Zilhs Ahmed Jewel,Juhr Ali*,Yunlong Png,Anumll MhenderBrt AeroJose Hernndez,Jinlong Xu,Zhikng Li,*

a Rice Breeding Platform,International Rice Research Institute(IRRI),Los Ba?os,Laguna 4031,Philippines

b Institute of Crop Sciences,Chinese Academy of Agricultural Science,Beijing 100081,China

c Institute of Crop Science,College of Agriculture and Food Science,University of the Philippines,Los Ba?os,Laguna 4031,Philippines

Keywords:Nutrient use efficiency Grain yield Nitrogen Phosphorus Green super rice

A B S T R A C T The developm ent of green super rice varieties w ith im proved nutrient use efficiency(Nu UE)is a vital target area to increase yield and m ake it m ore stable under rainfed conditions.In the present study,we followed an early backcross(BC)breeding approach by using a highyielding and w idely adapted Xian variety,Weed Tolerant Rice 1(WTR-1),as a recipient and a Geng variety,Hao-An-Nong(HAN),as a donor.Starting from the BC1F2 generation,the BC population w ent through one generation of selection under irrigated,low-input,and rainfed conditions,follow ed by four consecutive generations of screening and selection for high grain yield(GY)under six different nutrient conditions(NPK,75N,-N,-P,-NP,and-NPK),leading to the developm ent of 230 BC1F6 introgression lines(ILs).These 230 ILs w ere evaluated under the sam e six nutrient conditions for 13 agro-m orphological and grain yield component traits in com parison to four checks and parents.Significant trait variations w ere observed betw een the treatm ents and ILs.Positive correlations w ere identified for GY w ith biom ass,panicle length,flag-leaf area,flag-leaf w idth,filled grain num ber per panicle,1000-grain w eight,and tiller num ber under-N,-P,-NP,and-NPK conditions.Out of 230 ILs,12 w ere identified as prom ising under tw o or m ore nutrient deficiency conditions.The results dem onstrated an efficient inter-subspecific BCbreeding procedure w ith a first round of selection under rainfed-drought conditions,follow ed by four generations of progeny testing for yield performance under six nutrient conditions.The promising ILs can be useful resources for m olecular genetic dissection and understanding the physiological m echanism s of Nu UE.

1.Introduction

Rice rem ains the m ost im portant cereal crop and principal staple food in developing countries,particularly in Asia[1].Global rice production needs to keep pace w ith feeding the rapidly grow ing human population in a sustainable m anner.World rice production during the Green Revolution(GR)and post-GRm ore than doubled[2].This increased rice production was primarily achieved through the development and wide adoption of sem i-dw arf GR rice varieties that w ere highly responsive to high inputs(chem ical fertilizers,pesticides,and am ple irrigation w ater)[3].These breeding objectives predominated over the last three to four decades but now lie exhausted am id our hope to raise productivity per se sustainably.How ever,the crop's efficiency in harnessing applied inputs led to yields approaching a theoretical limit[4,5]. Single-year fertilizer N recovery efficiencies in researcher-m anaged experim ental plots averaged 46%for the rice crop[6].How ever,the experimental plots do not accurately reflect the recovery efficiencies available on-farm.Differences in farming operations and m anagement practices such as tillage,seeding,w eed and pest control,irrigation,and harvesting can affect nutrient use efficiency.Nitrogen recovery in crops grow n by farmers rarely exceeds 50%and is often m uch low er.The average nitrogen recovery efficiency in farm ers'fields ranges from 20%to 30% under rainfed conditions and from 30%to 40%under irrigated conditions[7].Most of the tim e,only half of the applied nutrients are used by rice plants and a significant part of the nitrogen in fields is lost through volatilization,leaching,de-nitrification,and soil erosion[8,9].

