Abstract
Genetic diversity is shaped by the interaction of drift and selection, but the details of this interaction are not well understood. The impact of genetic drift in a population is largely determined by its demographic history, typically summarized by its long-term effective population size (Ne). Rapidly changing population demographics complicate this relationship, however. To better understand how changing demography impacts selection, we used whole-genome sequencing data to investigate patterns of linked selection in domesticated and wild maize (teosinte). We produce the first whole-genome estimate of the demography of maize domestication, showing that maize was reduced to approximately 5% the population size of teosinte before it experienced rapid expansion post-domestication to population sizes much larger than its ancestor. Evaluation of patterns of nucleotide diversity in and near genes shows little evidence of selection on beneficial amino acid substitutions, and that the domestication bottleneck led to a decline in the efficiency of purifying selection in maize. Young alleles, however, show evidence of much stronger purifying selection in maize, reflecting the much larger effective size of present day populations. Our results demonstrate that recent demographic change — a hallmark of many species including both humans and crops — can have immediate and wide-ranging impacts on diversity that conflict with would-be expectations based on Ne alone.
The genetic diversity of populations is determined by a constant interplay between genetic drift and natural selection. Drift is a consequence of a finite population size and the random sampling of gametes each generation1. In contrast to the stochastic effects of drift, selection systematically alters allele frequencies by favoring particular alleles at the expense of others as a result of their effects on fitness. Researchers often study drift by excluding potentially selected sites2,3,4, or selection by focusing on site-specific patterns under the assumption that genome-wide diversity reflects primarily the action of drift5.
Drift and selection do not operate independently to determine genetic variability, however, in large part because linkage allows the effects of selection to be wide-ranging6,7,8. Linked selection, which refers to the effects of selection at one site on diversity at linked sites8, can take the form of hitch-hiking, when the frequency of a neutral allele changes as a result of positive selection at a physically linked site6, or background selection, where diversity is reduced at loci linked to a site undergoing selection against deleterious alleles9. Recent work in Drosophila, for example, has shown that virtually the entire genome is impacted by the combined effects of these processes10,11,12.
The impact of linked selection, in turn, is heavily influenced by the effective population size (Ne), as the efficiency of natural selection is proportional to the product Nes, where s is the strength of selection on a variant8,13,14,15. The effective size of a population is not static, and nearly all species, including flies16, humans17, domesticates18,19, and non-model species20 have experienced recent or ancient changes in Ne. Although much is known about how the long-term average Ne affects linked selection13, relatively little is understood about the immediate effects of more recent changes in Ne on patterns of linked selection.
Because of its relatively simple demographic history and well-developed genomic resources, maize (Zea mays) represents an excellent organism to study these effects. Archaeological and genetic studies have established that maize domestication began in Central Mexico at least 9,000 years bp21,22, and involved a population bottleneck followed by recent expansion23,24,25. Because of this simple but dynamic demographic history, domesticated maize and its wild ancestor teosinte can be used to understand the effects of changing Ne on linked selection. In this study, we leverage the maize-teosinte system to study these effects by first estimating the parameters of the maize domestication bottleneck using whole-genome resequencing data and then investigating the relative importance of different forms of linked selection on diversity in the ancient and more recent past. We show that, while patterns of overall nucleotide diversity reflect long-term differences in Ne, recent growth following domestication qualitatively changes these effects, thereby illustrating the importance of a comprehensive understanding of demography when considering the effects of selection genome-wide.
RESULTS
Patterns of diversity differ between genic and intergenic regions of the genome
To investigate how demography and linked selection have shaped patterns of diversity in maize and teosinte, we analyzed data from 23 maize and 13 teosinte genomes from the maize HapMap 2 and HapMap 3 projects26,27. As a preliminary step, we evaluated levels of diversity inside and outside of genes across the genome. We find broad differences in genic and intergenic diversity consistent with earlier results28(Figure 1). In maize, mean pairwise diversity (π) within genes was significantly lower than at sites at least 5 kb away from genes (0.00668 vs 0.00691, p < 2 × 10−44). Diversity differences in teosinte are even more pronounced (0.0088 vs. 0.0115, p ≈ 0). Differences were also apparent in the site frequency spectrum, with mean Tajima’s D positive in genic regions in both maize (0.4) and teosinte (0.013) but negative outside of genes (−0.087 in maize and −0.25 in teosinte, p ≈ 0 for both comparisons). These observations suggest that diversity in genes is not evolving neutrally, but instead is reduced by the impacts of selection on linked sites.
