HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Milligan, S. B. Articles by Bischoff, K. P. Search for Related Content PubMed Articles by Milligan, S. B. Articles by Bischoff, K. P. Agricola Articles by Milligan, S. B. Articles by Bischoff, K. P. Related Collections Biometrics Field evaluation techniques Sugarcane

CROP BREEDING & GENETICS

Early Stage Sugarcane Selection Using Different Plot Sizes

^a Seminis, P.O. Box 249, Felda, FL 33930
^b National Univ. of Cordoba, CC 509, 5000-Cordoba, Argentina
^c St. Gabriel Research Station, Louisiana State Univ. Agricultural Center, 5755 LSU Ag. Rd., St. Gabriel, LA 70776

^* Corresponding author (scott.bradley.milligan{at}seminis.com).

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
Most sugarcane (Saccharum spp.) cultivar development programs use single-row plots in their first clonal trials. We hypothesized that a larger plot size would increase the accuracy of selection and compared selection efficiencies of 1.82-, 3.35-, and 4.88-m single-row plots. The 1.82-m plots generated larger genetic and residual variances than the larger plot sizes for sugar and cane yields and stalk number and weight but not for sugar concentration. Broad-sense heritabilities for yield components for the examined plot sizes differed little. Genetic correlations for the same trait among the plot sizes were high for most traits except stalk number. Consequently, the correlated response in larger plots to selection for the same trait selected in small plots was not affected by plot size. The proportion needed to confidently retain the top 1% of the genotypes was also not affected by plot size but was very high (>80%) for sugar and cane yields. It was substantially lower for sugar concentration and stalk weight (<43%), whereas it was higher (43–55%) for stalk number. The estimated probability of retaining the top genotypes by selecting the top 33% of the population was 59 to 66% for sugar and cane yields, but it ranged from 77 to 90% for the other yield components. Thus, increasing the plot size would not improve the selection efficiency of the program, but the study highlighted that there was still room for improvement in the initial stage of selection.

Abbreviations: LSU AgCenter, Louisiana State University Agricultural Center

Early Stage Sugarcane Selection Using Different Plot Sizes

^a Seminis, P.O. Box 249, Felda, FL 33930
^b National Univ. of Cordoba, CC 509, 5000-Cordoba, Argentina
^c St. Gabriel Research Station, Louisiana State Univ. Agricultural Center, 5755 LSU Ag. Rd., St. Gabriel, LA 70776

^* Corresponding author (scott.bradley.milligan{at}seminis.com).

Most sugarcane (Saccharum spp.) cultivar development programs use single-row plots in their first clonal trials. We hypothesized that a larger plot size would increase the accuracy of selection and compared selection efficiencies of 1.82-, 3.35-, and 4.88-m single-row plots. The 1.82-m plots generated larger genetic and residual variances than the larger plot sizes for sugar and cane yields and stalk number and weight but not for sugar concentration. Broad-sense heritabilities for yield components for the examined plot sizes differed little. Genetic correlations for the same trait among the plot sizes were high for most traits except stalk number. Consequently, the correlated response in larger plots to selection for the same trait selected in small plots was not affected by plot size. The proportion needed to confidently retain the top 1% of the genotypes was also not affected by plot size but was very high (>80%) for sugar and cane yields. It was substantially lower for sugar concentration and stalk weight (<43%), whereas it was higher (43–55%) for stalk number. The estimated probability of retaining the top genotypes by selecting the top 33% of the population was 59 to 66% for sugar and cane yields, but it ranged from 77 to 90% for the other yield components. Thus, increasing the plot size would not improve the selection efficiency of the program, but the study highlighted that there was still room for improvement in the initial stage of selection.

Abbreviations: LSU AgCenter, Louisiana State University Agricultural Center

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
RESEARCHERS IN THE early selection stages of a sugarcane (Saccharum spp.) breeding program invariably screen a large number of individual genotypes using small plots. The use of small plots is necessitated by resource limitations of land and seed and the need to evaluate large numbers of genotypes. The challenges inherent in using small plots are well known and generally relate to competition effects for light and other nutrients (Duncan, 1969; Tovey et al., 1973; Jackson and McRae, 2001) or deal with the minimum effective sample size to measure a trait (Wassom and Kalton, 1953; Stickler, 1960).