Over the years,no systematic rice breeding has been reported for Nu UE.Therefore,there is a need to breed rice cultivars w ith im proved Nu UEfor sustainable rice production under lim ited fertilizer input resources.Recently,green super rice(GSR),defined as rice varieties that can produce high and stable yields under few er inputs(w ater,nutrients,and pesticides)and adverse conditions,has becom e a new concept for enhancing rice Nu UE and achieving sustainable rice production through breeding[3,10].Breeding varieties w ith higher Nu UEis essential not only to im prove yield and reduce production costs but also to avoid environm ental pollution and m aintain the sustainability of cropping system s[11].In the light of high energy costs and increasingly fluctuating resources,future agricultural system s need to be m ore productive and efficient,including for fertilizer and w ater.Also,im proved Nu UEis an essential prerequisite for extending crop production into m arginal soils w ith low er nutrient availability(for exam ple,of nitrogen,phosphorus,and potassium)[11].Thus,developing rice varieties w ith high grain yield under low-nutrient conditions has becom e a breeding priority[12].Cultivars w ith higher Nu UE,coupled with best m anagement practices,w ill contribute to sustainable agricultural system s that protect and prom ote soil,w ater,and air quality[13].Unfortunately,only a few system atic breeding efforts until now have been extended in this respect.In these cases,breeding populations w ere screened under varying rates of N and P fertilizers for the identification of varieties with higher Nu UE[14-18].

Here,we report an effort for developing GSRvarieties w ith im proved Nu UE using a unique breeding approach through selecting ILs w ith higher Nu UE in a BC breeding program,w hich is potentially going to be w idely adopted for im proving the Nu UEof rice varieties in the future.

2.Materials and m ethods

2.1.Plant materials

A BC1F2population developed from a cross betw een an elite Xian(indica)variety,Weed Tolerant Rice 1(WTR-1,as the recipient),and a Chinese Geng(japonica)variety,Hao-An-Nong(HAN,as the donor)[19].WTR-1 is a high-yielding variety from southern China w ith w ide adaptability across subtropical and tropical areas of Asia.WTR-1 w as crossed w ith HAN in the 2010 dry season(DS)and the F1plants w ere backcrossed to WTR-1 once in the 2010 w et season(WS).In the 2011DS,seeds from＞25 segregating BC1F1plants w ere bulk harvested w ithout selection to form a single BC1F2population.

The Nu UE introgression lines were developed by selective introgression breeding by a screening of early backcross generation bulk populations(BC1F2)derived in the recipient background of NUE-efficient WTR-1 derived from four donor entries(HAN,Y134,Zhong 413,and BG300)under irrigated(IG),lowinput(LI),and rainfed(RF)conditions.These selections w ere screened under six nutrient conditions for three successive rounds.The relative BC1F2bulk yield performances showed the following:(i)WTR-1/Zhong 413:high yielding for both IG and LI conditions,(ii)WTR-1/Y134:high yielding under IG conditions but low yielding under LI conditions,(iii)WTR-1/HAN:low yielding under IG conditions but high yielding under LI conditions,and(iv)WTR-1/Bg300:low yielding for both IG and LI conditions.How ever,w e selected the bulk population of WTR-1/HAN for further selective introgression breeding as it show ed the possibility of a different pathway for nutrient use and uptake.

2.2.Selection schemes for improving Nu UE

Fig.1 show s the phenotypic selection schem es of the BC1F2bulk population to develop ILs w ith im proved Nu UE at the International Rice Research Institute(14°13′N,121°15′E,at an elevation of 21 m above m ean sea level),Philippines,during the three w et seasons of 2011-2013 and tw o dry seasons of 2012 and 2013,respectively.The first round of single-plant selection for higher grain yield w as practiced on the BC1F2bulk population grow n under IG,LI,and RF conditions in the 2011WS,resulting in 46 selected BC1F2plants.Then,in the follow ing seasons,four rounds of line-based selection w ere carried out from progeny testing under six different nutrient conditions(NPK,75N,-N,-P,-NP,and-NPK).NPK conditions used a total of NPK fertilizer of 160-50-50 kg ha-1in the DSand 90-30-30 kg ha-1in the WS.The 75N,-N,-P,-NP,and-NPK conditions indicated that 75%of nitrogen,the absence of N,the absence of P,the absence of both N and P,and finally the absence of NPK w ere used,respectively,com pared w ith norm al NPK conditions.Selection w as practiced by selecting the best one to three plants from the highest-yielding lines.

Fig.1-Phenotypic selection schem e for the id entification of nutrient use efficient ILs under six nutrient cond itions.WTR-1,Weed Tolerant Rice 1(recipient);HAN,Hao-An-Nong(d onor);IG,irrigated cond itions;LI,low-inp ut conditions;RF,rainfed conditions;ILs,introgression lines;SI,selection intensity;Sea,season;Gen,generation;No.of lines,num ber of introgression lines,grow n und er six nutrient conditions;DS,d ry season;WS,w et season.

