A Large-Scale Screen for Artificial Selection in Maize Identifies Candidate Agronomic Loci for Domestication and Crop Improvement

Abstract

Maize (Zea mays subsp mays) was domesticated from teosinte (Z. mays subsp parviglumis) through a single domestication event in southern Mexico between 6000 and 9000 years ago. This domestication event resulted in the original maize landrace varieties, which were spread throughout the Americas by Native Americans and adapted to a wide range of environmental conditions. Starting with landraces, 20th century plant breeders selected inbred lines of maize for use in hybrid maize production. Both domestication and crop improvement involved selection of specific alleles at genes controlling key morphological and agronomic traits, resulting in reduced genetic diversity relative to unselected genes. Here, we sequenced 1095 maize genes from a sample of 14 inbred lines and chose 35 genes with zero sequence diversity as potential targets of selection. These 35 genes were then sequenced in a sample of diverse maize landraces and teosintes and tested for selection. Using two statistical tests, we identified eight candidate genes. Extended gene sequencing of these eight candidate loci confirmed that six were selected throughout the gene, and the remaining two exhibited evidence of selection in the 3′ portion of each gene. The selected genes have functions consistent with agronomic selection for nutritional quality, maturity, and productivity. Our large-scale screen for artificial selection allows identification of genes of potential agronomic importance even when gene function and the phenotype of interest are unknown.

INTRODUCTION

One prominent goal of plant molecular biology is to identify and characterize the genes responsible for phenotypic variation. Historically, quantitative trait locus (QTL) mapping has been used to localize genomic regions contributing to phenotypic variation, but this approach has rarely led to candidate gene isolation. In crop systems, for example, only a limited number of genes has been isolated, cloned, and characterized based on information from a QTL analysis (e.g., fw2.2, Frary et al., 2000; Hd1, Yano et al., 2000; tga1, Wang et al., 2005). Another approach to defining the genes that control phenotypic variation is association analysis (Thornsberry et al., 2001). This approach is especially promising when prior information identifies potential candidate genes thought to contribute to the trait of interest. In the absence of candidate genes, a full-genome scan is necessary, and the potential for false positives (Type I error) in crop plants with their large genome size is extremely high.

Both QTL and association approaches rely on segregating phenotypic and molecular genetic variation. In crop systems, however, some genes underlying agronomic traits are expected to be bereft of segregating genetic variants because artificial selection has substantially decreased genetic diversity. One can think of selection occurring in two stages: domestication and crop improvement. Domestication resulted in the original landrace varieties, which were adapted to a wide range of environmental conditions. These landraces provided the genetic material for modern plant breeders to select improved varieties and inbred lines by enhancing traits controlling agricultural productivity and performance, such as yield and resistance to biotic and abiotic stresses. Consequently, crop varieties experienced strong selection at genes controlling traits of agronomic importance during their domestication and improvement by plant breeding (Wang et al., 1999; Whitt et al., 2002; Palaisa et al., 2003; Gallavotti et al., 2004; Wang et al., 2005). One result from selection during domestication and improvement is that QTL and association methods may miss the most interesting class of genes (i.e., those genes that lack genetic diversity because of a history of selection on their key role in controlling desirable agronomic traits).

How then, can one identify this selected class of genes that contributes to agronomic traits? In this and a previous article (Wright et al., 2005), we have been using population genetics approaches to identify selected genes in maize (Zea mays subsp mays). Maize was domesticated from teosinte (Z. mays subsp parviglumis) through a single domestication event in southern Mexico between 6000 and 9000 years ago (Piperno and Flannery, 2001; Matsuoka et al., 2002). Maize and teosinte differ in many aspects of plant morphology and productivity. Despite its selection history, most maize genes retain high levels of nucleotide diversity (Tenaillon et al., 2001). Maize is an outcrossing species and exhibits a high level of recombination (Fu et al., 2002), and the historical population size was large (Vigouroux et al., 2002a), all factors that contribute to a rapid decay of linkage disequilibrium (LD) among current maize inbred lines (Remington et al., 2001; Tenaillon et al., 2001). Therefore, maize is a model crop for performing association analysis to identify genes controlling agronomic traits (Thornsberry et al., 2001; Rafalski and Morgante, 2004). Several studies in maize have also shown that there is a large contrast in nucleotide diversity between genes with a history of selection and neutral genes (Wang et al., 1999; Whitt et al., 2002; Gallavotti et al., 2004; Tenaillon et al., 2004).

We recently began to address the genome-wide effects of artificial selection in maize by an analysis of single nucleotide polymorphisms (SNPs) in 774 genes in 16 teosinte accessions and 14 maize inbred lines (Wright et al., 2005). In that study, we concluded that 2 to 4% or ∼1200 maize genes show evidence of selection. Because the sequencing involved a single amplification product from each gene, the power to detect selected genes may have been limited. All genes significant for selection retained minimal, if any polymorphism, among the maize inbred lines. This result suggests a simplified method to conduct large-scale screens for selected genes in maize, which is sequencing a short region of each gene in inbred lines and only sequencing in teosinte those genes with very low inbred polymorphism. Another major question unaddressed by Wright et al. (2005) is the extent to which selection on any particular gene occurred at domestication rather than during subsequent crop improvement. This question is of both academic interest in understanding genetic consequences of selection and of practical importance to plant breeders concerned with diversifying germplasm used in maize breeding programs. For example, improvement genes, while lacking variation in elite inbreds, could be studied via QTL methods in crosses with landraces, and the germplasm diversity for these genes may be enhanced using exotic maize resources. For domestication genes, the plant breeder would have to extend crosses to at least the teosintes to introduce novel alleles.

In this study, we conduct a large-scale screen to discover genes responsible for maize domestication and improvement. Based on previous studies (Wang et al., 1999; Whitt et al., 2002; Palaisa et al., 2003; Gallavotti et al., 2004; Tenaillon et al., 2004), it is clear that the domestication event resulted in a loss of genetic diversity between teosinte and maize landraces and that modern plant breeding reduced the genetic diversity in maize inbreds relative to maize landraces (Figure 1). Moreover, genetic diversity in neutral (unselected) genes is expected to be reduced only by bottleneck effects, thereby retaining more diversity than selected genes (Figure 1). This reasoning leads to the prediction that genes strongly impacted by domestication or improvement are enriched in the subset of genes that exhibit low nucleotide diversity in modern improved varieties (Wright et al., 2005). Hence, we sequenced a large set of randomly chosen genes in diverse maize inbreds and identified genes with zero nucleotide diversity as potential targets of selection. These candidate genes were sequenced in additional diverse sets of maize landraces and teosintes. Finally, statistical analyses contrasting DNA sequence diversity among maize inbreds, landraces, and teosintes were performed to document selection and to determine whether selection occurred primarily during domestication or crop improvement. For the selection candidates with the strongest evidence of selection, we performed extended sequencing throughout the available gene sequence to confirm selection and to define the regions of the genes under selection. The selected genes have functions consistent with agronomic selection for nutritional quality, maturity, and productivity.

