Genomic architecture of biomass heterosisin ArabidopsisMei Yanga,1, Xuncheng Wanga,1, Diqiu Rena, Hao Huanga, Miqi Xua, Guangming Hea,2, and Xing Wang Denga,2

aState Key Laboratory of Protein and Plant Gene Research, Peking–Tsinghua Center for Life Sciences, School of Advanced Agricultural Sciences and School ofLife Sciences, Peking University, Beijing 100871, China

Contributed by Xing Wang Deng, June 14, 2017 (sent for review April 4, 2017; reviewed by James A. Birchler and Zhongfu Ni)

Heterosis is most frequently manifested by the substantially increasedvigorous growth of hybrids comparedwith their parents. Investigatinggenomic variations in natural populations is essential to understandthe initial molecular mechanisms underlying heterosis in plants. Here,we characterized the genomic architecture associated with biomassheterosis in 200 Arabidopsis hybrids. The genome-wide heterozygosityof hybrids makes a limited contribution to biomass heterosis, and nolocus shows an obvious overdominance effect in hybrids. However,the accumulation of significant genetic loci identified in genome-wide association studies (GWAS) in hybrids strongly correlates withbetter-parent heterosis (BPH). Candidate genes for biomass BPH fallinto diverse biological functions, including cellular, metabolic, and de-velopmental processes and stimulus-responsive pathways. Importantheterosis candidates include WUSCHEL, ARGOS, and some genes thatencode key factors involved in cell cycle regulation. Interestingly, tran-scriptomic analyses in representative Arabidopsis hybrid combinationsreveal that heterosis candidate genes are functionally enriched instimulus-responsive pathways, including responses to biotic and abi-otic stimuli and immune responses. In addition, stimulus-responsivegenes are repressed to low-parent levels in hybrids with high BPH,whereas middle-parent expression patterns are exhibited in hybridswith no BPH. Our study reveals a genomic architecture for understand-ing the molecular mechanisms of biomass heterosis in Arabidopsis, inwhich the accumulation of the superior alleles of genes involved inmetabolic and cellular processes improve the development and growthof hybrids, whereas the overall repressed expression of stimulus-responsive genes prioritizes growth over responding to environmentalstimuli in hybrids under normal conditions.

biomass heterosis | GWAS | natural variation | Arabidopsis

Heterosis, also known as hybrid vigor, refers to the biologicalphenomenon that the hybrid progeny exhibits superior

performance over its parents in many traits, such as biomass,growth rate, yield, and fitness (1‒4). Three quantitative geneticmodels, namely dominance (5), overdominance (6, 7), and epis-tasis (8), are largely conceptual and do not explain the molecularmechanism of heterosis. Although omics studies have describedgenome-wide changes in gene expression (9‒13), small RNAs(14), DNA methylation (15), and histone modifications (12, 13)between hybrids and parents in plants, the underlying geneticmechanism remains elusive. With the increased availability ofgenome sequences and the revolution in computational methods,genome-wide association studies (GWAS) has been developedinto a powerful tool to explore the genetic loci and candidategenes responsible for traits in plants (16‒20). Using this approach,numerous superior alleles contributing to yield-related heterosishave been identified in rice (21). The mapping of heterosisquantitative trait loci (QTL) for yield in rice hybrids has recentlybeen reported (22). These studies have greatly improved thecurrent understanding of the genetic bases of heterosis in plants.However, genetic loci associated with heterosis may not be com-pletely preserved in rice, as natural genetic diversity decreasesconstantly during rice domestication (23). By contrast, explorationin native plant species that have not undergone domestication mayhelp uncover the full spectrum of the genetic basis of heterosis.

Arabidopsis, an undomesticated plant species, serves as an ex-cellent model to study the genetic mechanism of heterosis, owingto its short life cycle, extensive naturally occurring genetic varia-tion, and sufficient amount of heterosis (24‒27). To date, fewgenetic loci associated with heterosis have been identified throughQTL mapping in Arabidopsis (28‒30). Remarkably, a recentGWAS performed in Arabidopsis hybrids, generated by inter-crossing 30 accessions, discovered significant loci and candidategenes for hybrid performance in flowering time and rosette traits(31), which demonstrated the feasibility of GWAS to dissect thegenetic architecture of heterosis in Arabidopsis. However, in thiscase, limited genetic loci can be detected because sequence di-vergence among 30 accessions represented only a small proportionof the natural genetic variation in Arabidopsis.Here, we generated 200 Arabidopsis hybrids by crossing

Columbia-0 (Col-0) with other natural accessions collected world-wide and phenotyped these hybrids together with their parents forbiomass-related traits during early development. We conductedGWAS for biomass heterosis and discovered 750 associated single-nucleotide polymorphisms (SNPs) that showed collective contri-butions. We identified 779 candidate heterosis genes, some ofwhich encoded key regulators of the cell cycle and plant growthand development. By performing a transcriptomic analysis inrepresentative hybrids, we found that stimulus-responsive geneswere highly overrepresented among candidates and exhibited an

Significance

Heterosis, the phenotypic superiority of a hybrid over its par-ents, has been extensively exploited in agriculture to improvebiomass and yield. Despite its great agricultural importance, thegenetic components underlying heterosis remain largely unclear.Here, we characterize the genomic architecture of heterosis inArabidopsis that have not undergone domestication and iden-tify hundreds of genetic loci that collectively contribute to bio-mass heterosis using genome-wide association studies. Thefunctional investigation of candidate genes and transcriptomicanalysis in representative hybrids suggest that the accumulationof superior genes involved in basic biological processes and therepression of stimulus-responsive genes in hybrids contribute tobiomass heterosis in Arabidopsis, thus providing a comprehen-sive understanding of the genetic bases of heterosis in naturalpopulations of plant species.