Phenotyping of grain yield-attributed traits of Nu UE-ILs:The final evaluation of the 230 BC1F7ILs for Nu UE was conducted in replicated field experim ents under the six levels of nutrient inputs(NPK,75N,-N,-P,-NP,and-NPK)at IRRIin the 2014DS.Seeds of the 230 ILs,parents,and four checks used w ere PSB Rc82(low input tolerant and irrigated),NSIC Rc222(high yielding irrigated),Apo(rainfed drought tolerant/upland),and IR74371-70-1-1(rainfed drought tolerant/upland)that w ere sow n on the seedling nursery bed,and 20-day-old seedlings of each line w ere transplanted into a tw o-row plot w ith 24 plants or hills at a spacing of 20.0 cm×15.0 cm,w ith one seedling per hill and tw o replications for each line.The checks and parents w ere replicated in all six nutrient conditions.Weeds w ere controlled by using herbicides and hand pulling.At m aturity,three plants w ere diagonally sam pled in the m iddle row of each plot for the phenotypic evaluation of 13 agro-morphological and grain yield traits.These traits w ere plant height(PH,cm),flag-leaf length(FLL,cm),flag-leaf w idth(FLW,m m),flag-leaf area(FLA,cm2),tiller num ber(TN),heading date(HD),spikelet num ber per panicle(SN),panicle length(PL,cm),filled grain num ber per panicle(FGN),spikelet fertility(SF),1000-grain w eight(TGW),biom ass(BM,g),and grain yield per plant(GY,g).

2.3.Statistical data analysis

Data on the m easured m orphological,agronom ic,and yield traits w ere analyzed using Microsoft Excel softw are.Analysis of variance(ANOVA)w as used to determ ine the significant differences among different treatments(T)of nutrient input(NPK,75N,-N,-P,-NP,and-NPK),am ong lines(G),and G×T interactions using SAS program 9.1(SAS Institute Inc.,Cary,NC,USA).Duncan's m ultiple range test(DMRT)w as used to compare the differences betw een the mean values of the treatm ents and genotypes w ith the help of Rsoftw are[20].

3.Results

3.1.Development of introgression lines(ILs)

Fig.2 show s the selection schem e for developing WTR-1 ILs starting with the first round of selection of the WTR-1/HAN//WTR-1 BC1F2population w ith 576 plants under each of the IG,LI,and RFconditions.In the first round of selection,46 plants w ere visually selected based on the m ore desirable plant type and higher yield performance than WTR-1(selection intensity,SI=2.7%)in the 2011WS,including 17 plants from IG,15 plants from LI,and 14 plants from RF.In the next season,2012DS,the 46 BC1F3lines(1104 plants)w ere planted in single lines under the six different nutrient conditions,from w hich 73 plants w ere selected,including 7,17,13,4,19,and 13 plants from NPK,75N,-N,-P,-NP,and-NPK conditions,respectively.In the follow ing three consecutive seasons(2012WS,2013DS,and 2013WS),the same phenotypic selection w as performed,resulting in 121 BC1F4plants,202 BC1F5plants,and 230 BC1F6plants,respectively.Together,the cum ulated contribution of the four rounds of selection through progeny testing under NPK,75N,-N,-P,-NP,and-NPK conditions leading to the final 230 BC1F7(ILs)w as 0.107 for NPK,0.144 for 75N,0.180 for-N,0.088 for-P,0.208 for-NP,and 0.273 for-NPK.

Fig.2-Effects of nutrient cond itions on the average expression of grain yield per plant(in g,GY)and related agrom orphological and yield-attributed traits of 230 WTR-1 introgression lines.GY,grain yield per plant(g);HD,head ing date(d ays);SN,spikelet num ber per panicle;FGN,filled grain num ber per p anicle;SF,spikelet fertility(%);TGW,thousand-grain w eight(g);PH,plant height(cm);TN,tiller num ber p er p lant;PL,p anicle length(cm);FLL,flag-leaf length(cm);FLW,flag-leaf w idth(cm);FLA,flag-leaf area(cm 2);BM,biom ass(g).

