Safeguarding Our Genetic Resources with Libraries of Doubled-Haploid Lines

Abstract

Thousands of landraces are stored in seed banks as “gold reserves” for future use in plant breeding. In many crops, their utilization is hampered because they represent heterogeneous populations of heterozygous genotypes, which harbor a high genetic load. We show, with high-density genotyping in five landraces of maize, that libraries of doubled-haploid (DH) lines capture the allelic diversity of genetic resources in an unbiased way. By comparing allelic differentiation between heterozygous plants from the original landraces and 266 derived DH lines, we find conclusive evidence that, in the DH production process, sampling of alleles is random across the entire allele frequency spectrum, and purging of landraces from their genetic load does not act on specific genomic regions. Based on overall process efficiency, we show that generating DH lines is feasible for genetic material that has never been selected for inbreeding tolerance. We conclude that libraries of DH lines will make genetic resources accessible to crop improvement by linking molecular inventories of seed banks with meaningful phenotypes.

GENETIC resources of domesticated species will play a key role in feeding future generations (McCouch et al. 2013). Impressive examples exist where introgressions of favorable alleles from genetic resources into elite material have provided resistance to devastating diseases (Plucknett and Smith 1987), or allowed revolutionary changes in management practices (Khush 2001). Despite these success stories, examples where genetic resources have been used successfully to broaden and significantly improve the genetic base of modern breeding germplasm for quantitative traits are scarce, and molecular evidence suggests that most genetic diversity has remained idle in seed banks (Hoisington et al. 1999). Several reasons can be held accountable for this: first, the gap in performance between genetic resources and elite breeding material for agronomic traits widens steadily with continuous selection progress. Second, passport data or genomic information collected by seed banks can predict an individual’s membership to a specific race or accession (Yu et al. 2016), but has limited predictive power with respect to breeding values of individuals. Third, unless a species is strictly autogamous, accessions in gene banks represent collections of heterogeneous and heterozygous individuals, so that phenotypic evaluation in replicated multi-environment field trials is only possible for the entire population, but not for individual genotypes. Fourth, maintaining allogamous species in gene banks is technically complex and resource demanding, and many reports have documented severe inbreeding, loss of diversity, admixture, or mislabeling of accessions (Chebotar et al. 2003; Bergelson et al. 2016). Thus, the value of gene bank accessions can be capitalized upon only if efficient strategies for proper maintenance and utilization of their genetic diversity can be devised.

Some of the difficulties mentioned above can be alleviated by developing inbred lines from the original populations, to generate reproducible genetic units representing the diversity of the source material. However, for many species, the development of inbred lines by recurrent selfing is either not possible or extremely cumbersome due to self-incompatibility or high genetic load (Hallauer et al. 2010), and the selective forces operating during the inbreeding process are largely unknown. An alternative approach to tap the genetic diversity of genetic resources is gamete capture (Stadler 1944), where elite genetic material is pollinated with a random sample of gametes from the heterogeneous source material, with subsequent recurrent selfing of the offspring. Depending on the evaluation process of the offspring, and the performance gap between the parental germplasm, strong selection in favor of alleles contributed by the elite parent is most probable, thus limiting the usefulness of the gamete capture approach with respect to introducing novel genetic diversity (Sood et al. 2014).

A method to circumvent the recurrent selfing process is the in vitro or in vivo recovery of haploid gametes from a source population, with subsequent chromosome doubling to produce fully homozygous doubled-haploid (DH) lines. In some crop species, DH technology is already highly advanced, and has largely replaced recurrent selfing in elite germplasm improvement programs (Dwivedi et al. 2015). Thus, applying DH technology to create libraries of “immortalized” genetic units from genetic resources seems an obvious next step to make them amenable to crop improvement. This applies in particular to genetic material that has already undergone moderate anthropogenic selection, such as the landrace populations that are the ancestors of our elite breeding material.

To maintain the diversity of landraces with libraries of DH lines, they need to represent random gametic arrays of the source populations, and expenditures for their development must not be prohibitive. In this study, we address these topics using maize (Zea mays L.) landraces as a model system for asking the following questions: (i) is the allelic inventory and linkage disequilibrium (LD) of landraces reflected in DH libraries derived from them? (ii) What is the efficiency of DH production from landraces in comparison to elite breeding material? (iii) During which phase of the DH production is the loss of gametes, and, presumably, the genetic load most pronounced? (iv) Does purging of the landraces from their genetic load via the production of DH lines act on specific genomic regions? (v) How many DH lines are required to capture a given proportion of the genetic variation in a landrace?