Figure 1.

Effect of Domestication and Plant Breeding on Genetic Diversity of Maize Genes.

The colored circles represent different alleles. The shaded areas indicate bottleneck effects placed on all genes by the processes of domestication and improvement (plant breeding). Our model assumes that there will be three types of genes: neutral genes that show reduction of diversity by the general bottleneck effects, domestication genes in which diversity is greatly reduced by selection between the teosintes and landraces, and improvement genes in which diversity is greatly reduced by selection between the landraces and inbreds.

RESULTS

Genetic Diversity in Maize Inbreds, Landraces, and Teosintes

To efficiently identify potential selected genes, a panel of inbred lines was established to represent as much of the genetic diversity of modern maize inbreds as possible. Simple sequence repeat (SSR) polymorphism data were used to choose 14 maize inbreds lines with maximum allelic diversity (Liu et al., 2003), with the constraint that final composition would include seven temperate inbreds and seven tropical inbreds (Table 1). This constraint was added to allow us to contrast genetic diversity in the temperate and tropical maize germplasm breeding pools. Sequence alignments for a single PCR product were obtained for 1095 randomly selected maize genes. An average of ∼13 inbred sequences per gene using 14 diverse inbreds and an average length of alignment of 280 bp without gaps were obtained (see Supplemental Table 1 online; summarized in Table 2). Among these 1095 loci, we identified 6169 SNPs, an average of 5.6 SNPs per alignment. This corresponds to one SNP approximately every 50 bp per alignment length or every 150 bp between any randomly chosen pair of maize inbreds. We found 2848 SNPs shared in temperate and tropical inbreds, 1742 SNPs specifically within temperate inbreds and 1579 SNPs within the tropical inbreds. The number of segregating sites (S) was not significantly different between temperate and tropical inbreds (Mann-Whitney U test, P > 0.12). There also was not a significant difference between temperate and tropical inbreds in the nucleotide diversity measure (π; Student's t test, P > 0.22), which is average proportion of pairwise nucleotide differences per nucleotide site (Tajima, 1983). The average π in all maize inbreds was 0.0067 (Figure 2), confirming the high level of sequence diversity previously reported for maize inbred lines (Tenaillon et al., 2001; Wright et al., 2005).

Table 1.

Plant Materials Used in This Study

Type	Inbred/Race (Accession or Type)	Origin
Inbred line	B73 (temperate)	Iowa
Z. mays subsp mays	Hp301 (temperate)	Indiana
	Il14H (temperate)	Illinois
	Ky21 (temperate)	Kentucky
	M37W (temperate)	South Africa
	Mo17 (temperate)	Missouri
	Oh43 (temperate)	Ohio
	CML69 (tropical)	Mexico
	CML247 (tropical)	Mexico
	CML322 (tropical)	Mexico
	CML333 (tropical)	Mexico
	Ki3 (tropical)	Thailand
	Ki11 (tropical)	Thailand
	NC350 (tropical)	North Carolina
Landrace	Assinboine (PI213793)	Northern U.S.
Z. mays subsp mays	Bolita (OAX 68)	Southern Mexico
	Cateto Sulino (URG 11)	Uruguay
	Chalqueno (MEX 48)	Central Mexico
	Chapalote (SIN 2)	Western Mexico
	Conico (PUE 32)	Central Mexico
	Costeno (VEN 453)	Venezuela
	Cristalino Norteno (CHI 349)	Chile
	Dzit Bacal (GUA 131)	Guatemala
	Gordo (CHH 160)	Northern Mexico
	Guirua (MAG 450)	Colombia
	Nal-tel (YUC 7)	Southern Mexico
	Pissccotunto (APC 13)	Peru
	Sabanero (SAN 329)	Colombia
	Serrano (GUA 14)	Guatemala
	Zapalote Chico (OAX 70)	Southern Mexico
Teosinte	Balsas (Beadle and Kato Site 4)	Mexico
Z. mays subsp parviglumis	Balsas (CIMMYT 8783)	Mexico
	Balsas (CIMMYT 11355)	Mexico
	Balsas (INIFAP JSG 374)	Mexico
	Balsas (INIFAP JSG 378)	Mexico
	Balsas (INIFAP JSG y LOS 109)	Mexico
	Balsas (INIFAP JSG y LOS 119)	Mexico
	Balsas (INIFAP JSG y LOS 161)	Mexico
	Balsas (INIFAP JSP y Lo5130)	Mexico
	Balsas (Kato Site 4)	Mexico
	Balsas (USDA PI566686)	Mexico
	Balsas (Wilkes Site 6)	Mexico
	Jalisco (Benz 967)	Mexico
	Jalisco (INIFAP JSG y MAS 264)	Mexico
	Jalisco (INIFAP JSG y MAS 401)	Mexico
	Oaxaca (INIFAP JSG 197)	Mexico
T. dactyloides	WW-2120	Oklahoma
	MIA 34597	Colombia

Table 2.

Sequence Statistics for 1095 Genes in Diverse Maize Inbreds

Type	N	L	Total L	S	Total S	π
All maize inbreds	13.1	280.3	306,895	5.6	6,169	0.0067
Temperate inbreds	6.7	292.2	310,306	4.3	4,590	0.0065
Tropical inbreds	6.6	290.6	308,649	4.2	4,427	0.0061

N, average number of sequences in the final alignment; L, average length of the core alignments in which all sequences contain bases, excluding gaps; S, average number of segregating sites (SNPs) per alignment; total L and total S, sums for all 1095 alignments; π, average proportion of pairwise differences among all shared positions for all sequences in an alignment.

Figure 2.

Distribution of Genetic Diversity π for 1095 Maize Genes.

The genes were sequenced in the panel of 14 diverse maize inbreds (Table 1).

From the 1095 genes, we selected 35 genes with alignments >200 bp and zero nucleotide diversity, either from SNPs or from insertion/deletion polymorphisms that are also very common among maize inbreds (Bhattramakki et al., 2002). Although the lack of nucleotide diversity at these genes could reflect a history of selection during domestication or improvement, the low diversity in maize could also reflect low diversity in teosinte and/or the demographic effects of domestication, plant breeding, and/or chance events (e.g., genetic drift). In order to distinguish between selection and other effects and also to determine if selection occurred primarily during domestication or crop improvement, we sequenced the same region of these 35 genes in 16 diverse teosinte accessions and 16 diverse maize landraces (Tables 1 and 3; sequence alignments are presented in Supplemental Table 2 online). There were significant differences for number of haplotypes (h) among the three populations: inbreds < landraces < teosintes (Kruskal-Wallis H test, P < 0.001; multiple comparison Scheffé's F test, all P < 0.001) (Table 4). The Kruskal-Wallis H test for differences in π among the populations was also significant (P < 0.001). Although π in inbreds and landraces was significantly lower than π in teosintes (Scheffé's F test, P < 0.001), π in inbreds versus landraces was not significantly different (Scheffé's F test, P > 0.05). For comparison, we also sequenced the same three populations for four documented neutral (unselected) genes: adh1, bz2, fus6, and glb1 (Eyre-Walker et al., 1998; Hilton and Gaut, 1998; Tenaillon et al., 2001). For these neutral genes, h and π were not significantly different among the three populations (Kruskal-Wallis H test, P > 0.40 and P > 0.66, respectively). In addition, use of the Tajima's D test (Tajima, 1989) supported neutrality for adh1, bz2, fus6, and glb1in all three populations (P > 0.10).