Author contributions: G.H. and X.W.D. designed research; M.Y., D.R., and M.X. performedresearch; M.Y., X.W., and G.H. analyzed data; and M.Y., X.W., H.H., G.H., and X.W.D.wrote the paper.

Reviewers: J.A.B., University of Missouri; and Z.N., China Agricultural University.

The authors declare no conflict of interest.

Data deposition: All original data sets have been deposited in the Gene Expression Omnibus(GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession nos. GSE85759 and GSE100595.1M.Y. and X.W. contributed equally to this work.2To whom correspondence may be addressed. Email: deng@pku.edu.cn or heguangming@pku.edu.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1705423114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1705423114 PNAS | July 25, 2017 | vol. 114 | no. 30 | 8101–8106

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

July

30,

202

1

overall repression in hybrids with high biomass heterosis. Takentogether, these data provide comprehensive insights into the ge-netic bases of biomass heterosis in Arabidopsis.

ResultsCollective Contribution of Different Growth Traits to Biomass Heterosis inArabidopsis. Previous studies have suggested that the degree ofheterosis observed in hybrids is proportional to the parental geneticdistance (3). In this scenario, if multiple A. thaliana accessions arecrossed with a common maternal accession, then the biomass het-erosis in these hybrids should be positively correlated with the pa-rental genetic distance, and the differences in biomass heterosisamong hybrids should result from the sequence divergence of pa-ternal accessions. This correlation may provide clues to revealingthe genetic basis of biomass heterosis. Considering this hypothesis,we generated 200 intraspecific hybrids by crossing 200 A. thalianaaccessions with one common maternal accession, Col-0 (DatasetS1). The collected 201 accessions originated from Europe, Asia, andNorth America and displayed a broad range of geographic diversity(Fig. 1A). The hybrids and their parents were phenotyped for fourtraits, with one trait corresponding to shoot biomass (represented asshoot fresh weight) and three traits that potentially contributed toshoot biomass (leaf number, leaf area, and rosette diameter), at14 d after sowing (DAS) to avoid the impact of flowering on bio-mass heterosis (SI Appendix, Fig. S1). We observed strong positivecorrelations among biomass, leaf area, and rosette diameter in hy-brids (correlation coefficient r = 0.769–0.847; SI Appendix, TableS1), indicating that both leaf area and rosette diameter were themain contributors to hybrid biomass. Whereas the leaf numberalso partially contributed to biomass in hybrids (r = 0.435), no

correlation was detected between the leaf number and leaf area orrosette diameter (r = 0.051 and r = 0.06, respectively) (SI Appendix,Table S1). These data suggested that in hybrids, the genetic com-ponents associated with leaf area or rosette diameter are distinctfrom those associated with leaf number. Furthermore, comparedbetween hybrids and paternal accessions for each trait, significantlypositive correlations were detected (Fig. 1B and SI Appendix, Fig.S2), suggesting that the genetic variations among paternal acces-sions are associated with differences in growth vigor in hybrids.Better-parent heterosis (BPH) and middle-parent heterosis

(MPH) describe the degree of phenotypic difference between ahybrid and its better parent and between a hybrid and the meanof two parents, respectively. We used BPH and MPH to evaluatethe heterosis of the 200 hybrids and found that heterosis widelyoccurred during early Arabidopsis development. All 200 hybridsshowed positive MPH, and no hybrid was inferior to its betterparent for all four traits (SI Appendix, Fig. S3). Most hybrids(98%, 196) showed significant positive BPH for all four traits,although with large variation (Fig. 1C and SI Appendix, Fig. S3).Further analyses focused primarily on BPH. We found that bothbiomass and leaf area heterosis were higher than leaf number orrosette diameter heterosis (Fig. 1D). Furthermore, the BPH forbiomass, leaf area, and rosette diameter were highly correlatedwith each other (r = 0.859–0.886; Fig. 1E and SI Appendix, Table

Leaf

are

a_F 1

(mm

2 )

Leaf area_paternal (mm2)15 20 2510 5

3530

15

2520

40

10

r = 0.657

B

A

D E

-100 -50 0 50 100 150-150

0.0 0.5 1.075

50

25

Beijing

Alaska

Longitude

Latit

ude

Finland

Col-0New York

80

60

0

40

20

C

Num

ber o

f F1

hybr

ids

60 100 140 20 -20 Biomass BPH (%)

150

5010

00

BP

H (%

)

LA RD BMLN

150

5010

00

80 120 40 0Leaf area BPH (%)

r = 0.886

Bio

mas

sB

PH

(%)

Fig. 1. Heterosis in 200 Arabidopsis hybrids. (A) Geographic distribution of201 A. thaliana accessions used in this study. Each dot indicates the originalisolation location for an accession. The colors represent the levels of biomassheterosis, except for red, which represents the common maternal accessionCol-0. (B) High correlation of leaf area between paternal accessions andF1 hybrids. (C) BPH for biomass in 200 Arabidopsis hybrids at 14 DAS.(D) Comparison among heterosis for different traits in 200 Arabidopsis hy-brids. LN, LA, RD, and BM represent leaf number, leaf area, rosette diameter,and biomass, respectively. (E) BPH for biomass strongly and positively cor-relates with BPH for leaf area in 200 Arabidopsis hybrids.