Demography of maize domestication
We next estimated a demographic model of maize domestication (Figure 2). To minimize the impact of selection on our estimates29, we only included sites >5kb from genes. The most likely model estimates an ancestral population mutation rate of θ = 0.0147 per bp, which translates to an effective population size of Na ≈ 123, 000 teosinte individuals. We estimate that maize split from teosinte ≈ 15, 000 generations in the past, with an initial size of only ≈ 5% of the ancestral Na. After its split from teosinte, our model posits exponential population growth in maize, estimating a final modern effective population size of Nm ≈ 370, 000. Although our model provides only a rough approximation of migration rates, we included migration parameters during demographic inference because omitting these could bias our population size estimates. We observe that maize and teosinte have continued to exchange migrants after the population split, with gene flow from teosinte to maize was Mtm = 1.1 × 10−5 × Na migrants per generation, and from maize to teosinte we estimate Mmt = 1.4 × 10−5 × Na migrants per generation.
Because our modest sample size of fully sequenced individuals has limited power to infer recent population expansion, we investigated two alternative approaches for demographic inference. First, we utilized genotyping data from more than 4,000 maize landraces30 to estimate the modern maize effective population size. Because rare variants provide the best information about recent effective population sizes31, we estimate Ne using a singleton-based estimator32 of the population mutation rate θ = 4Neμ and published values of the mutation rate33 (see online methods for details). This yields a much higher estimate of the modern maize effective population size at Nm ≈ 993,000. Finally, we employed a model-free coalescent approach34 to estimate population size change using a subset of six genomes each of maize and teosinte. Though this analysis suggests non-equilibrium dynamics for teosinte not included in our initial model, it is nonetheless broadly consistent with the other approaches, identifying population isolation beginning between 10,000 and 15,000 generations ago, a clear domestication bottleneck, and ultimately rapid population expansion in maize to an extremely large extant size of ≈ 109 (Figure S2). Our assessment of the historical demography of maize and teosinte provides context for subsequent analyses of linked selection.
Hard sweeps do not explain diversity differences
When selection increases the frequency of a new beneficial mutation, a signature of reduced diversity is left at surrounding linked sites6. To evaluate whether patterns of such “hard sweeps” could explain observed differences in diversity between genic and intergenic regions of the genome, we compared diversity around missense and synonymous substitutions between Tripsacum and either maize or teosinte. If a substantial proportion of missense mutations have been fixed due to hard sweeps, diversity around these substitutions should be lower than around synonymous substitutions. We observe this pattern around the causative amino acid substitution in the maize domestication locus tga1 (Figure S1), likely the result of a hard sweep during domestication35,36. Genome-wide, however, we observe no differences in diversity at sites near synonymous versus missense substitutions in either maize or teosinte (Figure 3).
Previous analyses have suggested that this approach may have limited power because a relatively high proportion of missense substitutions will be found in genes that, due to weak purifying selection, have higher genetic diversity37. To address this concern, we took advantage of genome-wide estimates of evolutionary constraint38 calculated using genomic evolutionary rate profile (GERP) scores39. We then evaluated substitutions only in subsets of genes in the highest and lowest 10% quantile of mean GERP score, putatively representing genes under the strongest and weakest purifying selection. As expected, we see higher diversity around substitutions in genes under weak purifying selection, but we still find no difference in diversity near synonymous and missense substitutions in either subset of the data (Figure S3). Taken together, these data suggest hard sweeps do not play a major role in patterning genic diversity in either maize or teosinte.
Diversity is strongly influenced by purifying selection
In the case of purifying or background selection, diversity is reduced in functional regions of the genome via removal of deleterious mutations9. We investigated purifying selection in maize and teosinte by evaluating the reduction of diversity around genes. Pairwise diversity is strongly reduced within genes for both maize and teosinte (Figure 4A) but recovers quickly at sites outside of genes, consistent with the low levels of linkage disequilibrium generally observed in these subspecies26,40. The reduction in relative diversity is more pronounced in teosinte, reaching lower levels in genes and occurring over a wider region.
