Upland rice (Oryza sativa L.) is an important crop in Brazil,

Transcription

1 RESEARCH Results of 25 Years of Upland Rice Breeding in Brazil Flavio Breseghello,* Orlando Peixoto de Morais, Patrícia Valle Pinheiro, Ana Carolina Simões Silva, Emilio da Maia de Castro, Élcio Perpétuo Guimarães, Adriano Pereira de Castro, José Almeida Pereira, Altevir de Matos Lopes, Marley Marico Utumi, and Jaison Pereira de Oliveira ABSTRACT The objective of this study was to estimate the results of the upland rice (Oryza sativa L.) breeding program conducted by the Brazilian Agricultural Research Corporation (Embrapa) and collaborators over the period of 1984 to 2009 covering 25 annual steps of improvement. The best lines generated by this program are evaluated in value for cultivation and use (VCU) trials. This study used data from 603 VCU trials conducted in seven Brazilian States. The group of lines entering VCU in each year was taken as a sample of the elite program in that year. Best linear unbiased estimates (BLUEs) of the means of groups were computed, and the regression of the BLUEs on years was taken as an estimate of the efficiency of the breeding program. Traits studied were grain yield (GY), days to flowering (DTF), and plant height (PH). For GY, three phases were described: from 1984 to 1992, the gain for GY was nonsignificant, from 1992 to 2002, the yearly gain was 15.7 kg ha 1 (0.53%), and from 2002 to 2009, the yearly gain was 45.0 kg.ha 1 (1.44%). No evidence of yield plateau was observed. There was a reduction of 0.25 d yr 1 in DTF and 0.52 cm yr 1 in PH, reflecting the change from traditional to modern cultivars. Factors influencing the results of the breeding program in this period are discussed. F. Breseghello, O.P. Morais, P.V. Pinheiro, A.C.S. Silva, E.M. Castro, É.P. Guimarães, A.P. Castro, and J.P. Oliveira, Embrapa Rice and Beans, Santo Antonio de Goiás, GO, Brasil ; J.A. Pereira, Embrapa Mid-North. A.M Lopes, Embrapa Eastern Amazon. M.M. Utumi, Embrapa Rondônia. Received 6 June *Corresponding author (fl avio@cnpaf.embrapa.br). Abbreviations: BLUE, best linear unbiased estimate; BLUP, best linear unbiased predictor; DTF, days to flowering; GY, grain yield; PH, plant height; TPE, target population of environments; VCU, value for cultivation and use. Upland rice (Oryza sativa L.) is an important crop in Brazil, with almost 2 million ha planted each year. Although this area is less than half of the area planted in the 1980s as a pioneer crop in the Brazilian savannas (also known as Cerrado), this system is regarded as having a high potential for expansion (Pinheiro et al., 2006) in a scenario of increase in the international demand for rice. The Brazilian Corporation for Agricultural Research (Embrapa) conducts an upland rice breeding program since 1975, in collaboration with other public institutions, aiming at developing rice cultivars with improved agronomic and grain quality characteristics, resistant to biotic and abiotic stresses, and well adapted to the upland rice growing regions in Brazil (Pinheiro et al., 2004). Plant breeding is a long-term investment; hence, decisions made today will show their consequences years later in the performance of the resulting cultivars. Results achieved in the past can help to forecast trends and plan ahead for program adjustments. For this reason, it is important to monitor the efficiency of the breeding program through time. Genetic gain estimates are Published in Crop Sci. 51: (2011). doi: /cropsci Published online 14 Mar Crop Science Society of America 5585 Guilford Rd., Madison, WI USA All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher CROP SCIENCE, VOL. 51, MAY JUNE 2011

2 useful as indicators of the efficacy of the methodology used as well as the potential of the germplasm exploited. Two types of methods have been used for estimating genetic gains from plant breeding programs through time: (i) evaluation of old and recent cultivars in field trials, followed by a regression analysis of phenotypic values as a function of the year of release of the cultivars (e.g., Peng et al., 2000; Tabien et al., 2008; De Vita et al., 2007), and (ii) meta-analysis of historical data from series of field trials, normally conducted with the aim of testing the value for cultivation and use (VCU) of candidate lines for varietal release. Meta-analysis has been used more frequently in Brazil (e.g., Soares et al., 1999; Breseghello et al., 1999; Cargnin et al., 2008) than in other countries, probably due to the influential work of Prof. Vencovsky (Vencovsky et al., 1986). If proper statistical methods are used, meta-analysis allows adjustment for environmental and experimental design effects in the unbalanced data sets normally available in breeding programs (Breseghello et al., 1998). Both approaches (i.e., cultivar comparison and meta-analysis) are capable of computing valid estimates of genetic gains. Advantages of cultivar comparison are that materials can be evaluated under homogeneous crop management for several traits simultaneously under equal methods of measurement. On the other hand, metaanalysis of a large data set allows better sampling of the environmental variation in the target population of environments (TPE), both