-400

4080

120160200

0.1 0.15 0.2 0.25

Bio

mas

s B

PH

(%)

Genetic distance

r = 0.025P = 0.73

-400

4080

120160200

2 4 6 8

Bio

mas

s B

PH

(%)

SNP density (1 kb)

r = 0.151P = 0.036

A

C

B

DP < 2.2e-16

AssociatedSNPs

Random SNPs

2

6

4

0

Chromosome3 4 5 2 1

-log 1

0P

10

0

5

-5 Z (F

ST)

_Bio

mas

s

E

-400

4080

120160200

0 100 200 300 400

Bio

mas

s B

PH

(%)

Number of associated SNPs

r = 0.558P < 2.2e-16

Fig. 2. Characterization of genomic features associated with biomass heter-osis in Arabidopsis. (A) No correlation is detected between BPH for biomassand parental genetic distance. Parental genetic distance was calculated be-tween Col-0 and the paternal accessions using PLINK’s IBD analysis. (B) A weakbut significant positive correlation is detected between BPH for biomass andgenome heterozygosity in Arabidopsis hybrids. The genome heterozygosity ofhybrids was represented by SNP density per kilobase between two parents.(C) Manhattan plot of GWAS for biomass heterosis. Negative log10-trans-formed P values from a genome-wide scan were plotted against positions oneach of the five Arabidopsis chromosomes. (D) BPH for biomass strongly andpositively correlates with the number of 750 associated SNPs accumulatedin paternal accessions. Associated SNPs were identified using a Benjamini–Hochberg test with a threshold of 0.2. (E) Z (FST) of 10-kb genomic regionssurrounding 750 associated SNPs between the high-BPH group and the low-BPH group for biomass is significantly higher than control (random SNPs).

8102 | www.pnas.org/cgi/doi/10.1073/pnas.1705423114 Yang et al.

Dow

nloa

ded

by g

uest

on

July

30,

202

1

S2). Interestingly, although leaf number was not correlated withleaf area or rosette diameter in hybrids, the BPH for leaf numberwas significantly correlated with that for leaf area or rosette di-ameter (r = 0.48 and r = 0.494, respectively) and that for biomass(r = 0.598) (SI Appendix, Table S2). These results indicated apotential common genetic mechanism underlying the heterosisof different traits and the collective contribution of differenttraits to biomass heterosis in Arabidopsis.

Genome-Wide Heterozygosity of Hybrids Makes a Limited Contributionto Biomass Heterosis. According to the observation that leaf areacontributed most significantly to biomass heterosis during earlyArabidopsis development (Fig. 1E), further analyses focused onheterosis for these two traits. We observed no obvious correlationbetween the extent of biomass heterosis and the geographic dis-tance of paternal accessions, and many hybrids of the paternal ac-cessions from nearby locations showed significantly different levelsof heterosis (Fig. 1A). To further determine whether the geneticdistance was correlated with the heterosis of hybrid populations inthis study, a total of 722,000 SNPs with high reliability for 191 of the200 paternal accessions obtained from the 1001 Genomes Projectwere used for the pairwise analysis of the genetic distance betweenCol-0 and each of the 191 paternal accessions using PLINK’s identityby descent (IBD) analysis. We found that although the paternalaccessions used in this study had extensive genetic diversity relativeto Col-0, no correlation between parental genetic distance and het-erosis for biomass or leaf area in hybrids was detected (Fig. 2A andSI Appendix, Fig. S4A). Thus, these results suggested that the degreeof genetic distance between parents does not necessarily contributeto heterosis in intraspecific hybrids of Arabidopsis and that the nat-ural genetic variation associated with microenvironmental di-vergence may contribute to biomass heterosis in Arabidopsis.A prerequisite for heterosis manifestation is the genetic het-

erozygosity resulting from divergence between parental lines.Accordingly, we further investigated the effect of hybrid genomeheterozygosity on heterosis. We used SNP density (SNPs perkilobase) to represent hybrid genome heterozygosity. Notably, aweak but significant positive correlation was observed betweenthe genome-wide heterozygosity of hybrids and heterosis forbiomass and leaf area (Fig. 2B and SI Appendix, Fig. S4B). Thesedata indicated that not all heterozygous sites in hybrids derivedfrom parental genomic divergence are involved in biomass het-erosis, and heterosis should be contributed by the heterozygosityat specific genetic loci in hybrids.

Accumulation of Associated GWAS Loci Correlates with BiomassHeterosis in Arabidopsis. To identify loci or genes with the poten-tial contribution to Arabidopsis biomass heterosis, we conductedGWAS on heterosis for biomass and leaf area using an algorithmfor multivariate linear mixed models in Genome-Wide EfficientMixed-Model Association (GEMMA) software (32). No clearsignals resembling a peak were observed in the resulting Manhat-tan plot (Fig. 2C), indicating that Arabidopsis biomass heterosisresults from many alleles with additive effects. Using a modestsignificance threshold [false discovery rate (FDR) < 0.2], 750 as-sociated SNPs were identified. Remarkably, heterosis strongly andpositively correlated with the accumulation of associated SNPs inthe paternal accessions (Fig. 2D and SI Appendix, Fig. S5A), butnot with accumulated SNPs that were randomly selected from ourSNP library (SI Appendix, Fig. S5 B and C). Therefore, the com-bination of heterozygous loci containing these 750 associated SNPsin hybrids may contribute to biomass heterosis in Arabidopsis.As linkage disequilibrium (LD) decays within ∼10 kb in Ara-