Our previous comparison of synonymous and missense substitutions has low power to detect the effects of selection acting on multiple beneficial mutations or standing genetic variation, because in such cases diversity around the substitution may be reduced to a lesser degree41,42. Nonetheless, such “soft sweeps” are still expected to occur more frequently in functional regions of the genome and could provide an alternative explanation to purifying selection for the observed reduction of diversity at linked sites in genes. To test this possibility, we performed a genome-wide scan for selection using the H12 statistic, a method expected to be sensitive to both hard and soft sweeps43. Qualitative differences between maize and teosinte in patterns of diversity within and outside of genes remained unchanged even after removing genes in the top 20% quantile of H12 (Figure S7A). We interpret these combined results as suggesting that purifying selection has predominantly shaped diversity near genes and left a more pronounced signature in the teosinte genome due to the increased efficacy of selection resulting from differences in long-term effective population size.
Population expansion leads to stronger purifying selection in modern maize
Motivated by the rapid post-domestication expansion of maize evident in our demographic analyses, we reasoned that low-frequency — and thus younger — polymorphisms might show patterns distinct from pairwise diversity, which is determined primarily by intermediate frequency — therefore comparably older — alleles. Singleton diversity around missense and synonymous substitutions (Figure S4) appears nearly identical to results from pairwise diversity (Figure 3), providing little support for a substantial recent increase in the number or strength of hard sweeps occurring in maize.
In contrast, we observe a significant shift in the effects of purifying selection: singleton polymorphisms are more strongly reduced in and near genes in maize than in teosinte, even after downsampling our maize data to account for differences in sample size (Figure 4B). This result is the opposite of the pattern observed for π, where teosinte demonstrated a stronger reduction of diversity in and around genes than did maize. As before, this relationship remained after we removed the 20% of genes with the highest H12 values (Figure S7). While direct comparison of pairwise and singleton diversity within taxa is consistent with non-equilibrium dynamics in teosinte, these too reveal much stronger differences in maize (Figure S5) and mirror results from simulations of purifying selection (Figure S6).
DISCUSSION
Demography of domestication
Although a number of authors have investigated the demography of maize domestication23,24,25, these efforts relied on data only from genic regions of the genome and made a number of limiting assumptions about the demographic model. We show that diversity within genes has been strongly reduced by the effects of linked selection, such that even synonymous polymorphisms in genes are not representative of diversity at unconstrained sites. This implies that genic polymorphism data are unable to tell the complete or accurate demographic history of maize, but the rapid recovery of diversity outside of genes demonstrates that sites far from genes can be reasonably used for demographic inference. Furthermore, by utilizing the full joint SFS, we are able to estimate population growth, gene flow, and the strength of the domestication bottleneck without making assumptions about its duration. This model paves the way for future work on the demography of do-mestication, evaluating for example the significance of differences in gene flow estimated here or removing assumptions about demographic history in teosinte.
One surprising result from our model is the estimated divergence time of maize and teosinte approximately 15, 000 generations before present. While this appears to conflict with archaeological estimates44, we emphasize that this estimate reflects the fact that the genetic split between populations likely preceded anatomical changes that can be identified in the archaeological record. We also note that our result may be inflated due to population structure, as our geographically diverse sample of teosinte may include populations diverged from those that gave rise to maize.
The estimated bottleneck of ≈ 5% of the ancestral teosinte population seems low given that maize landraces exhibit ≈ 80% of the diversity of teosinte28, but our model suggests that the effects of the bottleneck on diversity are likely ameliorated by both gene flow and rapid population growth (Figure 2). Although we estimate that the modern effective size of maize is larger than teosinte, the small size of our sample reduces our power to identify the low frequency alleles most sensitive to rapid population growth31, and our model is unable to incorporate growth faster than exponential. Both alternative approaches we employ estimate a much larger modern effective size of maize in the range of ≈ 106 – 109, an order of magnitude or more than the current size of teosinte. Census data suggest these estimates are plausible: there are 47.9 million ha of open-pollinated maize in production45, likely planted at a density of ≈ 25, 000 individuals per hectare46. Assuming the effective size is only ≈ 0.4% of the census size (i.e. 1 ear for every 1000 male plants), this still implies a modern effective population size of more than four billion. While these genetic and census estimates are likely inaccurate, all of the evidence points to the fact that the modern effective size of maize is extremely large.