in time and space, and includes both released and nonreleased materials. The contribution of plant breeding for grain yield has been studied in several annual crops (Table 1). In most cases, those studies indicated that plant breeding resulted in productivity gains. Although estimates vary considerably, genetic gains for grain yield are normally a few tens of kilograms per year. It is common to present the gain per year as a percentage of the mean yield for the crop under study. In this way, most rates of progress correspond to less than 1% (e.g., Zhou et al., 2007a). Peng et al. (2000) evaluated the genetic gain for irrigated rice, based on 12 rice cultivars developed at the International Rice Research Institute (IRRI), in the period of 1966 to 1995, resulting in a genetic gain of 75 to 81 kg ha 1 yr 1, corresponding to 1% per year. Breseghello et al. (1999) estimated the genetic gain for irrigated rice in the northeast of Brazil as 0.8% per year. Soares et al. (1999) computed the genetic gain for upland rice in the state of Minas Gerais, for the shortcycle and the intermediate-cycle subprograms, finding the genetic gains of 23.8 and 46.4 kg ha 1 yr 1, respectively. No report has been published so far on the genetic gain of the national rice breeding program in Brazil. This study was conducted with the objective of estimating the genetic gain resulting from Embrapa s upland rice breeding program on grain yield, days to flowering, and plant height through the meta-analysis of a large data set of field trials conducted in the most important upland rice growing states in Brazil in the period of 1984 to MATERIALS AND METHODS Source of Data The data used in this study were taken from the records of the VCU trials conducted by Embrapa and collaborators, under the scope of the upland rice breeding program, including 603 trials conducted during the period of 1984 to 2009, thus representing 25 steps of improvement (Table 2). The average number of trials per year was 23.2, varying from eight trials in 1990 to 39 trials in The geographic area represented includes the states of Goiás, Mato Grosso, Tocantins, Pará, Rondônia, Maranhão, and Piauí, where most of the upland rice is planted in Brazil nowadays. The data set comprised 45,688 field plots for grain yield (an average of 1757 plots per year), from which 37,119 plots (81.2%) had data for days to flowering and 41,468 plots (90.8%) had data for plant height. On average, 40.5 lines were tested in each year, varying from 17 lines in 2008 to 79 lines in 1993, whereas the number of lines debuting in VCU (lines that entered VCU for the first time) varied from three in 1991 to 40 lines in In the first year of the series (1984), all lines were considered novel, although not all of them were actually debuting in the VCU in that year. Only trials with reliable data, unambiguous genotype identification, and acceptable experimental precision (coefficient of variation for grain yield [GY] < 25%) were used. Description of Field Experiments The typical VCU trial used was conducted in randomized complete block design with four replicates and approximately 20 entries, from which four were check cultivars and the rest were breeding lines under test for the first, second, or third year. Novel lines were selected for VCU based on preliminary and regional trials, which are similar to VCU trials, although in a reduced number of sites. Bad performance in the VCU resulted in elimination of the material, whereas good performance resulted in further testing or submission for release as commercial cultivars. Cultivars or inbred lines from other breeding programs have been tested in those trials as well, for the purpose of comparison with Embrapa s materials. In the VCU trials, dry seeds were hand- or machine sown in rows normally 5 m long. Plots were composed of four or five rows. One row in each side was considered a border, and the intermediate 4 m of the internal rows were evaluated and reaped. All trials were conducted in the rainy season, without irrigation (exceptionally, supplementary sprinkler irrigation may have been provided). Chemical fertilizers were applied in the soil at sowing, in intermediate to high rates compared to farmer s management. Nitrogen fertilizer was normally top-dressed at the end of the vegetative stage. Weed control was made by hand weeding, herbicides, or a combination of the two. Insect control was made with insecticides applied to the furrow in most trials and, if needed, complemented with canopy spraying. No fungicides were applied, for evaluation of genetic resistance to fungal diseases. All traits were evaluated at the plot level. Days to flowering was evaluated as the interval between sowing and approximately CROP SCIENCE, VOL. 51, MAY JUNE