bidopsis (33), we focused our analyses on 10-kb genomic regionsaround these 750 associated SNPs. For convenience of analysis,we divided the paternal accessions into two groups: the high-BPHgroup consisted of the top third of accessions ranked according toBPH for biomass or leaf area, and the low-BPH group consisted ofthe bottom third of these ranked accessions (SI Appendix, Fig.S6A). The 10-kb genomic regions around these 750 SNPs exhibi-ted extensive sequence variation between the high-BPH and low-BPH groups, which was reflected by the higher population-differentiation statistic (FST) than control (Wilcoxon’s rank–sumtest, P < 2.2 × 10−16) (Fig. 2E and SI Appendix, Fig. S6B). Thisextensive sequence variation may provide the basis for the col-lective contribution of heterozygous loci containing 750 associatedSNPs to biomass heterosis in Arabidopsis.

Identification and Functional Characterization of Candidate Genes forBiomass Heterosis in Arabidopsis. Based on the Arabidopsis genomeannotation in TAIR, a total of 779 protein-coding genes within the10-kb genomic regions surrounding 750 heterosis-associated SNPswere identified and thought to be candidate genes for biomass BPHin Arabidopsis (Dataset S2). The biological function of these can-didates was characterized based on GO (gene ontology) annota-tions. Of the 779 genes, 474 genes had functional annotations forbiological process categories in the ArabidopsisGO database. Thesegenes were divided into diverse functional categories, includingcellular processes, metabolic processes, response to stimulus, bi-ological regulation, and developmental processes (SI Appendix, Fig.

Col-0 Per-1Col-0 x Per-1 B

05

10152025

Bio

mas

s (m

g) **A

Aa-0Col-0 x Aa-0Col-0

D Total SNPs

119150 85907 67235

Col-0 x Per-1 Col-0 x Aa-0

Associated SNPs

182 106 6

Missense SNPs

55236 37360 39308

E

0 3 6 9 12 150 2 4 6 8 10

phosphorus metabolic processphosphate metabolic process

protein modification processpost-translational protein modification

oxidation reductionsignaling

phosphorylationprotein amino acid phosphorylation

defense response

-log10 (P value)

Col-0 x Per-1 Col-0 x Aa-0GO terms

C

Fig. 3. The phenotypes and genotypes of represen-tative hybrids with the highest BPH or the lowest BPH.(A) Growth vigor of two representative Arabidopsishybrids, Col-0 × Per-1 and Col-0 × Aa-0, at 14 DAS.(Bar, 1 cm.) (B) Col-0 × Per-1 exhibits the highest BPHfor biomass, whereas Col-0 ×Aa-0 shows no significantBPH for biomass. The data are presented as themeans ± SD; n > 30. **P < 0.01 between the hybridsand parents (Student’s t test). (C) Silhouette imagesshowing differences in the size of the first leaf ofplants in A. (Bar, 0.5 cm.) (D) Col-0 × Per-1 accumulatesmore total, missense, and GWAS-associated SNPs thanCol-0 × Aa-0. (E) Genes with multiple missense SNPs(>10) are significantly enriched in defense pathways,which were more evident in Col-0 × Per-1.

Yang et al. PNAS | July 25, 2017 | vol. 114 | no. 30 | 8103

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

July

30,

202

1

S7), indicating the collective contribution of these biological path-ways to biomass heterosis in Arabidopsis.Further detailed inspection revealed that several genes encoding

key regulators of cell cycle processes, including CELL DIVISIONCYCLE 20.1 (CDC20.1), CELL DIVISION CYCLE 20.2 (CDC20.2),CYCLIN D1;1 (CYCD1;1), CYCLIN D2;1 (CYCD2;1), CYCLIN P2;1(CYCP2;1), E2F TRANSCRIPTION FACTOR 1 (E2F1), DP-E2F-LIKE PROTEIN 3 (DEL3), HOBBIT (HBT) and KIP-RELATEDPROTEIN 2 (KRP2), were among the candidate heterosis genes.Notably, WUSCHEL (WUS), a gene required to control the stemcell pool in the shoot apical meristem (SAM), and AUXIN-REGULATED GENE INVOLVED IN ORGAN SIZE (ARGOS)were also in the candidate list. Other important candidate genesinvolved in plant growth and development included INDOLE-3-ACETIC ACID INDUCIBLE 28 (IAA28), a negative regulator ofauxin signaling, and GA REQUIRING 3 (GA3), which plays anessential role in the gibberellin biosynthetic pathway (Dataset S2).The natural divergence of these genes may occur during the evo-lution of Arabidopsis, and the accumulation of their superior allelescould be important contributors to the growth vigor in hybrids.