Materials and Methods

Genetic material

Seven European maize landraces were chosen for this study: Bugard (BU) from France, Gelber Badischer (GB), Schindelmeiser (SC) and Strenzfelder (SF) from Germany, Rheinthaler (RT) and Walliser (WA) from Switzerland, and Satu Mare (SM) from Romania. The landraces were selected from a larger set of 70 European flint maize landraces on the basis of (i) their adaptation to the agroclimatic conditions for maize cultivation in Central Europe, and/or (ii) evaluation of their testcross performance with two testers from the dent heterotic pool (Böhm et al. 2014). Landraces were screened in two seasons for expression of the R1-nj embryo marker in induction crosses with haploid inducer UH400 to reduce misclassification in the identification of haploid seeds.

Production of DH lines

The landraces, together with nine intrapool single crosses between elite lines from the flint heterotic pool of the maize breeding program of the University of Hohenheim, were used as source material for production of DH lines by the in vivo haploid induction method (Prigge and Melchinger 2012). The efficiency of DH line production was evaluated in five landraces (GB, RT, SF, SM, and WA), and compared with the elite flint crosses. To accomplish this, the process of DH production was subdivided into eight steps, as detailed in Supplemental Material, Figure S1 in File S1 and Supplemental Notes in File S1. In each of the eight steps, we recorded the number $N_i$ of units (seeds, seedlings, plants, D₀ plants, D₁ ears with seed set, and propagated DH lines) for each induction cross. The success rate for each working step i was determined by calculating the ratio ${\text{SR}}_i= \left(N_{i+1}/N_i\right).$ From these ratios, we calculated the haploid induction rate (HIR) = proportion of haploid seeds in the total number of seeds harvested in induction crosses, the survival rate (SR) = proportion of haploid seeds that germinated and survived until flowering, the reproduction rate (RR) = proportion of DH plants from which D₁ lines could be produced, the overall success rate (OSR) = proportion of DH lines produced from the identified haploid seeds (which corresponds to the product SR × RR), and the total production costs per DH line for each material group (Supplemental Notes in File S1).

Genotyping

From five landraces (BU, GB, RT, SC, and SF), leaf samples were collected from (i) bulks of two to six seedlings from each of the DH lines developed from these landraces (36 ≤ NDH ≤ 69, Figure S2 in File S1), and (ii) 23 individual seedlings of randomly chosen S₀ plants per landrace grown from the same seed lots used for induction crosses. Genomic DNA of these samples was extracted using a modified CTAB protocol (Murray and Thompson 1980). The DH lines were genotyped with the Illumina MaizeSNP50 BeadChip (Ganal et al. 2011), the S₀ individuals with the 600k Affymetrix Axiom Maize Genotyping Array (Unterseer et al. 2014). After quality control carried out for both SNP data sets with the R package GenABEL (Aulchenko et al. 2007), a total of 266 DH lines genotyped for 57,840 SNPs, and 114 S₀ plants genotyped for 616,201 SNPs remained (one individual from BU was deleted). In the rare event of missing data, imputation was carried out with software Beagle version 3.3.2 (Browning and Browning 2007). Since coding of SNP alleles may differ between the Illumina and the Affymetrix platform, coding of the set of 45,655 SNPs common to both arrays was translated from the Affymetrix to the Illumina coding scheme (see Supplemental Notes in File S1). In total, 28,133 SNPs fulfilled all quality criteria, and were polymorphic across the entire set of 380 genotypes, and used in all subsequent analyses.

Statistical tests for differences in success rates of DH production

Success rates among the five landraces and the elite germplasm for a given working step in the DH procedure were tested for significant differences with the G-test of independence (Sokal and Rohlf 1969) using the R-package “DescTools” (Signorell et al. 2016). To account for multiple comparisons, P-values were adjusted with the Bonferroni correction.