Table 3.

Sequence Diversity in Maize Inbreds, Landraces, and Teosintes at 35 Genes That Exhibited Zero Sequence Diversity in Maize Inbreds and Four Neutral Genes

	Inbreds					Landraces					Teosintes
Gene	N	L	S	h	π	N	L	S	h	π	N	L	S	h	π
AY108876	14	390	0	1	0.00000	15	482	1	2	0.00028	15	445	9	8	0.00428
AY106123	14	327	0	1	0.00000	14	373	5	3	0.00250	15	374	5	4	0.00514
AY106190	14	326	0	1	0.00000	14	365	1	2	0.00072	16	330	2	3	0.00197
AY108504	14	321	0	1	0.00000	15	336	1	2	0.00040	15	336	3	3	0.00119
AY108957	14	315	0	1	0.00000	16	329	0	1	0.00000	14	309	1	2	0.00046
AY108255	14	300	0	1	0.00000	15	319	0	1	0.00000	14	318	8	5	0.00473
AY107195	14	288	0	1	0.00000	14	274	2	2	0.00104	15	302	10	7	0.00826
AY105850	13	271	0	1	0.00000	16	311	3	2	0.00314	16	311	5	6	0.00271
AY110109	14	271	0	1	0.00000	16	280	6	4	0.00307	16	273	10	8	0.01029
AY106826	14	265	0	1	0.00000	15	277	2	2	0.00096	16	277	4	4	0.00253
AY109101	14	260	0	1	0.00000	16	285	2	3	0.00088	13	288	3	4	0.00285
AY105752	14	246	0	1	0.00000	16	293	2	4	0.00191	14	309	2	3	0.00217
AY105060	14	248	0	1	0.00000	15	297	2	3	0.00090	15	285	21	8	0.02185
AY108178	14	239	0	1	0.00000	15	244	6	2	0.00609	14	230	12	8	0.01529
AY109011	14	237	0	1	0.00000	16	275	1	2	0.00045	15	273	6	6	0.00433
AY107827	13	229	0	1	0.00000	13	240	3	2	0.00192	14	261	5	4	0.00564
AY107821	14	228	0	1	0.00000	16	253	3	3	0.00148	14	258	3	3	0.00554
AY107949	13	229	0	1	0.00000	16	223	0	1	0.00000	16	231	2	3	0.00155
AY106125	14	217	0	1	0.00000	14	248	1	1	0.00058	15	242	9	7	0.00543
AY107673	12	220	0	1	0.00000	9	312	2	4	0.00258	12	292	2	2	0.00280
AY108759	14	214	0	1	0.00000	14	224	1	2	0.00235	16	219	5	4	0.00285
AY108231	14	213	0	1	0.00000	14	239	0	1	0.00000	16	238	4	5	0.00326
AY106616	14	209	0	1	0.00000	16	197	8	4	0.01193	16	226	6	5	0.00830
AY106734	14	193	0	1	0.00000	15	194	0	1	0.00000	15	242	1	2	0.00055
AY107952	13	531	0	1	0.00000	12	476	1	2	0.00064	12	541	12	6	0.00370
AY108388	14	441	0	1	0.00000	13	479	13	3	0.00418	16	476	22	6	0.00840
AY108552	13	313	0	1	0.00000	14	295	0	1	0.00000	14	338	2	3	0.00085
AI737881	14	294	0	1	0.00000	16	323	5	4	0.00421	14	306	12	11	0.00991
AY108194	13	253	0	1	0.00000	10	274	1	2	0.00073	16	280	4	4	0.00179
AY106702	14	247	0	1	0.00000	15	258	2	3	0.00103	14	257	3	4	0.00167
AY106371	14	235	0	1	0.00000	12	282	0	1	0.00000	15	284	6	6	0.00604
AY106889	14	240	0	1	0.00000	8	141	0	1	0.00000	15	266	3	4	0.00150
AY107535	14	237	0	1	0.00000	14	265	0	1	0.00000	16	264	3	4	0.00218
AY107529	14	235	0	1	0.00000	16	255	4	3	0.00275	14	255	6	5	0.00612
AY107818	14	203	0	1	0.00000	15	258	3	3	0.00199	15	253	4	4	0.00256
adh1	14	1217	42	6	0.01237	16	1309	58	10	0.01480	7	1217	55	5	0.01808
bz2	14	524	11	5	0.00910	16	594	17	8	0.00885	9	522	15	7	0.00702
fus6	14	238	10	4	0.01057	15	234	8	5	0.01213	10	227	10	6	0.01429
glb1	14	958	52	11	0.01747	14	957	86	14	0.02169	12	939	99	12	0.02433

Gene, the GenBank accession number of the original unigene sequence; N, number of sampled sequences; L, length of the core alignments in which all sequences contain bases, excluding gaps; S, total number of segregating sites; h, number of unique sequences (haplotypes); π, average proportion of pairwise differences per base pair.

Table 4.

Sequence Statistics for 35 Genes with Zero Inbred Diversity

Type	N	L	h	π
Inbreds	13.8	271.0	1.0	0.0000
Landraces	14.3	290.7	2.2	0.0017
Teosintes	14.8	296.8	4.9	0.0048

N, average number of sequences in the final alignment; L, average length of the core alignments in which all sequences contain bases, excluding gaps; h, average number of unique sequences (haplotypes) in the alignments; π, average proportion of pairwise differences per base pair among all shared positions for all sequences in an alignment.

Statistical Tests for Selection

Two separate tests of selection were conducted for each of the 35 candidate genes: Hudson-Kreitman-Aguadé (HKA) tests (Hudson et al., 1987) and coalescent simulations of domestication (CS) tests (Tenaillon et al., 2004). The HKA test requires an outgroup sequence to compare rates of divergence between species to levels of polymorphism within species. The test also requires reference loci that are not believed to have been affected by selection. The ratio of divergence to polymorphism is then compared between a putatively selected gene and reference gene(s). If the putatively selected gene has a significantly higher ratio of divergence to polymorphism, it is reasonable to postulate that polymorphism in the putative selected gene has been diminished by selection. Simulation studies of the genetic effects of artificial selection have demonstrated that the HKA is an appropriate and powerful test to find genes affected by artificial selection (Innan and Kim, 2004).