3.2.Phenotypic performance of ILs for agro-morphological and yield traits in diverse environments

In the replicated progeny testing,the recipient and donor had sim ilar average grain yield across all six nutrient input conditions except under-NPK and-P.In the form er case,WTR-1 had significantly higher GY than the donor,HAN,w hile the opposite w as true in the latter case under-P(Table 1).When the effects of different nutrient treatm ents w ere compared,the m ean yields of the nutrient treatments could be divided into three levels:(1)the norm al NPK show ing the highest m ean yield of the tested m aterials,and show ing nonsignificant differences from 75N;(2)-P having the second highest average yield;and(3)-N,-NP,and-NPK showing sim ilarly low average GY for the tested lines.The nutrient treatm ents had sim ilar effects on PH,TN,and biom ass and had no effects on HD,TGW,and SF(Fig.2).How ever,the deficiency of nitrogen(-N)and phosphorus(-NP)caused significant reductions in SN,FGN,FLL,and FLA in com parison to the rem aining five Nu UE conditions.Im portantly,the ILs show ed sim ilar m ean values in-NPand-NPK but trem endous variation and transgressive segregation were observed for GY under all six nutrient conditions.

Table 2 show s the correlation coefficients betw een m ean yield perform ance and other agronomic traits measured under different nutrient conditions.High and positive correlations betw een grain yield and biom ass w ere observed across all six nutrient conditions,particularly under-N,-NP,and-NPK,w hich w as follow ed by FLW,FLA,and TN.Surprisingly,the contributions of the other tw o direct yield com ponents(FGN and TGW)to grain yield w ere,although positive,m uch less im portant,particularly under the lesser nutrient deficiency conditions(NPK and 75N).

The ILs show ed trem endous segregation for their GY under different nutrient treatm ents(Table 1).In fact,a significant portion of the ILs had significantly(10%)higher GY than WTR-1 under each of the nutrient treatments.In particular,more ILs show ed significantly higher yields under 75N and-NP conditions,w hile the opposite w as true under-P.Under the other conditions,the ILs show ed largely norm al distributionsw ith approxim ately equal num bers of high-and low-yielding ILs,providing tremendous opportunities for selection.Indeed,several prom ising ILs w ith significantly higher GY than WTR-1 w ere identified under each of the Nu UE treatm ents.These included Nue-114(50.6 g),Nue-115(42.9 g),Nue-3(42.1 g),Nue-51(42.1 g),Nue-112(40.4 g),and Nue-230(48.8 g)under NPK,-NPK,-N,-P,-NP,and 75N conditions.

Table 1-The m ean grain yield per plant(g)perform ance of 230 ILs and their p arents und er six nutrient input cond itions.

Table 2-Correlations of GY w ith other related traits of 230 WTR-1 introgression lines und er six nutrient conditions.

3.3.Selection of promising ILs w ith superior yield under two or more nutrient conditions

Table 1 show s the number of the selected 230 ILs from previous screening under IG,LI,and RFconditions that had significantly higher m ean grain yield under one or tw o nutrient conditions.Of the 230 selected ILs,the num ber of ILs show ing significantly higher GY than WTR-1 w as 52 under NPK,103 under 75N,33 under-P,63 under-N,52 under-NP,and 49 under-NPK.This translated into indirect selection efficiencies of 0.226,0.448,0.143,0.274,0.226,and 0.213 for improved yield performance under NPK,75N,-P,-N,-NP,and-NPK conditions,respectively.Of the 52 high-yielding ILs identified under normal NPK conditions,26,7,17,20,and 17 ILs also show ed significantly improved GY under 75N,-P,-N,-NP,and-NPK conditions,respectively.Of the 103 high-yielding ILs selected under 75N,20,31,34,31,and 26 ILs showed significantly higher GY under norm al NPK,-P,-N,-NP,and-NPK conditions,respectively.Of the 33 high-yielding ILs identified under-Pconditions,20,7,13,15,and 9 of these ILs perform ed w ell w ith significantly higher GY under norm al NPK,75N,-N,-NP,and-NPK conditions,respectively.Sim ilarly,of the 63 high-yielding ILs selected under-N conditions,13,31,17,22,and 20 of these ILs show ed significantly higher GY under-P,normal NPK,75N,-NP,and-NPK conditions,respectively.From the 52 high-yielding ILs selected under-NPconditions,w e w ere able to identify 22,15,34,20,and 20 ILs show ing significantly higher GY under-N,-P,norm al NPK,75N,and-NPK conditions,respectively.Finally,from the 49 high-yielding ILs under-NPK,20,27,9,31,and 17 ILs also show ed significantly higher yield under NPK,75N,-P,-N,and-NP conditions,respectively.This could be converted into an average secondary indirect selection efficiency of 0.335 for NPK,0.276 for 75N,0.388 for-P,0.327 for-N,0.454 for-NP,and 0.424 for-NPK,respectively.In other w ords,replicated progeny testing under-NP and-NPK resulted in the highest efficiency for selecting higher grain yield across all six nutrient conditions.