Molecular diversity analysis

Analyses of molecular diversity were conducted with SNP data from the DH and S₀ generations separately. Departure from Hardy-Weinberg equilibrium in the sample of S₀ plants from each landrace was tested for each polymorphic SNP. P-values were derived from Fisher’s (1934) exact test using the R-package “genetics” (Warnes 2013). In addition, we calculated the gene diversity statistic $H_s$ (Nei 1973) for each SNP. For graphical representation, results were summarized for windows of 10 Mb width, sliding in steps of 0.5 Mb. For a given window, $H_s$ or P-values were averaged across all SNPs, and plotted at the center of the window. An analysis of molecular variance (AMOVA) was performed using Arlequin V.3.5 (Excoffier and Lischer 2010). The total molecular variance among entries was partitioned between and within landraces. The phylogenetic structure of the five landraces was depicted separately for the DH lines and the S₀ plants by constructing neighbor-joining trees based on Rogers’ distances (Rogers 1972) using the R package “ape” (Paradis et al. 2004), and visualized with the software FigTree (“http://tree.bio.ed.ac.uk/”).

Statistical tests for genetic differentiation of the DH and S₀ generation

For each landrace, the genetic differentiation of the DH from the S₀ generation was examined by the F_ST statistic (Holsinger and Weir 2009). F_ST was calculated per SNP, and averaged (i) in windows of 10 Mb width, sliding in steps of 0.5 Mb, and (ii) across the entire genome. P-values for testing whether an observed F_ST value was significantly different from the null distribution, indicating genetic differentiation between the DH and S₀ generation, were determined with a permutation test. To generate the distribution of the test statistic under the null hypothesis, we (i) generated “pseudoS₀” genotypes (PS₀) from pairs of homozygous DH lines (see Supplemental Notes in File S1), and (ii) calculated F_ST values for 10,000 random subdivisions of combined sets of S₀ and PS₀ genotypes. Significantly different F_ST values were determined at a genome-wise type I error rate of 5% with a Bonferroni correction for multiple testing based on the number of nonoverlapping windows (200). For graphical representation, F_ST values were averaged across all SNPs in a given window, and plotted at the center of the window.

LD in the DH and S₀ generation

LD estimates (r²) between pairs of markers on the same chromosome were computed following Hill and Robertson (1968). LD values were calculated separately for the DH and S₀ generation from each landrace using the R package “synbreed” (Wimmer et al. 2012). For the three landraces with N_DH > 46, calculation of r² values was based on randomly sampled subsets of n_DH = 46 to keep the number of sampled gametes comparable to those in the S₀ generation. Sampling was repeated 100 times, and results were averaged for each landrace over chromosomes and repetitions. LD decay with increasing physical distance among markers was compared for the DH and S₀ generation by grouping the respective r² values according to the physical distance of marker pairs in steps of 0.05 Mb, and averaging over all 10 chromosomes following Strigens et al. (2013a).

Number of DH lines required for adequate sampling of landraces

We investigated the molecular diversity of the source population captured by the DH lines as a function of sample size (n_DH) based on $H_s.$ First, we calculated $H_s$ (averaged over all 28,133 common SNPs) in the set of 23 (22 for BU) S₀ individuals [corresponding to 46 (44) gametes], which served as reference for the genetic diversity present in each landrace. Second, we sampled with replacement n_DH DH lines from each landrace, starting with n_DH = 4, and increasing n_DH in steps of two up to the total number of DH lines available for the landrace. For each of these samples, $H_s$ was calculated with a total of 250 repetitions. Finally, we calculated, for the different samples of DH lines, the ratio $H_s\left(\text{DH}\right):$ $H_s\left({\text{S}}_0\right)$, and, for a given sample size, averaged the results over all repetitions.

Data availability

Genotypic data of S₀- and DH-imputed genotypes are available in the supplemental files “snp_S0.txt” and “snp_DH.txt,” respectively, in File S1. Marker positions referring to the physical map (B73 RefGen_v2) are described in the supplemental file “marker_map.txt” in File S1.

Results

Efficiency of DH library production from maize landraces

Totals of 383,000 and 35,327 seeds were harvested from induction crosses of landraces and elite crosses, resulting in totals of 388 and 137 DH lines, respectively (Table 1). Mean haploid induction rate (HIR) differed significantly (P < 0.05) between landraces and elite material (Table 1 and Table S1 in File S1), but induction rates in two landraces (SF and WA) were significantly higher even than in the elite material. Differences in HIR between landraces and elite material were small compared to differences in OSR of DH production. OSR was 2.5× higher in the elite material compared to the landraces. While HIR differed significantly, OSR was stable across landraces. Subdividing OSR into survival rate during the juvenile phase (SR) and reproduction rate (RR) (Table 1 and Table S1 in File S1) yielded comparable SR for landraces (67.8%) and elite germplasm (75.4%). On the other hand, RR was more than twice as large in elite germplasm (15.3%) than in the landraces (6.0%). The reduction in RR in the landraces was due mainly to failed or aborted seed set in D₀ plants and D₁ lines (N₇/N₆ and N₈/N₇ in Table S1 in File S1).