Candidate Genes for High BPH Are Enriched in Stimulus-ResponsivePathways. We further characterized heterosis candidates from agene expression aspect. Because leaf area heterosis and biomassheterosis were highly correlated, the candidate genes underlying

biomass heterosis should be expressed in the leaves. Accordingly,we narrowed the candidate genes for biomass heterosis by inves-tigating the expression of these genes in representative Arabidopsishybrid combinations. The leaves of two hybrid combinations,Col-0 × Per-1 and Col-0 × Aa-0, were selected for whole-genomeexpression profile analyses. Col-0 × Per-1 exhibited one of thehighest levels for biomass BPH (139.4%) and leaf area BPH(129.4%), whereas Col-0 ×Aa-0 showed one of the lowest levels forbiomass BPH (−2.4%) and leaf area BPH (−7.3%) (Fig. 3 A–C).Comparison of the genotypes between Col-0 and Per-1 orAa-0 showed that Col-0 × Per-1 contained more total and missenseSNPs and accumulated more heterosis-associated SNPs than didCol-0 × Aa-0 (Fig. 3D). Interestingly, genes with multiple missenseSNPs were significantly enriched in defense response pathways,which were much more evident in Col-0 × Per-1 (Fig. 3E). Of the779 genes adjacent to heterosis-associated SNPs, 453 genes weredetectably expressed in the leaves of at least one genotype inCol-0 × Per-1 and were considered candidate genes for biomassheterosis in this hybrid. Further GO analysis showed that thesecandidate genes were significantly enriched in response to stimuluspathways (153 genes), including immune responses, responses toboth biotic and abiotic stimuli, and responses to stress (defenseresponses) (Fig. 4A). These results suggested that the natural ge-netic variation in stimulus-responsive genes might be associatedwith biomass heterosis in Arabidopsis.

GO:0050896 (5.53e-05) BA

GO:0006955 (0.000405) immune response

GO:0009628 (0.00623) response to

abiotic stimulus

response to stimulus

GO:0009607 (0.00183) response to

biotic stimulus

GO:0006950 (0.0309) response to stress

GO:0045087 (0.000405)

GO:0051707 (0.00254) response to

other organism

GO:0006952 (0.0448) defense response

BLP LP MP HP AHP

Sig

nific

ance

leve

l

Low

High

is a

GO:0009814 (0.0309) defense response,

incompatible interaction

GO:0009627 (0.00551) systemic acquired

GO:0045087 (0.000405) innate immune

response

GO terms P1 > P2 P1 < P2 Up Down P1 > P2 P1 < P2 Up Down

response to stimulus 4.5e-05 5.4e-26 2.9e-25 1.6e-39

immune response 6.1e-34 3.1e-21 1.3e-32

response to stress 2.8e-06 5.3e-26 2.0e-32 2.4e-44 0.023

0-aA x 0-loC1-reP x 0-loC

60

80

100

erce

nt (%

)

Up DownCol-0 x Per-1 Col-0 x Aa-0

gsystemic acquired resistance

DC

defense response 1.5e-04 9.6e-42 2.0e-22 9.8e-53

response to abiotic stimulus 0.0093 0.0035 1.4e-12 6.0e-11 0.023

response to biotic stilumus 3.5e-32 1.5e-23 6.1e-320

20

40

Gen

e pe

P1 > P2 P1 < P2 P1 > P2 P1 < P2 Significance level Low High not significant

Fig. 4. Distinct expression patterns of stimulus-responsive genes between Col-0 × Per-1 and Col-0 × Aa-0. (A) GO analysis of 453 candidate genes for biomassheterosis, which were significantly enriched in response to stimuli. (B) Candidate genes exhibiting low-parent expression (LP) or below–low-parent expression(BLP) and high-parent expression (HP) or above–high-parent expression (AHP) in Col-0 × Per-1 largely show middle-parent expression (MP) in Col-0 × Aa-0.(C) Genome-wide comparison of expression patterns between Col-0 × Per-1 and Col-0 × Aa-0. P1 in Col-0 × Per-1 and Col-0 × Aa-0 represents Col-0; P2 inCol-0 × Per-1 and Col-0 × Aa-0 represents Per-1 and Aa-0, respectively. Up, up-regulation in hybrid; Down, down-regulation in hybrid. (D) GO analysis of geneswith expression variation in Col-0 × Per-1 and Col-0 × Aa-0.

8104 | www.pnas.org/cgi/doi/10.1073/pnas.1705423114 Yang et al.

Dow

nloa

ded

by g

uest

on

July

30,

202

1

Overall Repression of Stimulus-Responsive Genes in Hybrids with HighBiomass Heterosis. To gain more in-depth insight into how candidategenes contribute to Arabidopsis biomass heterosis, we compared theexpression patterns of these 453 candidate genes between Col-0 ×Per-1 and Col-0 ×Aa-0. Notably, most candidate genes showing high-or low-parent expression in Col-0 × Per-1 were expressed at themiddle-parent level in Col-0 × Aa-0 (Fig. 4B and SI Appendix, Fig.S8). This result further demonstrated that candidate genes exhibiteddistinct expression patterns between hybrids with different degrees ofbiomass heterosis. Most importantly, candidate genes displaying low-parent or below–low-parent expression in Col-0 × Per-1 were over-represented in response to stimuli (P = 4.6 × 10−4). However, can-didate genes expressed at high-parent or above–high-parent levels inCol-0 × Per-1 did not exhibit GO enrichment. These data indicated arole for stimulus-responsive genes in the establishment of biomassheterosis in Arabidopsis. This conclusion was further substantiated bya comparative analysis of variations in the expression patterns ofstimulus-responsive genes at the genome-wide level between Col-0 ×Per-1 and Col-0 × Aa-0 hybrids. In the Col-0 × Per-1 combination,regardless of the direction of expression divergence between parents,gene expression was predominantly down-regulated in the hybridplants. By contrast, in the Col-0 × Aa-0 combination, only genes inone direction of divergence between parents showed down-regulationin the hybrid (Fig. 4C and SI Appendix, Fig. S9). Further GO en-richment analysis between parents revealed that genes showinghigher expression in Col-0 or in Per-1 were significantly enriched instimulus-responsive pathways and that these response genes weredown-regulated in the Col-0 × Per-1 hybrid (Fig. 4D). Nevertheless,only genes with higher expression levels in Col-0 than in Aa-0 weresignificantly enriched in stimulus responses, and these response geneswere only partially down-regulated in the Col-0 × Aa-0 hybrid. No-tably, in both combinations, genes showing up-regulation in hybridswere not enriched in stimulus responses (Fig. 4D). Transcriptomicanalysis of another hybrid Col-0 × Ak-1 with high biomass BPHrevealed similar stimulus-responsive gene expression patterns (SIAppendix, Fig. S10). These data suggested that the bidirectional di-vergent expression of stimulus-responsive genes between parents andtheir overall down-regulation in hybrids contribute to better-parentbiomass heterosis in Arabidopsis.