To perform the HKA test, we obtained an orthologous sequence of Tripsacum dactyloides for 31 of 35 candidate genes. We used a multiple locus implementation of the HKA test (see Methods) to compare each of our candidate genes to the four neutral genes adh1, bz2, fus6, and glb1 at three population levels: teosintes, landraces, and inbreds. Given the results of these HKA comparisons, we identified a gene as a candidate domestication gene if it was significant for selection both in inbreds and landraces but not significant in teosintes; this pattern indicates that selection occurred primarily between teosintes and landraces (Figure 1). Similarly, we identified a candidate improvement gene if significant for selection in inbreds but not significant in landraces and teosintes, indicating that diversity was reduced by modern plant breeding (Figure 1). Using this approach, we detected six candidate genes for domestication and six for improvement by the HKA analysis (Table 5, Figure 3). For example, HKA tests on gene AY108876 were significant for both the inbred (P = 0.010) and landrace sample (P = 0.014) (Table 5), suggesting that polymorphism in this gene is aberrantly low in both the landrace and inbred samples. Given our model in Figure 1, we therefore conclude that this gene was selected during domestication, prior to improvement. Similarly, HKA tests on gene AY110109 were significant only for the inbreds (P = 0.046), leading to the conclusion that this gene was selected during improvement.

Table 5.

Results of the Tests of Selection and Homology Searches

	P Values in HKA			Candidate Status by HKA	P Values in CS		Candidate Status by CS
Gene	I	L	T		I versus T	L versus T		Homology Search
AY108876	0.010*	0.014*	0.225	Domestication	0.0152*	0.0145*	Domestication	Amino acid transporter
AY106123	0.155	0.022*	0.014*	–	0.0730	0.6184	–	Putative poly(A) binding protein binding protein
AY106190	0.047*	0.047*	0.046*	–	0.1579	0.5561	–	Hypothetical protein
AY108504	0.030*	0.013*	0.036*	–	0.1219	0.4699	–
AY108957	0.057	0.023*	0.028*	–	0.3032	0.1827	–
AY108255	0.323	0.158	0.848	–	0.0294*	0.0038**	Domestication	Transcriptional factor
AY107195	0.067	0.383	0.858	Improvement	0.0175*	0.1471	Improvement	Auxin response factor
	0.047*	0.250	0.833
AY105850	NA	NA	NA	–	0.0393*	0.5508	Improvement
AY110109	0.046*	0.644	0.822	Improvement	0.0211*	0.7939	Improvement	GTP binding protein
AY106826	0.102	0.308	0.541	–	0.0898	0.6554	–
AY109101	0.082	0.217	0.251	–	0.0847	0.6766	–	Ser/Thr protein kinase
AY105752	0.040*	0.116	0.109	Improvement	0.1915	0.7282	–	Putative casein kinase
AY105060	0.016*	0.034*	0.863	Domestication	0.0065**	0.0008***	Domestication
AY108178	0.007**	0.053	0.217	Improvement	0.0059**	0.2970	Improvement	Circadian clock
	0.010*	0.039*	0.208	Domestication
AY109011	0.028*	0.028*	0.234	Domestication	0.0403*	0.1042	Improvement	NLI interacting factor
AY107827	0.085	0.691	0.536	–	0.0731	1.0000	–
AY107821	0.030*	0.105	0.083	Improvement	0.1350	0.7468	–	Universal stress protein
AY107949	0.033*	0.011*	0.037*	–	0.1713	0.0878	–
AY106125	0.175	0.305	0.864	–	0.0160*	0.0435*	Domestication
AY107673	0.027*	0.027*	0.018*	–	0.2279	0.7457	–
	0.023*	0.029*	0.020*
AY108759	0.053	0.062	0.292	–	0.0662	0.2380	–
AY108231	0.530	0.308	0.879	–	0.0588	0.0140*	–
AY106616	0.013*	0.291	0.101	Improvement	0.0499*	0.5148	Improvement	Ankyrin repeat-like protein
AY106734	NA	NA	NA	–	0.3544	0.2543	–
AY107952	0.008**	0.011*	0.363	Domestication	0.0161*	0.0460*	Domestication	Putative fruit protein, oxidoreductase
AY108388	NA	NA	NA	–	0.0047**	0.6850	Improvement	Cys synthase
AY108552	0.029*	0.008**	0.026*	–	0.1654	0.0938	–	Arm repeat-containing protein
AI737881	0.052	0.279	0.781	–	0.0034**	0.0992	Improvement	Putative alcohol oxidase
AY108194	0.148	0.183	0.483	–	0.1020	0.4527	–
	0.164	0.264	0.589
AY106702	0.019*	0.025*	0.041*	–	0.0808	0.6868	–	Putative suppressor of actin 1, inositol 5-phosphatase 3-like
AY106371	0.023*	0.010*	0.196	Domestication	0.0375*	0.0052**	Domestication	Putative methyl binding domain protein
AY106889	0.025*	0.019*	0.031*	–	0.0991	0.1456	–
AY107535	0.082	0.026*	0.245	–	0.0938	0.0284*	–	Putative integral membrane protein
AY107529	NA	NA	NA	–	0.0479*	0.8760	Improvement	Putative Ser carboxypeptidase
AY107818	0.010*	0.017*	0.011*	–	0.1105	0.7098	–	Putative polygalacturonase

Gene, the GenBank accession of the original unigene sequence. HKA tests: P value of the candidate gene by multiple locus HKA against the four neutral genes by the maximum cell value test. I, inbreds; L, landraces; T, teosintes. For AY107195, AY108178, AY107673, and AY108194, two distinct T. dactyloides haplotypes were used. NA, T. dactyloides sequence was not obtained. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Figure 3.

The Genetic Diversity π in Landraces and Teosintes for Each of 35 Genes with Zero Diversity in Inbred Lines.

The black bars indicate genetic diversity in the teosintes, and the white bars indicate genetic diversity in the landraces. Where no white bars are present, the genetic diversity in the landraces was zero. Inbreds are not shown as all are π = zero. The two rows under the figure indicate the test results for selection by HKA and CS analyses. The “-” indicates no significance; n = not tested in the HKA test due to inability to amplify a T. dactyloides orthologous sequence; I, improvement candidate; D, domestication candidate. The GenBank accession of the original Maize Mapping Project/DuPont unigene sequence is indicated under the bar.

The HKA test is valid, but the distribution of test statistics can be influenced by demographic history (Wright and Gaut, 2005). By contrast, CS analyses incorporate summary statistics, including information about recombination, to estimate the duration and severity of the bottleneck, based on data from the four reference genes. The CS then tests whether the loss of diversity in inbreds versus teosintes and landraces versus teosintes at a candidate locus is too great to be explained by demographic effects alone. By the CS analyses, a gene is considered a domestication gene if the gene was significant for selection in both landraces versus teosintes and inbreds versus teosintes, and a gene is inferred to be an improvement gene if inbreds versus teosintes, but not landraces versus teosintes, were significant for selection. Because the CS analysis directly tests for a major consequence of artificial selection, the specific loss of diversity at target genes, in addition to expected diversity loss from bottleneck effects on all genes, the CS analysis complements the HKA test as a second independent and powerful test for selected genes.