Table 3 show s the number of low-yielding ILs under different conditions from all high-yielding ILs identified under each of the six nutrient conditions.Generally,a small portion of the ILs show ing superior yield under one nutrient condition w as yielding poorly under the other condition(s),suggesting a low correlation for grain yield performance between different nutrient conditions.The values on the first diagonal show the num ber of higher GY ILs in each of the nutrient conditions and the values in the upper triangular portion show the number of higher GY ILs shared in each pair of the nutrient conditions.

3.4.GY performance of groups divided by type of first-round selection

To exam ine how the types of first-round selection affected the selection efficiency for im proving Nu UE,the 230 ILs w ere divided into three groups,including 43 ILs originally selected from IGconditions,21 ILs from LIconditions,and 166 ILs from RF conditions.Surprisingly,ILs originally selected from LI conditions show ed the low est mean GY under-NP,-NPK,and-P,but the highest m ean GY under 75N and NPK as com pared w ith checks and parents,indicating that these ILs adapted them selves better under good nutrient conditions,but w ere very sensitive to P deficiency(Fig.3).In contrast,those ILs originally selected from IG conditions show ed relatively higher GY under-N and-NPK but perform ed poorly under 75N and NPK,suggesting that they w ere less sensitive to N deficiency,w hile those ILs originally selected from RF conditions appeared to perform slightly better under-P conditions.The selected ILs w ith high GY under IG conditions m ay not be perform ing better under low input conditions and vice versa.We indeed observed a negative correlation betw een yield performance under-P and-N conditions insome ILs.For example,five high-yielding ILs(Nue-104,Nue-46,Nue-57,Nue-77,and Nue-83)identified under-P conditions show ed poor yield perform ance under-N conditions(Table 4).When compared w ith the performance of WTR-1,the poor yield perform ance under-N could be prim arily attributed to greatly reduced biom ass,TN,and FGN.

Table 3-The num ber of WTR-1 introgression lines show ing significantly higher grain yield under tw o nutrient conditions.

Fig.3-Com parison of m ean GY perform ance of introgression lines in six nutrient conditions in w hich solid and broken lines indicate the values of the recip ient,WTR-1,and the donor,HAN,resp ectively.IG,irrigated;LI,low input;RF,rainfed(drought).

Table 4-Grain yield p er p lant(g)perform ance of five WTR-1 ILs and their parents w ith high GY in-Pconditions but low GY in-N conditions.

3.5.Promising ILs under different Nu UEconditions

Table 5 lists 12 ILs that show ed significantly higher grain yield than WTR-1 under tw o or m ore nutrient conditions and the same yield as WTR-1 under the other conditions.These included one prom ising line,Nue-115,w hich had significantly higher yield under all four nutrient deficiency conditions,and four ILs(Nue-114,Nue-112,Nue-229,and Nue-230)that had superior yield performance under norm al or 75N plus tw o nutrient deficiency conditions.These lines have now been prom oted to replicated m ulti-location yield trials for further testing for potential release as new varieties in the target environm ents of several Asian and African countries.

4.Discussion

Plant breeders currently face at least three great challenges for im proving Nu UE.In the first place,im proved Nu UE is yet to find a prom inent place in m ost rice breeding program s am ong the prim ary target traits that dom inate current breeding programs,w hich now emphasize enhancing yield potential under intensive cropping system s and tolerance of abiotic stresses.For exam ple,adequate tolerance of drought and salinity,in addition to high yield potential,w ould be the priority traits for breeding programs for the rainfed ecosystem and saline areas.The second difficulty com es from the factthat Nu UEis a complex trait,of low heritability,w hich is often affected by the intrinsic large variation in soil fertility and type.The third challenge is the apparent“l(fā)ack”of appropriate donors w ith high Nu UE because of few er past efforts to identify rice germ plasm accessions w ith high Nu UE.As a result,few w ell-docum ented breeding efforts have been m ade for developing rice cultivars tolerant of nitrogen and phosphorus deficiencies[21-24].In this respect,results from this study provided at least partial answ ers to the three practical questions and dem onstrated an innovative breeding m ethod involving four consecutive rounds of selection under six different Nu UE conditions for developing rice varieties w ith significantly im proved Nu UE.

