variants

Novel variants detected using the MFAR program are automatically annotated using the program SnpEff, which categorizes variants by impact and functional class. The number of high-impact novel variants ranged from 1 to 11 per sample, while the loss of translation start site ranged from 0 to 63 and the gain of a premature stop codon ranged from 0 to 55 per sample (Supplementary Table S3). The potential effect of missense changes on gene function was further evaluated using SIFT4g. In total, 2793 variants representing 959 unique gene IDs and 1882 unique transcript IDs were evaluated. This represents variants in the coding sequence (n = 1600) and variants located in the annotated 5′ and 3′ untranslated regions (UTRs) of the annotated genes. A total of 356 unique nonsynonymous variants were predicted to be deleterious to gene function, with 69 of these flagged as low confidence. Sequence alignments and gene ontology analysis of all SIFT annotations revealed hits to a range of proteins with associated GO terms (Supplementary Table S4). Annotations in loci harboring predicted deleterious changes included zinc finger-containing proteins, components of histone acetyltransferase, cyclin-dependent kinase 2-interacting protein (involved in DNA replication), and others.

4. Discussion

The overall goal of many plant mutation projects is to generate sufficient novel genetic variation to produce novel phenotypic variation. This can be in service of characterizing the function of genes, breeding new crop varieties, or both. Coffee mutants described here that were generated by the treatment of embryogenic calli using either sodium azide or EMS were previously phenotypically evaluated for variation in response to coffee leaf rust (CLR) using isolated coffee leaf discs [33]. While many functional genomics studies involving chemically induced mutations and the reverse-genetic method known as TILLING (Targeting Induced Local Lesions IN Genomes) have centered on abiotic stress resistance, biotic stress resistance has also been a focus of induced mutation and TILLING programs [21,49,50,51]. For example, exome capture was used to recover a genetic variant in the R gene Lr21 in a study aimed at recovering mutants affecting rust resistance in hexaploid wheat [52]. TILLING was also used to generate mlo-based powdery mildew resistance in hexaploid wheat, showing the efficacy of the approach in polyploids [53]. More commonly, induced mutations have been used in forward screens to uncover plants with enhanced resistance to diseases. For example, screens of EMS-mutagenized wheat revealed 16 lines with multi-year enhanced stripe rust resistance [54]. Physical irradiation, which can cause a range of genetic changes from single nucleotides to large indels, has been used for decades for forward genetic screens. The FAO/IAEA Mutant Variety Database contains 310 officially released mutant varieties with the attribute of resistance to biotic stresses that are generated using gamma or X-rays, representing more than 40 different plant species. These include crops with resistance to leaf, stripe, and other rusts, showing the efficacy of induced mutation approaches for disease resistance. The majority of mutant varieties are self-fertile plants with a short seed-to-seed time. Over half of the officially released mutant varieties annotated as having improved disease resistance are represented by four crops: rice, wheat, soybean, and barley (https://mvd.iaea.org/, accessed on 27 July 2024).

The mutant variety database showcases over 3400 officially released mutant varieties, yet no coffee mutant varieties were found regardless of phenotypic trait. Indeed, in terms of mutant varieties and TILLING populations, tree species are not well represented. While tree species are less important to global food security and are not considered model plants for basic biological research, therefore receiving less research funding, additional points can be considered. Compared to species like rice or barley, tree species are more recalcitrant with respect to breeding and the development of mutant populations. There are several factors that may contribute to this, including the space requirements for developing large mutant populations and longer generation times. For example, the first description of TILLING involved chemical mutagenesis of thousands of Arabidopsis thaliana seeds that was carried out in a single 500 mL beaker. Prior to mutagenesis, plants were subject to three rounds of backcrossing to the Columbia ecotype to generate a nearly isogenic population of thousands of seeds to be mutagenized [55]. This was intended to ensure that the genetic background of all seeds was homozygous and homogeneous such that any novel genetic variant recovered could be attributed to the mutagenic process. This was logistically possible because of the high fecundity of Arabidopsis and the relatively fast seed-to-seed time of approximately 8 weeks. By comparison, coffee is much more challenging. The seed-to-seed time is approximately 20-fold slower, and seed set is much lower, making the development of an isogenic coffee seed stock for mutagenesis impractical [12]. Indeed, more than 6 years were required to make the seed-mutagenized M₂ population described here. An attractive alternative is cell culture, where mutagenesis of embryogenic calli allows for the regeneration of plants from a single cell [30]. This obviates the need for meiotic propagation because regeneration from a single cell should avoid genetic mosaicism. Furthermore, the time from mutagenesis to the development of sufficient plant tissue for genotyping is reduced from years to months.

An ideal plant mutant population can be considered one whereby the number of induced sequence variants per line is as high as possible without adversely affecting fertility or fecundity. This approach maximizes the number of induced mutations and, potentially, phenotypes, which results in minimizing the cost and effort of genotyping. Some sterility is expected, and thus, a balance between mutation load and plant fitness is sought. Mutational load has been shown to be correlated with ploidy. Diploids have been reported with mutation frequencies as high as one induced mutation per approximately 100 kilobases. Polyploids can accumulate more mutations before the onset of lethality owing to genetic redundancy. Triploid banana mutagenized with EMS was reported to have a frequency of 1 variant per 57 kb, while tetraploids have been reported to have frequencies in the range of 1 per 40 kb, and hexaploids up to 1 per 24 kb [25]. This provides a framework for evaluating different crops. Another useful metric is the spectrum of nucleotide changes. In many species, EMS has been reported to cause >90% G:C to A:T transition changes. Fewer studies have been reported using sodium azide as the sole mutagenic treatment [21]. In barley, one TILLING study revealed 22 novel variants, with 21 being G:C to A:T changes, while another barley study reported 4 of 5 sequence-validated mutants as G to A transition changes, supporting earlier work stating that the majority of base substitutions caused by sodium azide are transition mutations [56,57,58]. Additional metrics that can be used to evaluate a mutant population are the zygosity of recovered variants and the acceptable minor allele frequency for a variant to be considered a putative induced mutation. For zygosity, 100% of all induced mutations identified in lines from mutagenizing embryogenic calli are expected to be heterozygous because no meiotic propagation was carried out prior to tissue collection and DNA sequencing. In seed-mutagenized M₂ material, 2/3 of variants are expected to be heterozygous, while the remaining third are expected to be homozygous. With regards to minor allele frequency, the most stringent approach is to accept variants that are never found in more than one mutant line.