3 Table 1. Summary of results from previous studies on genetic gain of plant breeding programs. Crop Region Period Method of estimation Genetic gain (ha 1 yr 1 ) Reference Rice, irrigated Brazil, mid-north Meta-analysis 18 kg (0.3%) Rangel et al., 2000 Rice, irrigated Brazil, northeast Meta-analysis 54.9 kg (0.8%) Breseghello et al., 1999 Rice, irrigated United States, Texas Cultivar comparison 42.2 kg (high N input) 26.3 kg (low N input) Tabien et al., 2008 Rice, irrigated Philippines Cultivar comparison 75 to 81 kg (1%) Peng et al., 2000 Rice, upland Brazil, State of Meta-analysis 23.8 kg (short cycle); Soares et al., 1999 Minas Gerais 46.4 kg (medium cycle) Rice, irrigated India Meta-analysis NS (upland, rainfed lowland and irrigated) Muralidharan et al., 2002 and rainfed Barley Argentina Cultivar comparison : NS Abeledo et al., : 41 kg Sorghum United States, Cultivar comparison 13.5 kg Mason et al., 2008 Nebraska Maize United States, Cultivar comparison 37.4 kg Mason et al., 2008 Nebraska Sunfl ower Argentina Meta-analysis Grain yield: NS De La Vega et al., 2007 Oil yield: 11.9 kg Bread wheat Brazil, State of Meta-analysis 48 kg (1.8%) Cargnin et al., 2008 Minas Gerais Winter wheat Southern China Cultivar comparison 14.0 kg (0.3%) to 40.8 kg (0.7%) Zhou et al., 2007a for different provinces Winter wheat Northern China Cultivar comparison 32.1 kg (0.5%) to 72.1 kg (1.2%) Zhou et al., 2007b for different provinces Durum wheat Italy Cultivar comparison 19.9 kg De Vita et al., 2007 Rice, Oryza sativa L.; barley, Hordeum vulgare L.; sorghum, Sorghum bicolor (L.) Moench.; maize, Zea mays L.; sunfl ower, Helianthus annuus L.; bread wheat or winter wheat, Triticum aestivum L.; durum wheat, Triticum turgidum L. subsp. durum (Desf.) Husn. NS, nonsignifi cant. 50% of the plants flowered. Plant height was evaluated as the distance from the ground to the tip of the highest panicle in the prematurity stage. Plots were hand harvested and threshed and grains were sun dried and stored in room temperature for a few weeks for grain moisture stabilization before weighing. Grain yield was adjusted to kilograms per hectare. Statistical Analysis Materials were grouped according to the first year when they have been included in the VCU trials. For example, the group 2000 includes all the materials for which the earliest VCU data available are from the year 2000, irrespective of how many years they stayed under testing or whether the lines have been discarded or released as cultivars. Each year, some lines were discarded and others were added to the VCU trials. However, there were always common lines and checks between consecutive years, allowing control of the environmental variation through the entire period. From the 493 lines evaluated in the period, 376 were from Embrapa s breeding program, whereas the remaining lines where from state-level programs or private-sector programs or were introductions from other countries. All materials originated from other breeding programs were pooled into a separate group. This group contributed to the adjustment of the model but was not used in the regression step, for the estimation of the genetic gain. A mixed-effects model was used, with groups as fixed effects and lines within groups, years, experiments within years, and replicates within experiments as random effects. The statistical model can be represented as: Table 2. Summary of the Embrapa s upland rice breeding program data set used in this study. Year No. of trials Total no. of lines tested No. of new lines from Embrapa No. of field plots Arithmetic mean Total Y ijkmn = μ + g j + l i /g j + a k + t m /a k + b n /at km + ε ijkmn CROP SCIENCE, VOL. 51, MAY JUNE 2011

4 in which Y ijkmn is the data at the plot level, μ is the intercept, g j is the fixed effect of group j, l i /g j is the random effect of line i within group j, a k is the random effect of year k, t m /a k is the random effect of trial m within year k, b n /at km is the random effect of block n within trial m in year k, and ε ijkmn is the random error N(0, σ 2 ). Analysis was done in SAS (SAS Institute, 2004), with the following commands: proc mixed; class group line year trial block; model GY DTF PH = group; random year trial(year) block(year trial) line(group)/s; lsmeans group/cov; run;. Proc Mixed solves the following matrix equation by restricted maximum likelihood to estimate best linear unbiased estimates (BLUEs) of the fixed effects and the corresponding matrix of covariances. y = Xβ + Zλ + ε in which y is the vector of observations, X and Z are the design matrices of fixed and random factors, respectively, β is the vector of fixed effects of groups, λ is the vector of random effects of years, trials within years, blocks within trial within year, and lines within groups, and ε is the vector of random errors. Genetic gain was computed as the generalized linear regression coefficients of the BLUEs on years, according to the model: θ = (X V 1 X) 1 (X V 1 Y) in which θ is the vector of solutions of the generalized linear regression, X is the matrix of incidence, formed by a column of 1 s and a column indicating the year in the series corresponding to the groups, V is the matrix of covariances of the BLUEs of groups, and Y is the vector of BLUEs of groups. Significance of slopes was evaluated by the t test. Relative genetic gain was estimated as the ratio between the slope and the intercept, in percentage, in which the intercept represents the initial stage of the breeding program. The best linear unbiased predictors (BLUPs) of some lines (those released as cultivars or in the pipeline) were computed as the sum of the BLUEs for the group plus the effect of the line within group. RESULTS Structure