Table 5-Tw elve prom ising higher yield ing ILs w ith superior yield perform ance in tw o or m ore nutrient cond itions.

4.1.Exploiting the“hidden”genetic diversity in different subspecific gene pools for improving Nu UE

In the BC breeding procedure for developing ILs w ith im proved Nu UE,w e used HAN,a Geng variety,as the donor,to improve the Nu UEof a widely adaptable Xian variety,WTR-1.WTR-1 w as released in Bangladesh as BRRIdhan69 in 2015 under the Green Super Rice(GSR)project for“boro”conditions after six years of rigorous testing for grain yield and grain quality.In each generation of the breeding scheme,increasing the selection intensity provides a w indow of opportunity to identify suitable ILs under both LI and IG conditions.This approach can be useful to understand the low heritability of complex traits.Our results indicated that HAN did not seem to have a high Nu UE but w as able to contribute m any useful genes/traits to WTR-1,resulting in the developm ent of m any WTR-1 ILs w ith greatly im proved Nu UE.This w as consistent w ith our previous results that there is rich hidden genetic diversity in the different subspecific gene pools that can be exploited for im proving com plex traits such as abiotic/biotic stress tolerance/resistance[23,25-27]using a modified BC breeding strategy[10,23,28].This is not surprising because it is now know n that different rice populations,particularly the tw o m ajor subspecies,each contain large num bers of unique genes and alleles(haplotypes)that are absent or rare in other populations or subspecies[19].How ever,the segregating introgression lines often show epistatic genetic com plem entation com bining the better of the tw o parents in a transgressive manner.Thus,results from this study indicated that accessions of different subspecies could be a good source of useful genetic diversity for im proving Nu UEin rice.

4.2.Promising lines w ith high Nu UE and selection efficiency for improved Nu UE

In this study,w e w ere able to identify m any ILs(out of the 230 ILs)that had significantly higher yield(＞30 g in GY)than WTR-1 under at least one nutrient deficiency condition,including 10 ILs under-NPK,7 ILs under-N,47 ILs under-P,9 ILs under-NP,plus 12 prom ising ILs w ith significantly im proved Nu UE under m ore than one nutrient deficiency condition(Table 5).While this is consistent w ith other findings[27,28]that highyielding varieties w ith high Nu UE can be developed under Nu UE deficiency conditions,it would be interesting to understand how the various types of selection schem es w e adopted affect our selection efficiency for im proving Nu UEin rice.For the first round of selection,w e had approximately the sam e selection intensity under IG,LI,and RF conditions,of w hich LI and RF w ere relevant to Nu UE.Under LI conditions,w e expected to select those plants that w ere tolerant of low NPK,because zero nutrient input w as given to the LIfield for 16 consecutive crop seasons before the selection.Surprisingly,the ILs selected under LIconditions did not have higher yield under-NP,-NPK,and-P conditions.Instead,these ILs had higher grain yield under-N,75N,and NPK conditions,when com pared w ith the ILs selected from IGand RFconditions(Fig.3).According to the soil fertility analyses,the total soil N(Kjeldahl)and the available P and available K contents of the LIfield had reached a fairly constant low level at 700 m g N kg--1,12.66 m g P kg-1,218.49 m g K kg-1in an average of first three consecutive seasons of 2012(DSand WS)and 2013(DS).Together,our results seem ed to suggest that these lines selected under LIconditions had low Puptake ability but high Pusability,and high N uptake and usability,w hich rem ain to be validated by m ore accurately designed experim ents.On the other hand,water deficiency was the prim ary factor limiting rice yield perform ance under RF conditions w hen the first round of selection w as practiced.It w as reported that plant w ater use efficiency is closely related to Nu UEbecause w ater uptake w ould facilitate nutrient uptake[24].However,those ILs selected under RF conditions show ed relatively higher GY under-P and 75N conditions and a m uch greater variation(72.1%)in GY and all m easured traits across the six nutrient conditions than those ILs selected from either IG or LI conditions(Fig.3).Furtherm ore,the IL(Nue-112)w ith the highest Nu UE w as also selected from RF conditions(Table 5).This suggested that drought tolerance and high Nu UE in this population were apparently under independent genetic control.How ever,it is im portant to test w hether this holds true for other rice populations.For those ILs originally selected under IG conditions,no overall yield advantages under nutrient deficiency w ere observed.Overall,the efficiency of the first round of selection w as RF＞LI＞IG for im proving Nu UE.Therefore,selecting rice cultivars w ith higher Nu UE under rainfed conditions is a prom inent approach for the identification of superior rice cultivars.