When considering all recovered sequence variants that are unique to a single line, the coffee mutants, after removal of sample AH1CC3SS102, which was deemed to be a failure, showed a range of between one mutation per 24,660 bps and one change per 83,429 bps, within the range of expectation for a tetraploid. However, the non-mutagenized controls harbored between one novel variant per 26,413 bps and one per 72,247, which is within the range of that detected in mutagenized material. When considering only GC to AT transition changes, the number of putative induced variants per bp dropped to between 1/150 kb and 1/404 kb, with the controls between 1/200 kb and 1/365 kb. No trend was observed with regard to mutation frequency and mutagen or treatment material. Likewise, regardless of treatment material, heterozygous variants predominated, with fewer than 20% homozygous novel variants being recovered per line regardless of treatment type (Supplementary Table S2). All novel variants reported are unique to a single line compared to a minimum of 60 other samples. This represents a minor allele frequency of 1.7% or less at the population level. In contrast, studies evaluating natural variations in coffee have used a minor allele frequency filter of less than 5% to 10% as a cut-off for candidate alleles [59]. However, the presence of low-frequency natural alleles has not been directly tested, and previous studies using a minor allele frequency filtration may have removed bona fide natural variants. Novel variants in the current study were selected with the criteria of having a minimum coverage of at least 20× and mapping quality of greater than 40. Errors associated with RADseq data include the under-representation of heterozygous calls, yet variant calls in the current data are mostly heterozygous. Taken together, the novel variants reported here cannot be unambiguously assigned as having been caused by the mutagenic treatment. Tissue culture has been shown to be mutagenic, which may explain the variation in both in vitro control and chemically mutagenized calli, but does not explain novel variant accumulation in seed-mutagenized material. Potential issues associated with next-generation sequencing, mapping, and variant calling in polyploids can also be considered [60,61,62]. In addition to applying a minimum mapping quality filter of 40, the evaluation of allele ratios did not reveal evidence of skewed ratios that would be indicative of reads mapping to more than one genome.

While further investigations are required to clarify the source of the recovered genetic variation, functional analysis showed the accumulation of mutations affecting start and stop codons and more than 300 missense mutations predicted to be deleterious by SIFT. In addition, as reduced representation genome sequencing samples only a small fraction of the genome and its component genes, it is expected that the mutant populations tested harbor many yet to be discovered gene lesions that could impact protein function. In the current analysis, approximately 4 Mbp was evaluated per sample at 20× or higher coverage, representing less than half a percent of the coffee genome. Yet, analysis revealed predicted deleterious lesions in genes potentially involved in gene regulation. It is thus expected that many more such variants exist in mutant lines. This likely explains the previously reported phenotypic variations, including differential responses to CLR [33]. Future efforts, including whole-genome sequencing of mutant lines, will allow for a more comprehensive understanding of the genetic variations responsible for the observed variations in the response to disease pressures.

5. Conclusions

This work highlights the challenges of establishing mutant populations in what can be considered recalcitrant species that lack the advantages of quickly propagating cereals. The development of additional coffee populations using different mutagen concentrations will be required in order to establish meaningful expectations for a successful mutagenic treatment in Coffea arabica L. The current study utilized reduced-representation genome sequencing to sample a subset of the genome for putative induced mutations. This has the advantage of allowing for the evaluation of more plants for the same throughput of sequencing when compared to a whole-genome sequencing approach. Yet, for species like coffee, reducing population sizes while conducting experiments to establish a suitable mutagenic treatment prior to scaling up the population size may be advantageous. Therefore, sequencing fewer plants, but at a whole-genome level, can be considered an alternative approach. For example, it has been described that, in a hypothetical species with a genome size of 1 Gb and a sequencing output 100 Gb, the same number of unique bases can be evaluated at 20x coverage when sequencing 5 individuals completely or 50 individuals using an exome capture approach that samples 100 Mbp of sequence per individual [19]. It is important, however, to consider that the accumulation of induced mutations can vary from plant to plant from the same mutant population, and thus, sampling too few individuals can lead to a potential over- or underestimation of mutation frequency. Determining the optimal number of plants to sample in this approach requires further experiments. In addition to resource and space considerations, the time to develop a mutant population must also be considered. Given the long seed-to-seed time in coffee, we propose treatment of embryogenic calli as the best approach because DNA can be collected and sequenced prior to meiotic propagation. Combining whole-genome sequencing of few individuals with mutagenesis of calli will provide the most efficient way to evaluate multiple mutagenic treatments. Further, the bioinformatic tools described in this study will allow for efficient and consistent recovery of novel variations in order to make meaningful comparisons between different mutagenic treatments with a variety and also between different plant species. The mutation frequencies achieved in one tree species, for example, can be useful as a guide for determining whether mutagenesis was successful when evaluating a different tree species of the same ploidy. The development of populations harboring a high density of novel variants will help to address the narrow genetic base found in many economically important crops.