Note that for identification of an improvement gene, we did not apply the CS test directly between landraces to inbreds. An implicit assumption of the CS test is that the ancestral population follows a neutral equilibrium model. Although the teosinte population demonstrates some deviation from neutrality (Wright et al., 2005), the equilibrium neutral assumption is reasonable for any single gene in teosinte. By contrast, we cannot reasonably assume that the landraces fit a neutral equilibrium model because they have recently experienced a population bottleneck.

We identified six candidate genes for domestication and nine candidates for improvement by the CS analysis (Table 5, Figure 3). There were four domestication and four improvement genes identified in common between the HKA and CS analyses (Table 5, Figure 3). Because these eight genes were identified by two distinct tests for selection, one that relies on divergence information and the other that corrects for demographic history, they are our strongest candidates for selected genes.

Extended Sequencing of Candidate-Selected Genes

One limitation to our approach for identifying selected genes is that the short length of the alignment restricts the power of the approach. Longer sequences would increase the power to identify selection. Although the short alignments are expected to result in more false negatives rather than false positives, it is also important to characterize the false positive rate. We conducted extended sequencing in the maize inbreds and the teosintes for the entire available sequence of the eight genes identified by both HKA and CS analyses (Table 6, Figure 4; sequence alignments are presented in Supplemental Table 3 online).

Table 6.

Sequence Diversity in Maize Inbreds and Teosintes at Eight Candidate Genes and Results of the Tests of Selection

	Inbreds								Teosintes
Gene	N	L	S	h	πtotal	πsilent	P Value in HKAtotal	P Value in HKAsilent	N	L	S	h	πtotal	πsilent	P Value in HKAtotal	P Value in HKAsilent	Candidate Status	Homology Search
AY108876	14	1055	1	2	0.00025	0.00050	<0.0068**	<0.0120*	16	1026	13	8	0.00245	0.00403	<0.0433*	<0.1849	Selected	Amino acid transporter
AY107195	14	3119	1	2	0.00005	0.00007	<0.0058**	<0.0087**	11	3097	81	11	0.00820	0.01171	<0.5889	<0.6761	Selected	Auxin response factor
AY110109	14	1466	1	2	0.00036	0.00000	<0.0051**	<0.0054**	14	1355	43	11	0.00697	0.00999	<0.3613	<0.5321	Selected	GTP binding protein
AY105060	14	1090	0	1	0.00000	0.00000	<0.0041**	<0.0053**	15	1112	59	13	0.01566	0.01890	<0.7005	<0.7631	Selected
AY108178	14	1259	0	1	0.00000	0.00000	<0.0054**	<0.0082**	13	1224	54	10	0.01174	0.01458	<0.3233	<0.4719	Selected	Circadian clock
AY106616	14	2745	84	9	0.01132	0.01888	<0.4395	<0.2205	7	2619	97	7	0.01509	0.02446	<0.6859	<0.7214	Selected	Ankyrin repeat-like protein
AY107952	14	2469	23	5	0.00333	0.00446	<0.1193	<0.0927	14	2599	38	12	0.00390	0.00502	<0.1453	<0.1678	Selected	Putative fruit protein, oxidoreductase
AY106371	14	1574	4	5	0.00091	0.00000	<0.0094**	<0.0061**	15	1615	65	12	0.01078	0.01197	<0.4603	<0.4047	Selected	Putative methyl binding domain protein

Gene, the GenBank accession of the original unigene sequence; N, number of sampled sequences; L, length of the core alignments in which all sequences contain bases, excluding gaps; S, total number of segregating sites; h, number of unique sequences (haplotypes); π_total and π_silent, average proportion of pairwise differences per base pair at all sites and silent sites, respectively. HKA tests: P value of the candidate genes by multiple locus HKA against the four neutral genes by the maximum cell value test. HKA_total and HKA_silent: HKA test at all sites and silent sites, respectively. *, P < 0.05; **, P < 0.01. Candidate status: genes shown in bold are selected throughout the gene length, and genes not in bold are only selected at the 3′ region.

Figure 4.

Sliding-Window Analysis of the Genetic Diversity π in Maize Inbreds and Teosintes.

For sliding-window analysis, π was calculated for segments of 100 bp at 10-bp intervals. Horizontal and vertical axes on the graphs indicate DNA sequence position and genetic diversity π, respectively. Red lines indicate genetic diversity in the inbreds, whereas blue lines indicate genetic diversity in the teosintes. For gene structure under the sliding-window graphs, white bars indicate the predicted exons, and black lines indicate introns or genomic regions. Left arrows and right arrows indicate the positions for predicted start codons and stop codons, respectively. The lines with two arrows under the gene structure indicate the sequencing regions in our initial screening. (A) AY108876; (B) AY107195; (C) AY110109; (D) AY105060; (E) AY108178; (F) AY106616; (G) AY107952; (H) AY106371.

Five candidate genes, AY107195, AY110109, AY105060, AY108178, and AY106371, exhibited low nucleotide diversity in maize inbreds and high (normal) levels of nucleotide diversity in teosintes throughout the entire length of the gene (Table 6, Figure 4). The HKA tests at all sites (HKA_total) verified selection at these five candidates (P < 0.01), whereas HKA_total in teosintes was not significant. The relative ratios of π at all nucleotide sites (π_total) in inbreds versus teosintes ranged from zero to 0.08, indicating that the inbreds have lost >92% of genetic variation in the teosinte sample. Thus, these five genes exhibit evidence of selection throughout their length, and extended sequencing unambiguously demonstrates that these are selected genes.

The evidence for selection on the remaining three genes is less striking. For AY108876, HKA_total was significant in both teosintes and inbreds, but the HKA test at silent sites (HKA_silent; synonymous and noncoding positions) was significant for selection only in the inbreds (Table 6). One interpretation of these results is that this gene experienced selection in teosinte (as evidenced by low diversity in teosinte) and has undergone additional artificial selection (as evidenced by loss of silent site polymorphism from teosinte to maize).

For the remaining genes, AY106616 and AY107952, neither HKA_total nor HKA_silent were significant for selection of the entire gene. However, the entire second exon and 3′ untranslated region at AY106616 was significant for selection (Figure 4F; P_inbreds < 0.0218 and P_teosintes < 0.4169). Because the region of reduced polymorphism in this gene extends for more than 1 kb, the distribution of polymorphism for this gene is also strongly consistent with selection. For AY107952, two small exons exhibit marked decreases in polymorphism between maize and teosinte (Figure 4). For only this single gene out of eight is the decreased polymorphism limited to the initially sequenced region. In summary, for seven of our eight candidate genes, the extended sequencing clearly supports selection over genetic drift for the cause of the low inbred diversity, affirming the validity of our approach to identifying selected genes.