Table 1. Number of seeds ($N_S$) harvested from induction crosses and number ($N_{D_1}$) of DH lines (D₁ generation) for five European maize landraces and elite crosses from the flint germplasm pool. HIR, SR, RR, and OSR in development of DH lines were calculated as described in Supplemental Notes in File S1.

Source Germplasm	$N_S$	$N_{D_1}$	HIR	SR	RR	OSR
			%†
Landraces (LR)
Gelber Badischer(GB)	113,596	59	1.24f	64.74c	5.81b	3.76c
Rheinthaler (RT)	44,557	43	1.86e	64.79c	7.33b	4.75c
Strenzfelder (SF)	41,779	61	2.81a	71.47b	6.76b	4.83c
Satu Mare (SM)	114,712	108	1.96d	69.54bc	6.50b	4.52c
Walliser (WA)	68,356	117	2.83a	68.26bc	7.31b	4.99c
Sum/mean	383,000	388	2.14*	67.76**	6.74**	4.57**
Elite crosses (EC)	35,327	137	2.54c	75.35a	15.33a	11.55a

^†Values followed by the same letter are not significantly different at Bonferroni corrected P < 0.05.

^*,**Mean of the landraces and elite crosses materials differed at the 0.05 and 0.01 probability level, respectively.

Molecular diversity and LD in DH and S₀ generations

The five landraces for which the S₀ generation was sampled (BU, SC, GB, RT, and SF) showed no significant deviations from Hardy-Weinberg equilibrium (Figure S3 in File S1). Analyses of molecular diversity gave very similar results for the DH and the S₀ generation. The phylogenetic relationship of the five landraces visualized with neighbor-joining trees showed a distinct grouping of the landraces, and was almost identical for the DH and S₀ generation (Figure 1). Molecular variance components obtained from the AMOVA were comparable for the two generations (Table 2), when taking into account that S₀ plants and DH lines are expected to differ by a factor of two (Hallauer et al. 2010), and revealed that most of the molecular variance was found within landraces, and not between them. Gene diversity H_s corresponded well between the two generations across the entire genome (Figure S2 in File S1).

Figure 1.

(A, B) Neighbor-joining tree constructed from Rogers’ distances for five European maize landraces (BU, GB, RT, SC, and SF) based on 28,133 polymorphic SNPs.

Table 2. Estimates of variance components obtained from the analysis of molecular variance of the original landraces (S₀ generation), and populations of DH lines (D₁ generation) derived from them for five European maize landraces (BU, GB, RT, SC, and SF) calculated on the basis of 28,133 SNPs.

Source	S0 Plants		DH Lines
	Dfa	VCb	Dfa	VCb
Among landraces	4	1144c	4	869c
Within landraces	109	1402	261	2916
Within BU	21	1698	35	3544
Within GB	22	1416	58	3008
Within RT	22	1022	43	2611
Within SC	22	1361	57	2642
Within SF	22	1525	68	2938
Total	114	2547	265	3785

^aDegrees of freedom.

^bVariance component.

^cSignificant at P < 0.01.

Allele frequency distributions were similar for the two generations for all landraces (Figure S4 in File S1). F_ST values for genome-wide subpopulation differentiation between the DH and S₀ generation of each landrace are given in Table 3. For all landraces, F_ST values differentiating the S₀ and D₁ generation were significantly (P < 0.01) different from the null distribution generated in the permutation test. However, the average F_ST of 0.019 was small in relation to the F_ST statistics obtained from the pairwise comparison of landraces in the S₀ generation with an average F_ST of 0.110. The F_ST values for the two generations along the ten maize chromosomes are given in Figure 2A. F_ST values never surpassed 0.05 in landrace SF, and in only one of the 200 nonoverlapping windows in landraces SC and GB. In landraces BU and RT, F_ST statistics ranged between 0.05 and 0.10 more often, but simultaneously surpassed a value of 0.05 in only five of the 200 nonoverlapping bins. Likewise, P-values for the F_ST statistic determined by the permutation test were significant in several windows (Figure 2B), but never in all landraces simultaneously, and, apart from windows on chromosomes 1, 3, 7, and 9, in less than four landraces.