of the Data Set All of the design variables were considered random factors for modeling part of the variance of the data as covariances between observations taken within groups. Lines within groups were considered as random samples from the elite breeding program in the year corresponding to the group. The structure of the matrix Z and the estimates of variance components are presented in Table 3. Years and trials within year responded by a large proportion of the variance, especially for GY and PH. The importance of trials (loosely corresponding to locations) in this data set is explained by the fact that the geographic area represented is very large, including many different soil types, climate patterns, and biotic and abiotic stresses. Blocks presented small importance for DTF and PH and moderate importance for GY. Lines within groups responded by a moderate proportion of the variance for GY, probably because lines entering VCU were previously selected for GY in preliminary trials. Contrastingly, a great proportion of the variance for DTF was related to lines, because in former years there were two contrasting groups of materials: the early flowering lines and the intermediatecycle lines (late-flowering lines have not been focused on by rice breeding in Brazil in this period). The BLUPs of all lines evaluated were computed, and Pearson s correlations between variables were estimated. Those correlations may result from simultaneous selection for different traits in the breeding program. Considering only Embrapa lines, no correlation was found between GY and DTF, indicating that early flowering lines were not less productive than intermediate-cycle lines, as one might expect. Mild negative correlation was found between GY and PH (r = 0.23 and p = 0.001), probably due to the fact that modern materials are shorter and higher yielding. A weak positive correlation was detected between PH and DTF (r = 0.11 and p = 0.05), also an expected result from combined selection for short and early-flowering plants. Genetic Gain for Grain Yield Taking the whole period of 26 yr (corresponding to 25 yr of improvement, considering 1984 as year zero), the mean gain for GY was 19.1 kg ha 1, corresponding to 0.67% per year. However, gains were clearly nonlinear. Those results were analyzed in the light of the history of the breeding program (see the Discussion Section) and the years 1992 and 2002 were taken as break points, dividing the period under study in three different phases (Fig. 1 and Table 4). In the first phase, from 1984 to 1992 (eight steps), no significant gain was observed for GY. Although there was a short stretch of gains from 1987 to 1990, those gains were not retained. The year of 1991 stands out of the series, with only three lines entering VCU trials and presenting very low GY. The mean of the group 1992 was nearly the same from the beginning of the series. In the second phase, from 1992 to 2002 (ten steps), some trend of increase in GY was observed. In those 10 yr, genetic improvement added an average of 15.7 kg ha 1 yr 1 (p < 0.05), representing 0.53% per year. The third phase, from 2002 to 2009 (seven steps), presented the highest estimate of genetic gain in the period, with a slope of 45.0 kg ha 1 yr 1 (p < 0.01), representing 1.44% per year. Group 2008 presented the highest mean of the period under study, and group 2009 had the second highest mean. No trend of ceasing gains for GY was observed. Parametrically, the total genetic gain in the period under study was kg ha 1, being kg ha 1 from the second phase and kg ha 1 from the third phase. Therefore, 2/3 of the total gain for GY in the 25-yr period was achieved in the last seven improvement steps. CROP SCIENCE, VOL. 51, MAY JUNE

5 Table 3. Estimates of components of variance for random factors in the mixed model. Random factor Number of columns in Z Grain yield Days to flowering Plant height Year Trial within year Block within trial within year Line within group Residual [3476] Z, the design matrix of random factors. Figure 1. Genetic gain for grain yield in the upland rice breeding program. Black dots represent the mean of groups of lines debuting in the value for cultivation and use (VCU) trials. Open circles represent best linear unbiased predictor (BLUP) for cultivars according to their group. Generalized linear regression lines and equations are given for group means on years for three phases within the period under study. ns, nonsignifi cant; *, signifi cant at the 0.05 probability level; **, signifi cant at the 0.01 probability level. Table 4. Estimates of parameters for the generalized linear regression of grain yield, days to flowering, and plant height on years in the period of 1984 to Grain yield (kg ha 1 ) Days to flowering Plant height (cm) Intercept Slope (gain yr 1 ) 6.55 NS * ** *** *** Standard deviation t value p-value < < Relative gain 0.53% 1.44% 0.29% 0.49% *Signifi cant at α = 0.05 by the t test. **Signifi cant at α = 0.01 by the t test. ***Signifi cant at α = by the t test. NS, not signifi cant CROP SCIENCE, VOL. 51, MAY JUNE 2011