DISCUSSION

Our overarching goal in this article was to identify selected genes in maize because they represent candidates to contribute to agronomic traits. These genes will permit us to reconstruct the selection history of maize and provide novel candidate genes for maize improvement. To identify these genes, we first determined that the average genetic diversity π for 1095 maize genes was 0.0067. This result confirms that the average maize gene retains high levels of sequence polymorphism. Our sample of inbred maize contains more nucleotide diversity than species-wide samples of sorghum (Sorghum bicolor), soybean (Glycine max), Arabidopsis thaliana, or Arabidopsis lyrata (Arabidopsis Genome Initiative, 2000; Wright et al., 2003; Zhu et al., 2003; Hamblin et al., 2004; Ramos-Onsins et al., 2004; Wright and Gaut, 2005). In addition to providing the basis for identifying candidates for selection, our sequence data identifying 6169 SNPs in 1095 genes also provide the groundwork for the development of a public SNP mapping resource for maize. As all genes selected for sequencing were from the Maize Mapping Project/Dupont unigene set (see Methods), the physical and genetic location of the majority of these genes is known by association to anchored BACs of the maize fingerprint contig map (www.genome.arizona.edu/fpc/maize).

In choosing the panel of inbreds for sequencing, we intentionally established a comparison of genetic diversity between temperate and tropical maize lines. Our results indicate that the amount of genetic diversity contained within the two germplasm pools is similar. Using SSR markers, Liu et al. (2003) reported that tropical inbreds contained greater genetic diversity than temperate inbreds. This discrepancy may be more apparent than real because of the different sampling strategies in the two studies. First, our temperate sample uniquely contains both popcorn (Hp301) and sweet maize (Il14H) inbreds, expected to extend the diversity within the temperate sample. Second, while we sampled a larger number of loci than Liu et al. (2003), our shallow sampling of only seven lines each from temperate and tropical germplasm pools is primarily detecting SNP that occur at moderate to high frequency in maize. If tropical maize contains more rare SNP variants than temperate inbred lines, these rare SNPs would go largely undetected by our sampling strategy. However, our data clearly indicate that introducing tropical maize inbred lines into U.S. maize breeding programs will not greatly increase the genetic diversity of most maize genes over using diverse temperate inbred lines.

In a previous publication, we demonstrated that maize genes containing SSR sequences that were invariant among maize inbred lines, but polymorphic among teosintes, were enriched for selected genes (Vigouroux et al., 2002b). Although successful in finding candidates, SSR screening for selected genes has many limitations: only ∼10% of EST contigs for maize contain an SSR sequence (M.D. McMullen, unpublished data), and the high mutation rate of SSR may allow some recovery of diversity since domestication (Vigouroux et al., 2002a). For example, the promoter region of a maize domestication gene tb1 is a well-characterized region of reduced nucleotide variation, but an SSR in this region exhibits high diversity (Tenaillon et al., 2002). In this study, we have generalized the approach of random searches of low diversity genes to those identified by DNA sequence data from a large set of randomly chosen genes. This approach can be used to test any maize gene for a role in domestication and crop improvement.

In theory, diversity screens for identifying selected genes can be applied to any animal or plant domesticate, but the power of the approach depends critically on relative levels and patterns of diversity in neutral genes, selected genes, and genes in the wild taxon. If neutral genes retain very little diversity after domestication, it is difficult to discriminate neutral from selected genes without sequence fragments that are much longer than the 500-bp fragments used in this study. Sorghum may be one crop for which it will be difficult to identify selected genes without long sequence fragments because diversity levels in the domesticate are very low for all loci sampled to date (Hamblin et al., 2004). By contrast, maize has relatively high levels of polymorphism among plants (Wright and Gaut, 2005) so that screens for selection can be efficient with short sequences. Another important issue that will vary from taxon to taxon is the probability that selected genes were targets of selection or were hitchhiking with another target of selection. The distinction between target and hitchhiking depends critically on LD. In maize, LD generally decays very rapidly (Remington et al., 2001; Tenaillon et al., 2001), thus improving the possibility that an identified gene was a target rather than hitchhiked. Generally, self-pollinating species like soybean (Zhu et al., 2003) are expected to have high levels of LD, but it is important to note that this is not always true (Morrell et al., 2005). Thus, even self-pollinating taxa may be suitable species for diversity screens.

The HKA and CS analyses test selection on two separate selection criteria: the HKA test compares genetic diversity of the candidate gene with neutral genes and explicitly incorporates divergence information (Hudson et al., 1987), whereas the CS test examines reduction of diversity within a gene compared with its estimated demographic history (Tenaillon et al., 2004). Therefore, the two tests are complementary rather than redundant, and our strongest selection candidates are the four domestication and four improvement genes identified by both criteria (Table 5, Figure 3). The CS tests assigned three additional genes as improvement genes that could not be tested by HKA due to our inability to obtain an orthologous T. dactyloides sequence (Table 5, Figure 3).

We recognize at least four limitations in our analysis. First, the fact that average length of sequence alignment is short leads to a conservative test for selection. From Figure 3 it is clear that the candidate genes with no inbred diversity that failed tests of selection were of low diversity in the teosintes. There must be a high number of segregating sites in the teosintes before significant loss of diversity can be detected. Longer sequences would provide more power to test for selection in genes with lower genetic diversity levels in all three populations. However, our extended sequencing results clearly indicate that the short alignment length does not result in an appreciable false positive rate. For the eight genes identified by both the initial HKA and CS tests, the extended sequencing demonstrated that six genes were significant for selection throughout their sequence, and one additional gene was selected for an extended region. For only one of the eight genes, we could not distinguish between selection and genetic drift as the cause of the initial identification of selection based on a short alignment. Second, despite the rapid decay of LD in maize (Remington et al., 2001; Tenaillon et al., 2001), the genes we have identified may only be hitchhiking with neighboring selected genes. The selective sweep at tb1 extended only 60 to 90 kb upstream of the gene and does not contain other genes (Clark et al., 2004). However, for the Y1 locus, a gene under recent and strong selection in maize breeding programs, the effects of selection were evident up to 600 kb downstream of the target gene in the yellow endosperm subset of maize lines (Palaisa et al., 2004). The extent of the low diversity region around selected genes needs to be determined for many additional genes before definitive statements of selection can be made, but it is reasonable to expect that hitchhiked regions may be larger for improvement genes than domestication genes due to a more recent history of selection. However, it is interesting to note that for the six genes with evidence of selection throughout their sequence, three are domestication candidates (AY108876, AY105060, and AY106371) and three are improvement candidates (AY107195, AY110109, and AY108178). Third, if a causal site under selection within a gene, such as a single amino acid difference, is at moderate allele frequency in the population and at low LD with other polymorphisms within that gene before selection, this selected gene will remain polymorphic at the linked sites after selection and evade detection by either the HKA or CS analysis (Innan and Kim, 2004). Fourth, we acknowledge multiple test issues. However, the use of a standard correction such as the Bonferroni with genomic scans, such as this study, would result in an unacceptably high false negative rate, eliminating many biologically significant genes from further study (Storey and Tibshirani, 2003). Our extended sequencing empirically demonstrates that requiring candidates to be significant by both HKA and CS analyses provides a stringent selection criterion that resulted in a very low false positive rate.