Table 3. F_ST statistics for pairwise comparison of the original landraces (S₀ generation) and derived DH lines (D₁ generation) for five European maize landraces (BU, GB, RT, SC, and SF). F_ST values for the S₀ generation are shown above the diagonal, and those for the D₁ generation below the diagonal. F_ST statistics for comparison of the S₀ and D₁ generation of each landrace are given on the diagonal. All values were calculated on the basis of 28,133 SNPs and significantly (P < 0.01) greater than zero.

Landrace	BU	GB	RT	SC	SF
Bugard (BU)	2.4	16.5	19.0	16.1	15.0
Gelber Badischer (GB)	14.8	1.5	9.5	7.5	7.1
Rheinthaler (RT)	16.3	6.0	3.0	8.4	8.1
Schindelmeiser (SC)	16.1	7.0	6.8	1.7	2.7
Strenzfelder (SF)	14.4	6.4	6.0	3.2	1.0

Values were multiplied by 10²

Figure 2.

(A) F_ST statistic for evaluating genetic differentiation between the original landrace (S₀ generation), and the population of DH lines (D₁ generation) derived from it, averaged across all markers in a sliding window of 10 Mb width along the chromosomes for five European maize landraces (BU, GB, RT, SC, and SF). (B) Bins with significant F_ST statistics based on a permutation test (see Materials and Methods and Supplemental Notes in File S1). The heat map at the bottom, calculated based on the 28,133 SNPs analyzed, indicates the marker density within the windows (Mb⁻¹). Centromeres are indicated by gray vertical lines.

Curves for LD decay as a function of increasing physical distance between marker pairs were almost identical for the DH and S₀ generations (Figure 3). The r² values decayed rapidly and approached 0.10 at a distance of 3.0 Mb. The only exception was the S₀ generation of landrace RT, where LD remained at a higher level than in the other landraces. RT also showed the highest genome-wide and window based F_ST values for the comparison of the two generations, with strong differentiation across the entire genome (Figure 2 and Table 3).

Figure 3.

(A, B) LD in each generation vs. the physical distance between linked markers for five European maize landraces (BU, GB, RT, SC, and SF) calculated based on 28,133 polymorphic SNPs. The r² values for marker pairs were binned according to their physical distance, each bin corresponding to an interval of 0.05 Mb width.

The gene diversity H_s captured by random samples of DH lines of different size relative to the S₀ generation is given in Figure 4. For landraces SC and SF, >30 DH lines were needed to capture 95% of the gene diversity present in the sample of S₀ plants, whereas only ∼15 DH lines were needed for landraces BU and GB. For RT, the ratio H_s(DH): H_s(S₀) rapidly exceeded 1.0, suggesting that gene diversity of the sample of S₀ plants of RT was rather limited in comparison with the DH lines extracted from this landrace, as can also be seen from Figure S4 in File S1.

Figure 4.

Gene diversity (H_S) in random samples of n_DH DH lines (D₁ generation) relative to H_S in the original landrace (S₀ generation) for five European flint maize landraces (BU, GB, RT, SC, and SF) determined for 28,133 SNPs.

Discussion

DH libraries represent the allelic inventory of landraces

Using high-density genotyping in five landraces of maize originating from different geographic regions of Central Europe, we showed that libraries of DH lines capture the molecular diversity of genetic resources in an unbiased way. Both the allelic inventory and LD of the DH and S₀ generation were highly congruent for four of the five landraces. Genome-wide F_ST values between pairs of landraces varied, and generally exceeded the F_ST values for the differentiation of the DH and S₀ generation substantially (Figure 2A and Table 3 ). One pair of landraces (SF and SC) showed only mild differentiation (F_ST = 2.7), which was expected because, according to oral tradition by breeders, SF was derived from other German landraces, including SC. One landrace (RT) showed a slightly elevated differentiation between the DH and S₀ generation based on F_ST values and LD decay distance, but this was most likely the result of inappropriate sampling of S₀ plants rather than the effect of DH production or intrinsic genetic load, as the S₀ plants from RT showed a significantly reduced molecular variance and increased LD compared to the DH sample and the other landraces (Figure 3 and Table 2). We hypothesize that, had the DH lines been derived directly from the sampled S₀ plants, differentiation between the DH and S₀ generation would have been even less pronounced.