Hard sweeps do not shape genome-wide diversity in maize
Our findings demonstrate that classic hard selective sweeps have not contributed substantially to genome-wide patterns of diversity in maize, a result we show is robust to concerns about power due to the effects of purifying selection37. Although our approach ignores the potential for hard sweeps in noncoding regions of the genome, a growing body of evidence argues against hard sweeps as the prevalent mode of selection shaping maize variability. Among well-characterized domestication loci, only the gene tga1 shows evidence of a hard sweep on a missense mutation36, while published data for several loci are consistent with soft sweeps from standing variation47,48 or multiple mutations49. Moreover, genome-wide studies of domestication28, local adaptation50 and modern breeding51,52 all support the importance of standing variation as primary sources of adaptive variation. Soft sweeps are expected to be common when 2Neμb ≥ 1, where μb is the mutation rate of beneficial alleles with selection coefficient sb42. Assuming a mutation rate of 3 × 10−833 and that on the order of ≈ 1 – 5% of mutations are beneficial53, this implies that soft sweeps should be common in both maize and teosinte for mutational targets >> 10kb — a plausible size for quantitative traits or for regulatory evolution targeting genes with large up‐ or down-stream control regions47 e.g., Indeed, many adaptive traits in both maize54 and teosinte55 are highly quantitative, and adaptation in both maize28 and teosinte56 has involved selection on regulatory variation.
The absence of evidence for a genome-wide impact of hard sweeps in coding regions differs markedly from observations in Drosophila57 and Capsella58, but is consistent with data from humans59,60. Comparisons of the estimated percentages of nonsynonmyous substitutions fixed by natural selection10,58,61,62 give similar results. While differences in long-term Ne likely explains some of the observed variation across species, we see little change in the importance of hard sweeps in genes in singleton diversity in modern maize (Figure S4), perhaps suggesting other factors may contribute to these differences as well. One possibility, for example, is that, if mutational target size scales with genome size, the larger genomes of human and maize may offer more opportunities for noncoding loci to contribute to adaptation, with hard sweeps on nonsynonymous variants then playing a relatively smaller role. Support for this idea comes from numerous cases of adaptive transposable element insertion modifying gene regulation in maize47,63, 64, 65 and studies of local adaptation that show enrichment for SNPs in regulatory regions in teosinte56 and humans66 but for nonsynonymous variants in the smaller Arabidopsis genome. Our results, for example, are not dissimilar to findings in the comparably-sized mouse genome, where no differences are seen in diversity around nonsynonymous and synonymous substitutions in spite of a large Ne and as many as 80% of adaptive substitutions occuring outside of genes67. Future comparative analyses using a common statistical framework (e.g.14) and considering additional ecological and life history factors (c.f.15) should allow explicit testing of this idea.
Demography influences the efficiency of purifying selection
One of our more striking findings is that the impact of purifying selection on maize and teosinte qualitatively changed over time. We observe a more pronounced decrease in π around genes in teosinte than maize (Figure 4A), but the opposite trend when we evaluate diversity using singleton polymorphisms (Figure 4B). The efficiency of purifying selection is proportional to effective population size68, and these results are thus consistent with our demographic analyses which show a domestication bottleneck and smaller long-term Ne in maize23,24,25,61 followed by recent rapid expansion and a much larger modern Ne. Simple foward-in-time population genetic simulations qualitatively confirm these results, and further suggest that the observed patterns are likely cause by sites under relatively weak purifying selection S6.
Although demographic change affects the efficiency of purifying selection, it may have limited implications for genetic load. Recent population bottlenecks and expansions have increased the relative abundance of rare and deleterious variants in domesticated plants69,70 and human populations out of Africa31,71, and such variants may play an important role in phenotypic variation71,72,73. Nonetheless, demographic history may have little impact on the overall genetic load of populations74,75, as decreases in Ne that allow weakly deleterious variants to escape selection also help purge strongly deleterious ones, and the increase of new deleterious mutations in expanding populations is mitigated by their lower initial frequency and the increasing efficiency of purifying selection75,76,77.