A BLAST analysis was used to identify the protein function of the selected genes and to provide clues as to the traits under selection (Table 5). For the domestication genes, AY107952 has homology to fruit protein of kiwifruit (Ledger and Gardner, 1994). AY106371 encodes a putative methyl binding domain protein in maize. DNA methylation is a common factor in epigenetic gene regulation in plants (Martienssen, 1998; Bender, 2004) and may affect expression of an undetermined target gene. AY108876 has significant homology to an amino acid transporter, suggesting a role in amino acid synthesis or metabolism. Other genes for amino acid synthesis have also been identified as candidates for selection in maize (Wright et al., 2005), suggesting that nutritional quality has been a major target of human selection. This result complements the finding that many genes in the starch synthesis pathways in maize also show evidence of human selection (Whitt et al., 2002). AY105060 has no homology to known genes and proteins. One of the advantages of our scan is that genes of unknown function like AY105060 can only be identified as selection candidates by an unbiased approach such as used in this article. Of the improvement genes, AY107195 encodes an auxin response transcription factor (Ulmasov et al., 1997) and may have been selected for altered growth response. A second auxin response transcription factor, AY104948, was identified as a selected gene by Wright et al. (2005). Taken together, the data suggest that auxin-regulated growth responses may have been a major target of artificial selection for enhanced maize productivity. AY106616 encodes ankyrin repeat-like protein, which mediates protein–protein interactions. AY110109 encodes a GTP binding protein; this class of proteins is involved in signal transduction, cell differentiation, or membrane vesicle transport (Kang et al., 1995). AY108178 encodes an F-box protein with homology to circadian clock genes, ZTL and FKF1, controlling flowering time in Arabidopsis (Nelson et al., 2000; Somers et al., 2000) and may have been selected for an aspect of maize maturity. Another F-box protein, AY104147, was also identified as selected by Wright et al. (2005). From conducting multiple genomic searches for selection, key gene families and biological processes essential to maize improvement are becoming apparent.

Early studies on potential domestication genes in maize focused on the obvious morphological and developmental differences between maize and teosinte (Dorweiler et al., 1993; Hanson et al., 1996; Doebley et al., 1997). Recent studies have indicated that a number of the genes for enzymes in the starch synthesis pathway also exhibit signs of selection, demonstrating that selection has affected specific biochemical pathways (Whitt et al., 2002). However, performing individual tests for candidate genes postulated in controlling adaptive processes is laborious and time consuming. Taking an unbiased approach to the identification of selected genes is a more efficient path to understanding the genetic consequences of domestication and crop improvement. The results of this study suggest that genes controlling a wide range of traits may have been targets of selection (Table 5), and that our approach can also identify unknown function genes as important to domestication or improvement demonstrates an important advantage of our approach over a priori selection of candidates to test.

The productivity of crop species advances by the selection of favorable alleles at genes controlling the traits targeted by the plant breeder. However, as we have demonstrated, the genes that have undergone the most stringent selection (and thereby the greatest reduction in diversity) have little remaining genetic variation and cannot easily be further improved by standard plant breeding or even identified by QTL analysis because all inbred lines will have similar alleles. Indeed, where classical QTL mapping experiments between inbreds may fail to identify these genes as important genetic factors for agronomic traits, our strategy of genome screening for selected genes proved successful for detecting novel candidate domestication and improvement genes, providing new target genes for crop improvement. There are two approaches to reintroduce variation at these genes into maize breeding programs. The alleles from teosintes for domestication genes and from landraces for improvement genes could be introduced into maize breeding programs. It is specifically these genes that need to be added to maize breeding from exotic sources to broaden the genetic base of maize breeding efforts. Alternatively, transgenic alteration of the expression patterns of selected genes can be tested for desired effects on the relevant agronomic traits.

METHODS

Plant Materials

We used three diverse sets of maize (Zea mays) materials for DNA sequence analysis: inbred lines, landraces, and teosinte accessions (Table 1). The 14 maize inbreds, composed of seven temperate and seven tropical inbreds, were chosen based on SSR polymorphism data to maximize allelic diversity (Liu et al., 2003). The 16 maize landraces represent all areas in which maize was grown at the time of the discovery of the New World (Tenaillon et al., 2001). The 16 teosinte accessions were chosen based on geographic criteria to represent all areas where Z. mays subsp parviglumis is found. Each teosinte accession was self-pollinated twice or three times to derive partially inbred material. Two accessions of the wild relative, Tripsacum dactyloides (gammagrass), were used as an outgroup species for statistical analyses. The DNA was extracted with standard protocols (Saghai-Maroof et al., 1984) with minor modification.

Sequence Analysis

All genes for DNA sequencing were selected from the Maize Mapping Project/Dupont unigene set (http://www.agron.missouri.edu/files_dl/MMP/Cornsensus). The PCR primers were designed using the Primer3 program (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) to amplify 300 to 500 bp surrounding the 3′ untranslated region. The PCR was performed using Red Taq (Sigma-Aldrich) in a DNA Engine Tetrad thermocycler (MJ Research) with touchdown PCR (one cycle of 45 s at 95°C, 45 s at 65°C, and 55 s at 72°C; 1°C decrement in annealing temperature per cycle until annealing temperature is 55°C; then 25 cycles of 45 s at 95°C, 45 s at 55°C, and 1 min at 72°C). Following PCR amplification, unincorporated primers and deoxynucleotide triphosphates were removed by ethanol precipitation prior to sequencing. The PCR products were sequenced with each forward and reverse primer using BigDye terminator version 3.1 or dRhodamine terminator cycle sequencing kits (Applied Biosystems) and analyzed on ABI 3100 or 3700 sequencers (Applied Biosystems). The PCR products from T. dactyloides accession WW-2120 were cloned into pGEM-T vector (Promega), and eight clones were sequenced in both directions to eliminate Taq polymerase errors.

For extended sequencing of the eight selected genes, several sets of PCR primers were designed using the Primer3 program (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) to amplify 500- to 1000-bp products. The genomic PCR was performed using PCR Master Mix (Promega) or TaKaRa LA Taq with GC buffer (TaKaRa Bio) with a DNA Engine Tetrad thermocycler (one cycle of 2 min at 95°C; 35 cycles of 1 min at 95°C, 1 min at 50 to 65°C [dependent on primers] and 30 s to 3 min [dependent on PCR product] at 72°C; final extension of 3 to 5 min at 72°C). Following PCR amplification, unincorporated primers and deoxynucleotide triphosphates were removed by exonuclease I and shrimp alkaline phosphatase (United States Biochemical) and ethanol precipitation prior to sequencing. The PCR products were sequenced using BigDye terminator version 3.1 and analyzed on an ABI 3100 sequencer. The PCR products from two T. dactyloides accessions, WW-2120 and MIA 34597, were cloned into pGEM-T vector, and several clones were sequenced to eliminate Taq polymerase errors.