6 Figure 2. Genetic gain for (a) days to fl owering and (b) plant height in the upland rice breeding program. Black dots represent the mean of groups of lines debuting in the value for cultivation and use (VCU) trials. Open circles represent best linear unbiased predictor (BLUP) for cultivars according to their group. Generalized linear regression line and equation are given for group means on years. ***, signifi cant at the probability level. Genetic Gain for Days to Flowering The cycle of the lines from sowing to flowering presented a trend of reduction at a rate of 0.25 d yr 1 (p < ). In the whole period, this change represented a reduction of 6.35 d in the duration of the crop (Fig. 2a). The reduction in DTF was due more to the selection against lateflowering lines than for selection for very early-flowering lines. Since 2001, the DTF mean remained almost constant, around 80 d. The cycle of released cultivars showed an interesting trend (Fig. 2a). Until 1995, there were two clearly distinct groups: the early-flowering cultivars, with DTF 75 d, and the intermediate-cycle cultivars, with DTF 85 d. No cultivars from that period presented DTF between 75 and 85, whereas after the year 2000 all released cultivars and lines in the pipeline for release presented flowering time within that range. Genetic Gain for Plant Height Plant height was reduced at a mean rate of 0.52 cm yr 1 (p < ), which represents a reduction of 13 cm for the whole period (Fig. 2b). Until 1995, most group means were above 100 cm, whereas after that year most group means were below that value. In the period 2003 to 2007, mean PH increased at a pace of approximately 3 cm yr 1. The groups 2008 and 2009 reverted most of this problem, presenting mean plant height of approximately 95 cm. The cultivars released by the program presented PH varying from 88 to 120 cm. The range of variation was larger in the early phase, before The reduction of the mean PH observed through time was due to selection against tall plants rather than selection for very short plants. DISCUSSION This study describes the facts related to the largest upland rice breeding program in Brazil, for a period of a quarter century, in which changes in plant type, grain type, genetic background, and breeding methods were observed (Pinheiro et al., 2004). A very large data set was used, representing a wide range of years, locations, input levels, and genotypes, within the broad TPE represented by the upland rice crop in the center-north of Brazil. The data set used in this meta-analysis was highly unbalanced, since different lines were evaluated in different locations and years. The 15 most important check varieties represented 10,471 data points (22.9% of the data), whereas 57 lines had fewer than 10 data points each. The mixed model analysis used in this study is considered robust against data unbalance and generates estimates for fixed effects and predictors for random effects that are unbiased and have minimum error variance (BLUE and BLUP; Searle, 1971). The expectation of the BLUEs of groups are free of effects of design factors, including years; CROP SCIENCE, VOL. 51, MAY JUNE

7 hence, they can be taken as estimates of the genetic value of the program in a given year. The slope of the linear regression of those estimates on the time represented the mean genetic gain of the program. The genetic gain for grain yield computed in this study was satisfactory, especially considering that it accelerated in recent years (Fig. 1). Compared to previous estimates from other crops, such as wheat (Triticum aestivum L.), sunflower (Helianthus annuus L.), sorghum [Sorghum bicolor (L.) Moench], and maize (Zea mays L.) (Table 1), the gain of 45 kg ha 1 (1.44%) per year, observed in the third phase, is relatively high. However, if the whole period is considered, the mean gain of 19.1 kg ha 1 yr 1 (0.67%) is not as impressive. A previous study on the genetic gain of upland rice (Soares et al., 1999), for a state level breeding program in collaboration with Embrapa, estimated the progress per year at 23.8 kg ha 1 for early-flowering lines and 46.4 kg ha 1 for intermediatecycle lines. That state was not covered in the present study; nevertheless, the results found here are similar in magnitude. Irrigated rice, for comparison, has presented limited progress for yield potential in the post-green revolution era. Muralidharan et al. (2002) found no gains for yield in a meta-analysis of data from 21 yr of INGER (International Network for Germplasm Evaluation of Rice) trials in India although, due to the reduction observed in crop duration, yield per day has increased. The genetic gain for irrigated rice breeding has been computed for two different regions in Brazil, considering only semidwarf germplasm. In the mid-north (the states of Maranhão and Piauí), the gain per year was only 18 kg ha 1 (Rangel et al., 2000), whereas for the northeast the gain was estimated at 54.9 kg ha 1 (Breseghello et al., 1999). Those results are of similar magnitude compared to the progress in this study for upland rice. However, when the transition from traditional varieties to modern varieties is included in the series, as in Peng et al. (2000), the mean gain of irrigated rice is significantly higher. Wheat is a crop with similar mean yield compared to upland rice; thus it is interesting to compare the genetic gains for both species. Gains