Rapid changes in linked selection
Our results demonstrate that consideration of long-term differences in Ne cannot fully capture the dynamic relationship between demography and selection. While a number of authors have tested for selection using methods that explicitly incorporate or are robust to demographic change62,78,79 and others have compared estimates of the efficiency of adaptive and purifying selection across species80 or populations81, previous analyses of the impact of linked selection on genome-wide diversity have relied on single estimates of the effective population size14,15. Our results show that demographic change over short periods of time can quickly change the dynamics of linked selection: mutations arising in extant maize populations are much more strongly impacted by the effects of selection on linked sites than would be suggested by analyses using long-term effective population size. As many natural and domesticated populations have undergone considerable demographic change in their recent past, long-term comparisons of Ne are likely not informative about current processes affecting allele frequency trajectories.
AUTHOR CONTRIBUTIONS
TMB and JRI devised this study. TMB, LW, JRI, and KC analyzed the data. AD performed early-stage simulations. TMB, JRI, and MBH wrote the manuscript.
COMPETING INTERESTS STATEMENT
The authors declare no competing financial interests.
ONLINE METHODS
BASH, R, and Python scripts
All scripts used for analysis are available in an online repository at https://github.com/timbeissinger/Maize-Teo-Scripts.
Plant materials
We made use of published sequences from inbred accessions of teosinte (Z. mays ssp. parviglumis) and maize landraces from the Maize HapMap3 panel as part of the Panzea project26,27,82. From these data, we removed 4 teosinte individuals that were not ssp. parviglumis or appeared as outliers in an initial principal component analysis conducted with the package adegenet83 (Figure S8), leaving 13 teosinte and 23 maize that were used for all subsequent analyses (Table S1). We also utilized a single individual of (Tripsacum dactyloides) as an outgroup. All bam files are available at /iplant/home/shared/panzea/hapmap3/bam_internal/v3_bams_bwamem.
Physical and genetic maps
Sequences were mapped to the maize B73 version 3 reference genome84 (ftp://ftp.ensemblgenomes.org/pub/plants/release-22/fasta/zea_mays/dna/) as described by27. All analyses made use of uniquely mapping reads with mapping quality score ≥ 30 and bases with base quality score ≥ 20; quality scores around indels were adjusted following85. We converted physical coordinates to genetic coordinates via linear interpolation of the previously published 1cM resolution NAM genetic map86.
Estimating the site frequency spectrum
We estimated both the genome-wide site frequency spectrum (SFS) as well as a separate SFS for genic (within annotated transcript) and intergenic (≥ 5kb from a transcript) regions. We used the biomaRt package87,88 of R89 to parse annotations from genebuild version 5b of AGPv3. We estimated single population and joint SFS with the software ANGSD90, including all positions with at least one aligned read in ≥ 80% of samples in one or both populations. We assumed individuals were fully inbred and treated each line as a single haplotype. Because ANGSD cannot calculate a folded joint SFS, we first polarized SNPs using the maize reference genome and then folded spectra using δaδi4.
Demographic inference
We used the software δaδi4 to estimate parameters of a domestication bottleneck from the joint maize-teosinte SFS, using only sites > 5kb from a gene to ameliorate the effects of linked selection. To minimize the number of parameters estimated, we employed a simple demographic model which posits a teosinte population of constant effective size Na. At time Tb generations in the past, this population gave rise to a maize population of size Nb which grew exponentially to size Nm in the present (Figure 2). The model includes migration of Mmt individuals each generation from maize to teosinte and Mtm individuals from teosinte to maize. We estimated Na using δaδi’s estimation of θ = 4Naμ from the data and a mutation rate of μ = 3 × 10−833. We estimated all other parameters using 1,000 δaδi optimizations and allowing initial values between runs to be randomly perturbed by a factor of 2. Optimized parameters along with their initial values and upper and lower bounds can be found in table S2. We report parameter estimates from the optimization run with the highest log-likelihood.
We further made use of a large genotyping data set of more than 4,000 partially imputed maize landraces30 to estimate the modern maize Ne from singleton counts. We filtered these data to include only SNPs with data in ≥ 1, 500 individuals, and then projected the SFS down to a sample of 500 individuals by sampling each marker without replacement 1,000 times according to the observed allele frequencies. We then estimated Ne from the data assuming μ = 3 × 10−8 33 and the relation 32, where S is the total number of singleton SNPs and L is the total number of SNPs in the dataset.