Base calling, quality assessment, and trimming of trace files were conducted with PHRED (Ewing and Green, 1998; Ewing et al., 1998), and sequence assembly was performed by PHRAP. The multiple sequences for each gene were aligned with ClustalW (Thompson et al., 1994) and edited manually.

Statistical Analyses

For each locus, two individuals, B73 and Mo17, were sequenced twice. The inclusion of replicates permitted assessment of empirical error rates. With an SNP identification criterion based on a PHRAP quality score >30, we found 45 mismatched sites between either B73 or Mo17 sequences. In total, 628,071 replicate nucleotide sites were compared, yielding an empirical error rate of ∼1 error in 7290 bp. However, the errors were nonrandomly clustered among loci. The 45 errors were found in only 11 loci, and five loci accounted for 90% of the errors. This distribution suggests that some of the errors may be due to sample switching. Such switching is not problematic for identifying SNPs and measuring diversity but could cause problems for association analyses. After removal of errors apparently due to sample switching, the sequencing error rate is essentially zero.

For population genetic analyses, sequences were removed if they had an average PHRAP quality score <30 or were <80% of the average length of the sequences in the alignment. Only loci with sequences from four or more individuals were analyzed for SNP identification and population genetic analysis. Only SNP variants with a PHRAP quality score >30 were used for analysis. For population genetic analyses of the candidate genes, we determined the number of polymorphic sites (S), the number of unique sequences (haplotypes; h), and the average proportion of pairwise nucleotide differences per nucleotide site (π ; Tajima, 1983) for each gene in DnaSP version 4.00 (Rozas et al., 2003). DnaSP was also used to determine Tajima's D (Tajima, 1989) for the four neutral genes. For sequence data comparisons, Mann-Whitney U, Student's t, Kruskal-Wallis H, and Scheffé's F tests were performed using Statcel (Yanai, 1998).

The HKA tests (Hudson et al., 1987) were conducted with adh1, bz2, fus6, and glb1 (Eyre-Walker et al., 1998; Hilton and Gaut, 1998; Tenaillon et al., 2001) as neutral control genes and with sequence from T. dactyloides representing the outgroup using HKA software (http://lifesci.rutgers.edu/∼heylab/HeylabSoftware.htm#HKA). The HKA test of selection is similar to a χ² test and asks whether the relative levels of intraspecific polymorphism and interspecific divergence for a locus are consistent across loci. The table for observation values was arranged for an interest gene and four neutral genes, including one for variation in a species (maize inbreds, landraces, or teosintes), one for variation in T. dactyloides, and one for the divergence between the species. Like a χ² test, two tables for expectations under a null model and standardized discrepancies between observations and expectations were generated. Rejection of the null hypothesis requires that the sum of standardized discrepancies between observations and expectations be greater than expected by chance (Hudson et al., 1987). If selection has occurred at the interest gene in a species, only one cell in the standardized discrepancy table may show a large value, and the overall test may not detect the discrepancy. The HKA software analyzes the single cell departure from the null hypothesis with multiple loci. In this outlier test, the test statistic is the maximum standardized discrepancy (MSD) observed for any polymorphism value. If the species has experienced a selective sweep at the interest locus, the standardized discrepancy for that observation will be high. This test statistic was then compared with a neutral distribution of the MSD that was generated by independent HKA coalescent simulations (Hudson et al., 1987). For each simulation, the MSD observed for any locus, for polymorphism in either species, is the MSD measure for that simulation and is recorded. If the observed MSD value is >95% of the entire frequency distribution of simulated MSD values, then the outlier test is significant (Wang and Hey, 1996). If this test was not possible due to low deviation for the interest gene, the overall P value based on simulated MSD distribution was used. Any gene in which the HKA tests indicated significance in teosintes was considered a low diversity gene for the teosintes and excluded from consideration for selection in maize.

Coalescent Simulations

CS tests were used to model the impact of a bottleneck on sequence diversity, using protocols detailed previously (Tenaillon et al., 2004). The method first estimates the demographic history of reference genes. Two demographic scenarios were estimated separately: a domestication bottleneck based on comparisons between teosinte and landrace data and an improvement bottleneck based on comparisons between teosinte and inbred data. For both scenarios, the maximum likelihood of the ratio k of bottleneck duration and strength was estimated with data from four reference genes (adh1, bz2, fus6, and glb1; Eyre-Walker et al., 1998; Hilton and Gaut, 1998; Tenaillon et al., 2001). For the landraces, the estimated neutral multilocus parameter was k = 4.65, and for the inbreds, the estimated parameter was k = 1.25. We also estimated the maximum likelihood of domestication and improvement bottlenecks for each candidate gene. The CS statistic is a likelihood ratio of the best fitting demographic scenario for the candidate gene against the multilocus reference estimate, and it tests whether the candidate gene has a history concordant with the reference loci at P = 0.05. For all likelihood estimation and all tests, initial simulation parameters and multilocus likelihoods were calculated as described previously (Tenaillon et al., 2004) except that (1) only Hudson's estimator (Hudson, 2001) was used as the recombination-population parameter in simulations and (2) k was explored over a grid of 32 values, with a fixed bottleneck duration of 1000 generations. Fixing the duration of the bottleneck does not affect estimation of k (Tenaillon et al., 2004). For all simulations, the goodness-of-fit statistic was the number of segregating sites in maize.

Homology Search

We utilized both nucleotide–nucleotide (BLASTN) and translated query versus protein database (BLASTX) BLAST routines (http://www.ncbi.nlm.nih.gov/BLAST) to identify protein functions for candidate genes. All reported homologies exceed a 1e⁻²⁰ threshold for at least one of the two BLAST routines. The gene prediction was performed using the automated annotation system RiceGAAS (Sakata et al., 2002; http://ricegaas.dna.affrc.go.jp/).

Accession Numbers

All sequence data from this article were deposited in GenBank (BV106362 to BV123527), and all alignments are available from www.panzea.org. The 35 initial sequence alignments and the eight extended sequencing alignments are also available in Supplemental Tables 2 and 3 online.

Supplemental Data

The following materials are available in the online version of this article.

• Supplemental Table 1. Sequence Diversity in Temperate and Tropical Maize Inbreds for 1095 Maize Unigenes.

• Supplemental Table 2. Sequence Alignments in Maize Inbreds, Landraces, Teosintes, and Tripsacum at 35 Genes That Exhibited Zero Sequence Diversity in Maize Inbreds.

• Supplemental Table 3. Extended Sequence Alignments in Maize Inbreds, Teosintes, and Tripsacum at Eight Candidate Genes.