DiscussionIn this study, we performed GWAS for biomass heterosis establishedduring early development in Arabidopsis and identified 750 associ-ated SNPs showing collective contribution. A total of 779 candidategenes around the 10-kb genomic regions of heterosis-associatedSNPs were identified and implicated in diverse biological func-tions. One of the most important candidate genes wasWUS, which isexpressed in the stem-cell–organizing center of meristems andfunctions in controlling the size of the stem cell population (34). Itwill be intriguing to experimentally investigate whether this genecontributes to biomass heterosis by inducing larger SAM in hybrids.Interestingly, some candidate genes encoded key regulators of thecell cycle. CYCD1;1, CYCD2;1, CYCP2;1, E2F1, DEL3, andKRP2 are critical factors for the regulation of the G1-to-S transitionin the cell cycle (35, 36). In addition, HBT is a core component ofthe anaphase-promoting complex (APC), an E3 ubiquitin ligase thatis activated by CDC20 during the G2-to-M transition (35, 36). Be-cause the cell cycle plays a critical role in regulating organ growthand size (37, 38), and greater leaf growth in the Arabidopsis hybridwas tightly associated with increased cell number (39), which mayreflect the acceleration of the cell cycle, the cell cycle-relatedcandidate genes identified in this study may make importantcontributions to biomass heterosis in Arabidopsis. Furthermore,some phytohormone-responsive genes involved in the regulationof cell cycle processes were identified as biomass heterosis genecandidates. For example, ARGOS, an auxin-inducible gene, con-trols Arabidopsis leaf size by accelerating the cell cycle (40). Theexpression of ARGOS was activated in Arabidopsis hybrids (39),

and the overexpression of the maize ortholog of ARGOS, ZmAR-GOS1, affected the hybrid yield (41).Through a transcriptomic analysis, heterosis candidates for a

representative hybrid with high BPH were narrowed to 453 genes,which displayed significant enrichment in response to biotic andabiotic stimuli and immune processes. Remarkably, the stimulus-responsive genes showed distinct expression patterns betweenCol-0 × Per-1 with the highest heterosis and Col-0 × Aa-0 with thelowest heterosis and exhibited overall repression in Col-0 × Per-1.Miller et al. suggested the effect of stress response genes on growthheterosis in Arabidopsis by comparative transcriptomic analysis inhybrids (10). Our results provided evidence from both genetic andtranscriptional aspects for the contribution of stimulus-responsivegenes to biomass heterosis in Arabidopsis. Notably, stimulus-responsive genes, particularly defense response genes, are highlydivergent in sequence among natural Arabidopsis accessions. Thus,although the candidate genes for biomass heterosis are enriched instimulus-responsive pathways, a specific set of genes involved inthese pathways may vary depending on hybrid combinations, whichshould be identified by additional analyses of the mRNA expressionin different organs and at different developmental stages.Recent studies have uncovered the tradeoff between stress re-

sponses and growth in Arabidopsis (42‒45). Therefore, Arabidopsisbiomass heterosis may result from the divergence of stimulus-responsive genes between parents and the overall repression ofthese genes in hybrids (Fig. 5). This mechanism should be in-dependent of plant species, organs, or developmental stages. Po-tential evidence is that BPH for both leaf number and leaf area ispositively correlated and contributes to biomass BPH, whereas thetwo traits themselves are not correlated in hybrids (SI Appendix,Tables S1 and S2). The regulatory mechanisms for the repressionof stimulus-responsive genes may involve the parental divergenceof genetic components, such as missense variation in repressors forstimulus-responsive pathways. Epigenetic components, includingsmall RNAs, may also be involved. Indeed, several key genes in-volved in small-RNA biogenesis and functional processes, in-cluding ARGONAUTE 1 (AGO1), DICER-LIKE 4 (DCL4), andHUA ENHANCER 1 (HEN1), were among the candidate genesidentified in this study (Dataset S2). Notably, a previous study hasdemonstrated an important role for HEN1 in biomass heterosis inArabidopsis (15).These findings differed from those of a recent GWAS in

Arabidopsis hybrids that were generated after intercrossing

R1

R2 G2

G1

P1

R2

R1 G1

G2

P2

F1

R1R2

G2

G1R1 and R2, Response

genes group1 and 2

G1 and G2, Growth

associated with R1 and R2

Fig. 5. A model for biomass heterosis in Arabidopsis. The expression levels ofone group of stimulus-responsive genes (R1) are lower in one parent (P1) thanin another parent (P2), associated with stronger growth in P1 than in P2(represented by G1). The expression levels of another group of stimulus-responsive genes (R2) are higher in P1 than in P2, associated with weakergrowth in P1 than in P2 (represented by G2). The growth and response arebalanced in both parents. The differential expression levels of two groups ofresponse genes between parents may result from divergent negative regula-tors, such as transcriptional repressors, and both groups of response genes(R1 and R2) showed low-parent expression due to the trans-action of negativeregulators in the hybrid. As a result, balanced growth-response is disrupted,and the hybrid shows stronger growth (G1 and G2) than both parents.