4.3.Traits contributing to Nu UEof rice

Our results indicated that high Nu UE in rice is a very com plex trait controlled by different genetic,physiological,and m olecular mechanisms.At least tw o observations implied this conclusion.First,the contributions of the three yield components to GYwere relatively w eak and varied considerably w ith each nutrient deficiency condition,although high and positive correlations between GY and BM w ere consistently observed across all nutrient deficiency conditions.This was in contrast to previous reports that higher yields from enhanced grain filling ratio(GFR),spikelet number per panicle(SNP),number of panicles(NP),and TGW under nutrient deficiency conditions could be attributed to better nutrient absorption and uptake by plants after the application of fertilizer sources[29-34].Earlier reports also indicated that grain num ber and w eight w ere primarily controlled by P and N[35-37],because P is more used in the formation of grain productivity and N influences tillering ability,which contributed to higher yields through m ore panicles[38,39].Indeed,we observed that,of the six nutrient conditions,NPK produced the highest PH,and the lowest PH w as recorded in-N conditions.FLL,FLW,and FLA traits w ere low er in-NPconditions but were higher in-Pconditions.TN w as high in NPK and low in 75N.Similarly,higher TN leading to higher GY w as also reported in rice[40].Apparently,increased GY could be attributed to possible higher photo-assimilates and dry matter accumulation from increased leaf area in response to N fertilizer[41,42].TN,LL,LW,and LA were also reported significantly improved by the mining of nutrients,particularly N,through better root development,leading to improved translocation of carbohydrate capacity from expanded leaf rate[42-44].Second,w e observed that som e ILs show ed higher GY under one nutrient condition but low er GY in another(Table 4),w hile high GY of different ILs could be attributed to different combinations of yield com ponents.For exam ple,the high GY of Nue-112 under NPK conditions was mainly attributed to higher FGN,SN,PL,and BM but not TGW because the TGW of Nue-112 w as significantly low er than for WTR-1.How ever,its high GY under other conditions could be attributed to the simultaneous improvement of FGN,PL,BM,and TGW or the balance betw een them.Similarly,the high GY of Nue-115 could prim arily be attributed to TGW,FLW(-NPK),FLW(-P),and TGW(-N),respectively.Thus,the 230 ILs and their tremendous variation in Nu UE provide useful materials for studying the genetic and physiological mechanism s of Nu UEof rice,w hich is in progress.

5.Conclusions

We screened inter-subspecific BC1F2populations w ith a first round under RF,LI,and IG conditions,follow ed by a system atic phenotypic selection across six nutrient conditions over four rounds that helped us to identify introgression lines w ith im proved Nu UE.The study revealed that HAN did not possess a high Nu UE but w as able to contribute m any useful genes/traits to WTR-1,resulting in the development of m any WTR-1 ILs w ith greatly im proved Nu UE.These results confirm that rich hidden genetic diversity exists in different sub-specific gene pools that can be exploited for im proving complex traits such as abiotic/biotic stress tolerance/resistance using a modified BC breeding strategy.How ever,through this study,w e w ere able to develop as m any as 230 ILs that are being used for studying the genetic and physiological mechanisms of Nu UEof rice for different target Nu UErelated traits.The study dem onstrated an efficient intersubspecific BC breeding procedure w ith a first round of selection under rainfed-drought conditions,follow ed by four generations of progeny testing for yield performance under six different nutrient conditions.Overall,the efficiency of the first round of selection w as RF＞LI＞IG for im proving Nu UE.How ever,the five prom ising introgression lines(Nue-115,Nue-114,Nue-112,Nue-229,and Nue-230)w ith im proved Nu UEcould provide great help in rice breeding program s.

Acknow ledgm ents

The authors would like to thank and acknowledge the Bill&Melinda Gates Foundation(BMGF)for providing a research grant to Z.L.for the Green Super Rice project under ID OPP1130530.We w ould also like to thank the Departm ent of Agriculture of the Philippines for providing funds to J.A.under the Next-Gen project.