In a study investigating the mechanism underlying in vivo haploid induction in maize, Zhao et al. (2013) detected, in one out of 42 haploids, a 44-Mb heterozygous fragment from the inducer parent. Contrary to this observation, we found no evidence for presence of inducer genome in the 266 DH lines analyzed in our study, when comparing their DNA fingerprints to the genotyping results of inducer UH400 obtained with the Illumina MaizeSNP50 BeadChip (Hu et al. 2016). Likewise, Han et al. (2016) detected no traces of the inducer genome in a study with 633 DH lines derived from elite source germplasm. Thus, we hypothesize that either genome fragments present in the haploid stage are eliminated during later phases in the development of DHs, or transfer of the inducer genome occurs at such low frequency that it is of no practical relevance.

No evidence for targeted selection caused by genetic load

In elite germplasm improvement programs, DH technology has been applied at large scale, and no evidence for selection of specific genomic regions has been observed (e.g., Bauer et al. 2013). Thus, it is mainly the genetic load intrinsic to the landraces that might hamper their representation through the use of DH libraries. Two scenarios may arise: (i) the genetic load is due to recessive alleles that appear at low frequency, but have large detrimental effects and/or (ii) the genetic load is the cumulative result of many detrimental alleles with small effects (Willis 1999; Mezmouk and Ross-Ibarra 2014). In this study, mostly random and/or small allele frequency differences were observed for the DH and S₀ generation (Figure S4 in File S1). For all five landraces, the comparison of allele frequencies was consistent across the entire frequency spectrum, and obtaining fully homozygous DH lines in one generation gave no evidence for systematic directional selection in specific genomic regions. With the given sample sizes and marker density, it cannot be excluded that selection against detrimental alleles with very low frequency, and/or small effects, might have remained undetected in the comparison of the DH and S₀ generation. Nevertheless, the loss of favorable or neutral alleles through partial purging of the genetic load during the DH production process seems rather unlikely. Directional selection at one locus changes allele frequencies at a second, neutral, locus proportional to their LD. Averaged across the genome, LD was small (r² ≤ 0.1) for pairs of marker loci spanning a physical distance of 3 Mb or more. Even in peri-centromeric regions with low recombination rate, an over-representation of allelic differences between the DH and S₀ generation was not observed, as might have been expected from a recent study on maize, where an association of recombination frequency and deleterious polymorphisms was found (Rodgers-Melnick et al. 2015).

Our results on the molecular diversity of landraces, and their derived DH libraries, are in agreement with a study on phenotypic diversity of DH lines derived from three of the populations studied here (BU, GB, and SC) (Strigens et al. 2013b). When DH lines were intermated to resynthesize their respective original landrace, the original, and the resynthesized, landrace showed significant differences in agronomic trait performance only in rare cases. Combined with the results from this study, it seems justified to assume that the genetic load in maize landraces is, to a large extent, the result of the cumulative action of many genes with small effects distributed across the genome, and, given the low LD prevalent in the landraces, that it does not interfere with the unbiased sampling of diversity in the form of DH libraries.

Genetic load in landraces affects fertility traits

Assuming that differences in DH production efficiency between flint elite germplasm and landraces are mainly the result of genetic load, the rigorous analysis of the DH production process from haploid seeds to fertile DH lines yielded insights during the developmental phase of which the loss of genotypes, and, presumably, genetic load in the landraces, was most pronounced. While haploid induction rate and survival during the juvenile phase did not differ substantially between landraces and elite germplasm, fertility traits relevant for the last steps of DH production were significantly reduced and variable across landraces (Table 1). It is those last steps of line establishment and propagation that are common to all DH production systems, irrespective of the method employed. Consequently, interpolation of our results should be possible to other allogamous crops such as sugar beet, rape seed, and a number of vegetable species (Murovec and Bohanec 2012; Dwivedi et al. 2015). Moreover, development of efficient in vivo haploid induction systems in other cereals are within reach by the recent cloning of MATRILINEAL, a gene encoding a pollen-specific phospholipase, that triggers haploid induction in maize (Gilles et al. 2017; Kelliher et al. 2017; Liu et al. 2017).