reported for bread wheat in Brazil are of 48 kg ha 1 yr 1 (Cargnin et al., 2008), approximately the same value found for the third phase in this study ( ). Estimates of gain for wheat in China (Zhou et al., 2007a, b) varied from 14 to 72 kg ha 1 yr 1. On average, those results indicate that upland rice breeding in Brazil must sustain the gains of the third period for keeping up with the pace of wheat improvement. Phases of the Breeding Program Three phases of the breeding program were recognized in this study. A similar trend was found by Abeledo et al. (2003) for barley (Hordeum vulgare L.) improvement in Argentina, with a phase of null gains followed by a phase of linear progress. Phase I ( ) was a period of trial and error, when a large number of lines were tested but no genetic gain for GY was achieved. In this period, lines varied widely for DTF and PH, revealing a lack of a clear ideotype to be selected in the TPE. Brazilian landraces of upland rice belong to the tropical japonica subspecies, descending from varieties introduced by Portuguese settlers (Pereira, 2002). This germplasm is characterized by tall plants with droopy leaves, moderate tillering, and large panicles, which are vigorous and weed competitive. Those varieties were desirable for low input farming, with manual weeding and harvesting; however, they presented low harvest index and were prone to lodging. Most cultivars from the first phase (e.g., Guarani, Rio Paraguai, Xingu) presented characteristics of the traditional germplasm and are now considered obsolete. Phase II ( ) was marked by a strong influence of introduced materials, especially from CIAT and the United States. CIAT s influence was more important on plant type although having also influenced grain shape and amylose content. United States materials were introduced with the main purpose of improving grain quality. The preference for shorter plants in the second phase of the program reflects a change of focus toward a high-input, fully mechanized cropping system, in which lodging resistance was a priority. During phase II, a great deal of novel variability was introduced within the scope of a collaborative breeding program between Embrapa and CIAT (Morais et al., 2006). CIAT upland materials, although based on japonica background, had some introgression of indica genome (e.g., from the parent Colombia 1) and revealed highly competitive in Brazil, resulting in seven cultivars released (Progresso, Maravilha, and Canastra, resulting from lines introduced in Phase I, and BRS Bonança, BRS Carisma, BRS Talento, and BRSMG Curinga, from lines introduced in Phase II). This wave of introgression in the Brazilian program resulted in a reduction in the mean stature, abundant and upright tillering, short and erect fl ag leaves, and lower panicles in the canopy, contributing to a significant reduction in lodging proneness. Those traits were readily incorporated into the ideotype sought by Brazilian s breeders. Gains for GY in Phase II were faster until 1998; however, most of it was reversed in a linear decay in the period of 1998 to The reason for that decay in mean yield was the ceasing of the supply of high yielding CIAT lines, while the lines resulting from the use of CIAT materials as parents in Embrapa s program were not yet in VCU trials. Nevertheless, the reduction in DTF and PH observed in phase II added to the overall cropping system productivity, with a trend of increasing input use at farmer level. Tabien et al. (2008) also found a significant reduction in cycle duration and plant stature in U.S. rice varieties, which CROP SCIENCE, VOL. 51, MAY JUNE 2011

8 implies that those traits are favorable both in irrigated and in rainfed rice systems. Interestingly, grain characteristics were an important limiting factor for the adoption of cultivars directly derived from CIAT introductions. Moderate levels of kernel chalkiness and slightly sticky cooked grains have been perceived by consumers as relevant shortcomings. The three most important Embrapa cultivars currently in the market, BRS Sertaneja, BRS Monarca, and BRS Pepita, all resulted from crosses between CIAT parents and other materials used to correct those problems. Phase III ( ) represents the current stage of Embrapa s breeding program, with emphasis on selection for GY while keeping other traits within a suitable range. The genetic gain for GY in the third phase was the highest (Table 4). Plant stature and cycle are under stabilizing selection; therefore, those traits are expected to remain steady in the near future, with DTF around 80 d and PH around 95 cm. The linear increase in PH in the period from 2003 to 2007 raised concerns about reverting long term reduction for this trait and, for that reason, truncated selection for PH 100 cm is being applied. On the other hand, very short plants tend to be less productive and less weed competitive, so PH is expected to stabilize between 90 and 100 cm. There was a clear trend of reduction in the number of lines tested at VCU level through time (Table 2), reflecting a more focused program with a clear ideotype. Several lines from the third phase are in the pipeline for release, for which GY estimates are above the best current cultivars. The