Yang et al. PNAS | July 25, 2017 | vol. 114 | no. 30 | 8105

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

July

30,

202

1

30 accessions, which reported multiple heterosis-associated loci andcandidate genes for flowering time and rosette traits (31). It is likelythat the Arabidopsis hybrid populations used for GWAS were de-rived from different experimental designs and were different inscale. It is also likely that there are different mechanisms underlyingdifferent heterotic traits. The heterosis candidate genes identified inthis study were also different from those reported in rice (22). Riceis a domesticated crop subjected to extensive human selection. Thegenetic loci identified in that study should reflect improvements ofyield-related traits in hybrids instead of a general mechanism forbiomass or growth heterosis throughout plant species, which couldbe identified in undomesticated natural populations.In conclusion, our study demonstrated the efficiency of combin-

ing GWAS and transcriptome data to dissect the comprehensivegenomic architecture of heterosis. The combinational contribu-tion of accumulated superior alleles of the genes involved in basicbiological processes and the repressed expression of stimulus-responsive genes should enable more vigorous growth of hybridsrelative to their parents.

Materials and MethodsA total of 201 A. thaliana accessions obtained from the Arabidopsis BiologicalResource Center were used in this study (Dataset S1). Accession Col-0 was usedas the common maternal line and was crossed with the 200 other accessionsthrough hand-pollination. Details of plant materials and growth conditions,phenotyping, and population genetic analyses are described in SI Appendix,SI Materials and Methods.

For GWAS, 722,000 SNPs were used. GWAS was conducted between thestrongly and positively correlated biomass BPH and leaf area BPH using analgorithm for multivariate linear mixed models in the GEMMA software (32)(www.xzlab.org/software.html). To correct for multiple testing, a Benjamini–Hochberg test was performed. Thresholds of 0.2 were applied to identifymodestly significant SNPs associated with biomass heterosis.

The details and procedures of RNA-seq and data analysis and quantitativeRT-PCR are provided in SI Appendix, SI Materials and Methods.

ACKNOWLEDGMENTS. This work was supported by grants from the NationalKey Research and Development Program of China (2016YFD0100801), theNational Natural Science Foundation of China (31330048, 31621001), Peking-Tsinghua Center for Life Sciences (to X.W.D), State Key Laboratory of Proteinand Plant Gene Research, and in part by the Postdoctoral Fellowship ofPeking–Tsinghua Center for Life Sciences.

1. Chen ZJ (2013) Genomic and epigenetic insights into the molecular bases of heterosis.Nat Rev Genet 14:471–482.

2. Schnable PS, Springer NM (2013) Progress toward understanding heterosis in cropplants. Annu Rev Plant Biol 64:71–88.

3. Chen ZJ (2010) Molecular mechanisms of polyploidy and hybrid vigor. Trends Plant Sci15:57–71.

4. Lippman ZB, Zamir D (2007) Heterosis: Revisiting the magic. Trends Genet 23:60–66.5. Xiao J, Li J, Yuan L, Tanksley SD (1995) Dominance is the major genetic basis of

heterosis in rice as revealed by QTL analysis using molecular markers. Genetics 140:745–754.

6. Li L, et al. (2008) Dominance, overdominance and epistasis condition the heterosis intwo heterotic rice hybrids. Genetics 180:1725–1742.

7. Li ZK, et al. (2001) Overdominant epistatic loci are the primary genetic basis of in-breeding depression and heterosis in rice. I. Biomass and grain yield. Genetics 158:1737–1753.

8. Yu SB, et al. (1997) Importance of epistasis as the genetic basis of heterosis in an eliterice hybrid. Proc Natl Acad Sci USA 94:9226–9231.

9. Fujimoto R, Taylor JM, Shirasawa S, Peacock WJ, Dennis ES (2012) Heterosis of Ara-bidopsis hybrids between C24 and Col is associated with increased photosynthesiscapacity. Proc Natl Acad Sci USA 109:7109–7114.

10. Miller M, Song Q, Shi X, Juenger TE, Chen ZJ (2015) Natural variation in timing ofstress-responsive gene expression predicts heterosis in intraspecific hybrids of Arabi-dopsis. Nat Commun 6:7453.

11. Groszmann M, et al. (2015) Hormone-regulated defense and stress response networkscontribute to heterosis in Arabidopsis F1 hybrids. Proc Natl Acad Sci USA 112:E6397–E6406.

12. He G, et al. (2010) Global epigenetic and transcriptional trends among two ricesubspecies and their reciprocal hybrids. Plant Cell 22:17–33.

13. He G, et al. (2013) Conservation and divergence of transcriptomic and epigenomicvariation in maize hybrids. Genome Biol 14:R57.

14. Groszmann M, et al. (2011) Changes in 24-nt siRNA levels in Arabidopsis hybridssuggest an epigenetic contribution to hybrid vigor. Proc Natl Acad Sci USA 108:2617–2622.