Optimal representation of the intrinsic diversity of a landrace

The generation of a DH library is a onetime investment providing immortal, ready-to-use, genetic units that represent the diversity of the source material, and recover ancestral recombination. DH lines derived from landraces can be used readily for establishing prebreeding programs bridging nonelite and elite germplasm, as proposed by Gorjanc et al. (2016). Generating DH libraries from landraces capitalizes upon the fact that, in allogamous species, a large proportion of the genetic variation lies within landraces, and not between them (Sood et al. 2014). Depending on the genetic material studied, up to 80% of the molecular and phenotypic variation can lie within landraces, while only 20% differentiate the respective populations (Sanchez et al. 2000). The AMOVA of the DH and S₀ generations of the five landraces in this study corroborated these findings (Table 2). Thus, for improvement of quantitative traits, it seems more appropriate to sample the diversity of a few preselected landraces comprehensively, rather than sampling few individuals from each of a large number of landraces. Under the naïve assumption that each SNP corresponds to a QTL with equal additive effects, and absence of dominance effects, as applies to DH lines, gene diversity H_s should be proportional to the additive genetic variance of the population of S₀ plants, as well as DH lines derived from it (Falconer and Mackay 1996). Whether H_s can be used as a proxy to choose landraces, that are expected to release the largest amount of genetic variation in the DH library produced from them, warrants further investigation. Nevertheless, this information can be used to determine the number of DH lines that warrant adequate sampling of the genetic variation in the original landrace. From the results obtained in this study (Figure 4), it seems safe to conclude that equivalent numbers of independent gametes sampled in the DH or S₀ generation lead to comparable representation of the genetic diversity of the original landrace.

DH libraries—a tool for maintaining genetic resources

Expenditures for the generation of a representative DH library vary among crops and source populations. When landraces are maintained as DH lines rather than in the form of heterozygous populations, additional expenses are incurred through the DH production process, and increased space requirements in seed banks for storing individual lines. Based on the efficiency of the DH production process in maize, the cost for a representative DH library of 40 lines was estimated to be <US $ 3000 per landrace (Table S2 in File S1). These costs seem small considering the large investments currently spent to establish molecular inventories of seed bank accessions. Costs for storage of DH libraries will exceed those of heterozygous populations, but not necessarily at a linear rate, because fewer seeds would be required to preserve the genetic integrity of the material. Multiplication costs will be higher for heterozygous populations because far >100 full-sib families need to be generated through controlled crossings in each round of multiplication to avoid loss of alleles due to drift (Falconer and Mackay 1996). With DH lines, the multiplication process is simplified requiring the self-pollination of ∼10 plants per line. Taking the expenditure of the currently practiced multiplication process in seed banks as a base line (Taba et al. 2004), ∼50 DH lines per landrace can be maintained and multiplied without increase in costs. Our data show that already 40 DH lines represent the genetic diversity of a landrace to a large extent, and, in comparison to the current maintenance practice, there is no risk of changes in allele frequencies due to unintentional selection, assortative mating, or drift. Linking molecular inventories to meaningful phenotypes is the most challenging and complex task when mining biodiversity for useful alleles (McCouch et al. 2013). Since the genotypes of DH lines are genetically fixed, they can be reproduced ad libitum, which allows their immediate evaluation for breeding with any degree of precision required. We believe that, for preselected landraces, this gain in information overcompensates the moderate additional costs associated with the production and storage of DH libraries. Moreover, the generated DH libraries offer additional advantages as they can be used readily to investigate research questions such as the intrinsic genetic load in open-pollinated populations. Due to their high reproducibility, they can also be instrumental in investigating diversity not captured by SNP data, such as epigenetic modifications or plant–microbe interactions and their role in phenotypic variation of genetic resources. When extending their sample size to a few hundred, they are an excellent biological resource for high-resolution association mapping of quantitative trait loci, as they fully recover the ancestral recombination events of their source populations.

It seems realistic to assume that production costs of DH libraries will fall with advances in biotechnology based on CENH3, MATRILINEAL, or other methods (Ravi and Chan 2010; Melchinger et al. 2013; Kelliher et al. 2017). For (partially) allogamous species, such as maize, sugar beet, rape seed, and rye, the DH method has already started its success in improvement of elite germplasm (Dwivedi et al. 2015). Building libraries of “immortalized,” homozygous genotypes from ancestral populations will be a major step forward in making genetic resources directly accessible for crop improvement.

Supplementary Material

Supplemental material is available online at www.genetics.org/lookup/suppl/doi:10.1534/genetics.115.186205/-/DC1.