introduction of those new lines is expected to contribute to increase the upland rice yield at farmer level in Brazil. An alternative way of looking at the progress of the breeding program would be based on the variation of GY of the top-yielding cultivars released by the program. In this sense, Embrapa s program has consistently pushed the yield ceiling upward through time (Fig. 1). Excepting the cultivar Mearim, which has very long cycle duration, the topyielding cultivars frequently superseded previous records for GY, with a trend of acceleration in the later phases. Factors Promoting the Acceleration of the Genetic Gain for Grain Yield Some explanations can be offered for the faster gain for GY in Phase III. One of them is the application of a modified pedigree breeding scheme, in which crosses are selected for grain yield in F 2 followed by plant selection within F 1:3 families. This procedure is repeated in the F 5 generation (Morais et al., 2006). This scheme combines early selection for yield potential at the family level with ideotype selection at the single-plant level. Furthermore, the program adopted a recurrent selection philosophy in which selected plants are recombined to build elite populations with high potential for line extraction. New variability is tested carefully before introduction in the elite gene pool, avoiding yield setbacks. Breseghello et al. (2009) estimated the genetic progress through recurrent selection on one upland rice synthetic population, finding a gain of 103 kg ha 1 yr 1 (3.6% of the population mean) and showing the high potential of this approach for rice breeding. Finally, intense selection pressure is being applied on yield, since most of the elite population presented adequate plant type, cycle, and grain quality. The example from hybrid maize breeding shows that continuous gain for yield can be achieved through systematic selection in elite germplasm (Duvick, 2005). Embrapa s upland rice breeding program is progressively becoming a maize-like program. Factors Limiting the Genetic Gain for Grain Yield Some factors represented a restraint in the rate of genetic progress for yield. Disease resistance, especially blast resistance, has been a major selection factor during the period under study. Rice blast (Magnaporthe oryzae) is the most aggressive pathogen of upland rice in Brazil (Filippi et al., 2007) such that blast susceptibility has always been a major factor for discarding breeding lines or preventing crosses between high-yielding cultivars that lack blast resistance. On the other hand, considering that VCU trials are not treated with fungicides, gains for disease resistance probably contributed to the observed gains for GY. There is a project under way for implementing markerassisted selection in the routine of the breeding program, which could allow further enhancement of genetic gains for grain yield. The genetic gains reported here resulted exclusively from classical methods, since the application of molecular markers have not yet influenced the quality of the elite lines evaluated in final trials. Concomitantly with the changes in GY, DTF, and PH, a major shift in grain type was produced during the period under study (Fig. 3). The grain type of the traditional upland varieties was long and wide with high chalkiness and low amylose content, resulting in relatively sticky rice after cooking. In the 1980s, those characteristics begun to be rejected in the market, such that the upland rice product had a price penalty compared to irrigated rice. In response to that, high-quality varieties from the United States (e.g., Bluebelle, Lebonnet) were used in crosses for steering Embrapa s upland rice program toward long slender kernels. Strong selection was applied for grain shape, intermediate amylose content and gelatinization temperature, high vitreousness, soft and nonsticky cooking quality, high milling yield at different harvesting dates, glabrous and well closed husks, absence of awns, and nonshattering (neither too hard to thresh) panicles. Selection for grain quality during most of the period covered by this study certainly had a significant trade-off CROP SCIENCE, VOL. 51, MAY JUNE

9 Figure 3. Samples of the old upland rice grain type (cultivar Guarani, released in 1986) and the modern grain type (cultivar BRS Primavera, released in 1998). on GY progress. The current elite breeding population has converged to a very strict grain ideotype, and for this reason introduction of novel variability must be done with caution. Another limiting factor has been the vast geographic region represented by the TPE of Embrapa s upland rice breeding program, which encompasses a large range of soil, climate, and crop management variation. Considering that elite lines must perform well for the average of those locations, the variance related to genotype environment interaction is not well exploited (Pacheco et al., 2005). Subprograms focused on more specific TPEs could result in faster genetic gain for GY. However, the low geographic density of the upland rice crop in Brazil makes seed companies normally favor cultivars of broad adaptation with more marketing options. CONCLUSIONS The upland rice breeding program conducted by Embrapa and