As a final estimate of demography, we employed MSMC34 to complement our model-based demographic inference. We used six each of maize and teosinte (BKN022, BKN025, BKN029, BKN030, BKN031, BKN033, TIL01, TIL03, TIL09, TIL10, TIL11 and TIL14), treating each inbred genome as a single haplotype. We called SNPs in ANGSD90 using a SNP p-value of 1e–6 against a reference genome masked using SNPable (http://lh3lh3.users.sourceforge.net/snpable.shtml). We then removed heterozygous genotypes and filtered sites with a mapping quality < 30, a base quality < 20, or a |log2(depth)| < 1. We ran MSMC with pattern parameter 20 × 2 + 20 × 4 + 10 × 2 (Figure S2A) for population size inference. To estimate the rate of cross-coalescence we used four maize and four teosinte haplotypes with pattern parameter 20 × 1 + 20 × 2 (Figure S2B).
Diversity
We made use of the software ANGSD90 for diversity calculations and genotype calling. We calculated diversity statistics in maize and teosinte in 1 kb non-overlapping windows using filters as described above for the SFS. We used allele counts to estimate the number of singleton polymorphisms in each window, and used binomial sampling to create a second maize data set down-sampled to have the same number of samples as teosinte. We called genotypes in maize, teosinte, and Tripsacum at sites with a SNP p-value < 10−6 and when the genotype posterior probability > 0.95. We identified substitutions in maize and teosinte as all sites with a fixed difference with Tripsacum and ≤ 20% missing data. Substitutions were classified as synonymous, or missense using the ensembl variant effects predictor91. For each window with > 100bp of data we computed the genetic distance between the window center and the nearest synonymous and missense substitution as well as the genetic distance to the center of the nearest gene transcript.
Selection scan
We scanned the genome to identify sites that have experienced recent positive selection using the H12 statistic43 in sliding windows of 200 SNPs with a step of 25 SNPs.
Simulations
We used the program bneck_selection_ind included in version 0.4.4 of the forward-in-time population genetic simulation library fwdpp92 https://github.com/molpopgen/fwdpp]. All simulations used a population mutation rate of θ = 20, a population recombination rate of ρ = 20, and simulated 150,000 burn-in generations at an ancestral population size of N1 = 15, 000 to establish equilibrium, after which the population instantly changed to size N2 and then grew exponentially for 1,000 generations to size N3. To simulate a constant size population emulating teosinte, we set N2 = N3 = 15, 000. For maize we simulated a bottleneck similar to that estimated in Figure 2 by setting N2 = 750, followed by exponential growth to a large modern population size of N3 = 150, 000. For each taxon, we performed 1,000 simulations for each of five values of the strength of purifying selection: s = (0,10−6, 10−5, 10−4, 10−3}. All mutations were assumed to be codominant. To mimic nonsynonymous changes at a coding locus, we assumed that of mutations were selected. We calculated summary statistic across all sites using version 0.3.4 of msstats (https://github.com/molpopgen/msstats/releases).
Supporting Information
ACKNOWLEDGEMENTS
We are indebted to Graham Coop and Simon Aeschbacher for their constructive input during this study. We thank Robert Bukowski and Qi Sun for providing early-access data from maize HapMap3. Funding was provided by NSF Plant Genome Research Project 1238014 and the USDA-Agricultural Research Service.
References
- [1].↵
- [2].↵
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].↵
- [44].↵
- [45].↵
- [46].↵
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
- [60].↵
- [61].↵
- [62].↵
- [63].↵
- [64].↵
- [65].↵
- [66].↵
- [67].↵
- [68].↵
- [69].↵
- [70].↵
- [71].↵
- [72].↵
- [73].↵
- [74].↵
- [75].↵
- [76].↵
- [77].↵
- [78].↵
- [79].↵
- [80].↵
- [81].↵
- [82].↵
- [83].↵
- [84].↵
- [85].↵
- [86].↵
- [87].↵
- [88].↵
- [89].↵
- [90].↵
- [91].↵
- [92].↵