15. Shen H, et al. (2012) Genome-wide analysis of DNA methylation and gene expressionchanges in two Arabidopsis ecotypes and their reciprocal hybrids. Plant Cell 24:875–892.

16. Meijon M, Satbhai SB, Tsuchimatsu T, Busch W (2014) Genome-wide association studyusing cellular traits identifies a new regulator of root development in Arabidopsis.Nat Genet 46:77–81.

17. Atwell S, et al. (2010) Genome-wide association study of 107 phenotypes in Arabi-dopsis thaliana inbred lines. Nature 465:627–631.

18. Huang X, et al. (2012) Genome-wide association study of flowering time and grainyield traits in a worldwide collection of rice germplasm. Nat Genet 44:32–39.

19. Huang X, et al. (2010) Genome-wide association studies of 14 agronomic traits in ricelandraces. Nat Genet 42:961–967.

20. Kump KL, et al. (2011) Genome-wide association study of quantitative resistance tosouthern leaf blight in the maize nested association mapping population. Nat Genet43:163–168.

21. Huang X, et al. (2015) Genomic analysis of hybrid rice varieties reveals numeroussuperior alleles that contribute to heterosis. Nat Commun 6:6258.

22. Huang X, et al. (2016) Genomic architecture of heterosis for yield traits in rice. Nature537:629–633.

23. Doebley JF, Gaut BS, Smith BD (2006) The molecular genetics of crop domestication.Cell 127:1309–1321.

24. Koornneef M, Alonso-Blanco C, Vreugdenhil D (2004) Naturally occurring geneticvariation in Arabidopsis thaliana. Annu Rev Plant Biol 55:141–172.

25. Weigel D (2012) Natural variation in Arabidopsis: From molecular genetics to eco-logical genomics. Plant Physiol 158:2–22.

26. Genomes Consortium (2016) 1,135 Genomes reveal the global pattern of poly-morphism in Arabidopsis thaliana. Cell 166:481–491.

27. Barth S, Busimi AK, Friedrich Utz H, Melchinger AE (2003) Heterosis for biomass yieldand related traits in five hybrids of Arabidopsis thaliana L. Heynh. Heredity 91:36–42.

28. Mitchell-Olds T (1995) Interval mapping of viability loci causing heterosis in Arabi-dopsis. Genetics 140:1105–1109.

29. Kusterer B, et al. (2007) Heterosis for biomass-related traits in Arabidopsis in-vestigated by quantitative trait loci analysis of the triple testcross design with re-combinant inbred lines. Genetics 177:1839–1850.

30. Meyer RC, et al. (2010) QTL analysis of early stage heterosis for biomass in Arabi-dopsis. Theor Appl Genet 120:227–237.

31. Seymour DK, et al. (2016) Genetic architecture of nonadditive inheritance in Arabi-dopsis thaliana hybrids. Proc Natl Acad Sci USA 113:E7317–E7326.

32. Zhou X, Stephens M (2014) Efficient multivariate linear mixed model algorithms forgenome-wide association studies. Nat Methods 11:407–409.

33. Kim S, et al. (2007) Recombination and linkage disequilibrium in Arabidopsis thaliana.Nat Genet 39:1151–1155.

34. Schoof H, et al. (2000) The stem cell population of Arabidopsis shoot meristems inmaintained by a regulatory loop between the CLAVATA and WUSCHEL genes. Cell100:635–644.

35. Dewitte W, Murray JA (2003) The plant cell cycle. Annu Rev Plant Biol 54:235–264.36. Inze D, De Veylder L (2006) Cell cycle regulation in plant development. Annu Rev

Genet 40:77–105.37. Gonzalez N, Vanhaeren H, Inzé D (2012) Leaf size control: Complex coordination of

cell division and expansion. Trends Plant Sci 17:332–340.38. Polyn S, Willems A, De Veylder L (2015) Cell cycle entry, maintenance, and exit during

plant development. Curr Opin Plant Biol 23:1–7.39. Groszmann M, et al. (2014) Intraspecific Arabidopsis hybrids show different patterns

of heterosis despite the close relatedness of the parental genomes. Plant Physiol 166:265–280.

40. Hu Y, Xie Q, Chua NH (2003) The Arabidopsis auxin-inducible gene ARGOS controlslateral organ size. Plant Cell 15:1951–1961.

41. Guo M, et al. (2014) Maize ARGOS1 (ZAR1) transgenic alleles increase hybrid maizeyield. J Exp Bot 65:249–260.

42. Todesco M, et al. (2010) Natural allelic variation underlying a major fitness trade-offin Arabidopsis thaliana. Nature 465:632–636.

43. Lozano-Duran R, et al. (2013) The transcriptional regulator BZR1 mediates trade-offbetween plant innate immunity and growth. eLife 2:e00983.

44. Fan M, et al. (2014) The bHLH transcription factor HBI1 mediates the trade-off be-tween growth and pathogen-associated molecular pattern-triggered immunity inArabidopsis. Plant Cell 26:828–841.

45. Malinovsky FG, et al. (2014) Antagonistic regulation of growth and immunity by theArabidopsis basic helix-loop-helix transcription factor homolog of brassinosteroidenhanced expression2 interacting with increased leaf inclination1 binding bHLH1.Plant Physiol 164:1443–1455.

8106 | www.pnas.org/cgi/doi/10.1073/pnas.1705423114 Yang et al.

Dow

nloa

ded

by g

uest

on

July

30,

202

1