partners resulted in increase in grain yield and reduction in plant height and duration of plant cycle in the period of 1984 to The gain for grain yield was highest in recent years, showing no sign of yield plateau. During the same period, great changes have been promoted in plant type and grain quality, adding value to modern upland rice cultivars. Acknowledgments We thank all the research assistants and field workers that helped generating the data used in this study. Special thanks to Dr. Antônio Carlos de Oliveira, for the review on the statistical analysis. References Abeledo, L.G., D.F. Calderini, and G.A. Slafer Genetic improvement of barley yield potential and its physiological determinants in Argentina ( ). Euphytica 130: Breseghello, F., O.P. Morais, E.M. Castro, A.S. Prabhu, P.Z. Bassinello, J.A. Pereira, M.M. Utumi, M.E. Ferreira, and A.A. Soares Recurrent selection resulted in rapid genetic gain for upland rice in Brazil. Int. Rice Res. Newsl. 34:4. Breseghello, F., O.P. Morais, and P.H.N. Rangel A new method to estimate genetic gain in annual crops. Genet. Mol. Biol. 21: Breseghello, F., P.H.N. Rangel, and O.P. Morais Ganho de produtividade pelo melhoramento genético do arroz irrigado no nordeste do Brasil. (In Portuguese, with English abstract.) Bras. J. Agric. Res. 34: Cargnin, A., M.A. Souza, and V. Fronza Progress in breeding of irrigated wheat for the Cerrado region of Brazil. Crop Breed. Appl. Biotechnol. 8: Duvick, D.N The contribution of breeding to yield advances in maize (Zea mays L.). Adv. Agron. 86: De La Vega, A.J., I.H. Delacy, and S.C. Chapman Progress over 20 years of sunflower breeding in central Argentina. Field Crops Res. 100: De Vita, P., O.L.D. Nicosia, F. Nigro, C. Platani, C. Riefolo, N. Di Fonzo, and L. Cattivelli Breeding progress in morpho-physiological, agronomical and qualitative traits of durum wheat cultivars released in Italy during the 20th century. Eur. J. Agron. 26: Filippi, M.C.C., G.B. Silva, and A.S. Prabhu Indução de resistência à brusone em folhas de arroz por isolado avirulento de Magnaporthe oryzae. (In Portuguese, with English abstract.) Fitopatol. Bras. 32: Mason, S.C., D. Kathol, K.M. Eskridge, and T.D. Galusha Yield increase has been more rapid for maize than for grain sorghum. Crop Sci. 48: Morais, O.P., P.H.N. Rangel, P.R.R. Fagundes, E.M. Castro, P.C.F. Neves, V.A. Cutrim, A.S. Prabhu, C. Brondani, and A.M. Magalhães, Jr Melhoramento genético. p In A.B. Santos, L.F. Stone, and N.R.A. Vieira (ed.) A cultura do arroz no Brasil. 2nd ed. (In Portuguese.) Embrapa Arroz e Feijão, Santo Antônio de Goiás, Brazil. Muralidharan, K., G.S.V. Prasad, and C.S. Rao Yield performance of rice genotypes in international multi-environment trials during Curr. Sci. 83: Pacheco, R.M., J.B. Duarte, R. Vencovsky, J.B. Pinheiro, and CROP SCIENCE, VOL. 51, MAY JUNE 2011

10 A.B. Oliveira Use of supplementary genotypes in AMMI analysis. Theor. Appl. Genet. 110: Peng, S., R.C. Laza, R.M. Visperas, A.L. Sanico, K.G. Cassman, and G.S. Khush Grain yield of rice cultivars and lines developed in the Philippines since Crop Sci. 40: Pereira, J.A Cultura do arroz no Brasil: Subsídios para a sua história. (In Portuguese.) Embrapa Meio Norte, Teresina, Brazil. Pinheiro, B.S., E.M. Castro, and C.M. Guimaraes Sustainability and profitability of aerobic rice production in Brazil. Field Crops Res. 97: Pinheiro, B.S., E.M. Castro, O.P. Morais, and F. Breseghello Developing aerobic rice in Brazil. In K. Toriyama, K.L. Heong, and B. Hardy (ed.) Rice is life: Scientific perspectives for the 21st century. Proc. World Rice Res. Conf., Tokyo and Tsukuba, Japan. 4 7 Nov [CD-ROM]. International Rice Research Institute, Los Baños, Philippines, and Japan International Research Center for Agricultural Sciences, Tsukuba, Japan. Rangel, P.H.N., J.A. Pereira, O.P. Morais, E.P. Guimarães, and T. Yokokura Ganhos na produtividade de grãos pelo melhoramento genético do arroz irrigado no meio-norte do Brasil. (In Portuguese, with English abstract.) Bras. J. Agric. Res. 35: SAS Institute SAS/STAT 9.1 user s guide. SAS Inst., Cary, NC. Searle, S.R Linear models. John Wiley & Sons, Hoboken, NJ. Soares, A.A., P.G. Santos, O.P. Morais, P.C. Soares, M.S. Reis, and M.A. Souza Progresso genético obtido pelo melhoramento do arroz de sequeiro em 21 anos de pesquisa em Minas Gerais. (In Portuguese, with English abstract.) Bras. J. Agric. Res. 34: Tabien, R.E., S.O.P.B. Samonte, and A.M. McClung Fortyeight years of rice improvement in Texas since the release of cultivar Bluebonnet in Crop Sci. 48: Vencovsky, R., A.R. Morais, J.C. Garcia, and N.M. Teixeira Progresso genético em 20 anos de melhoramento de milho no Brasil. p In Congresso Nacional de Milho e Sorgo, 16, Belo Horizonte, Brazil. 4 8 Aug (In Portuguese.) Embrapa CNPMS, Belo Horizonte, Brazil. Zhou, Y., X.X. Sui, X.C. Xia, X.K. Zhang, Z.H. He, and G.S. Zhang. 2007b. Genetic improvement of grain yield and associated traits in the northern China winter wheat region from 1960 to Crop Sci. 47: Zhou, Y., H.Z. Zhu, S.B. Cai, Z.H. He, X.K. Zhang, X.C. Xia, and G.S. Zhang. 2007a. Genetic improvement of grain yield and associated traits in the southern China winter wheat region: 1949 to Euphytica 157: CROP SCIENCE, VOL. 51, MAY JUNE