Each year, the Louisiana State University Agricultural Center's (LSU AgCenter) sugarcane cultivar development program selects approximately 3000 genotypes from the initial seedling population to plant in its first clonal trial. Each clone is planted using two stalks, placed side by side, into single-row plots 1.82 m long with a 60-cm distance between plots within a row and 1.82 m between rows. From the first clonal population, about one-third of the clones are selected and advanced to the second clonal trials. Second clonal trials are planted with six stalks (two stalks side by side) into single-row, 4.88-m-long plots using 1.2-m spacing between plots within the row. Single-row 1.82-m plots typically provide two or three stools of cane to assess the cane yield potential. Because observable end-row effects (shorter and more numerous stalks) are detectable within 1 m from the end, the entire 1.82-m plot is subject to an end-row border effect.

The use of the larger plots after the first clonal stage is based on the assumption that larger plots, such as the 4.88-m plots, assess yield better than the 1.82-m plots used in the first clonal trials. Researchers have investigated the effects of competition in sugarcane small-plot evaluation trials (Arceneaux, 1929, 1939; Skinner, 1961; Skinner and Hogarth, 1978; Jackson and McRae, 2001). Most of the cited studies generally evaluated plots used in the advanced stages of the selection program where the number of entries is relatively small and seed is not a limiting factor. Only one report dealt directly with the plot size (Jackson and McRae, 2001). Jackson and McRae (2001) compared one-row and two-row unbordered plot yields with the yields from large, well-bordered plots where they collected data from the center two rows of six-row plots. The plots used in the study were, however, considerably larger (10.6 m long) than used in the initial stages of selection in the LSU AgCenter program. It also reflects some of the range of first clonal plot sizes used in different sugarcane breeding programs.

Jackson and McRae (2001) observed inflated genetic and residual variances in unbordered plots compared with bordered plots, apparently due to competition. Nevertheless, heritabilities of the yield components were not significantly affected. Genetic correlation of cane yield between the unbordered plots and the well-bordered plots was variable and generally weak (r_g = 0.49). Correlation between unbordered and bordered plots for sugar content was much higher (r_g = 0.91). They concluded that selection among genotypes using smaller unbordered plots should place greater emphasis on sugar content than on cane yield. They noted that selection based on sugar content gave equal or larger gains compared with other selection criteria, including the relative economic value (essentially the predicted profit).

The purpose of this work was to determine effects of alternative plot sizes on the qualities of yield estimate and selection effectiveness. The plot configurations compared were limited to ones deemed practical in the current selection program and were hence considerably smaller than those compared in the Jackson and McRae (2001) study.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
Twenty-seven clones randomly selected from the 1993 crossing series, first clonal trial were used in this study (Schexnayder et al., 1993). The clones originated from 14 locally adapted, randomly selected biparental crosses. Trials were planted at two locations in the fall of 1997, one at the LSU AgCenter's St. Gabriel Research Station near St. Gabriel, LA, and the other at the USDA-ARS Spanish Trail Farm near Chacahoula, LA. The St. Gabriel location was replanted in the fall of 1998. We planted the 27 clones plus three commercial check cultivars in a randomized block design using two replications per location with a split-plot treatment arrangement using three plot sizes as main-plot treatments. The three plot sizes were a 1.82-m long plot with a 0.6-m end gap (Stage 1 plot size convention), a 3.35-m plot with a 0.9-m end gap (plot size intermediate between those used in Stages 1 and 2), and a 4.88-m plot with a 1.2-m end gap (Stage 2 plot size convention). We used two stalks for the 1.82-m plot, four stalks for the 3.35-m plot, and six stalks for the 4.88-m plot. Sugarcane experimental plots in Louisiana are typically planted with two running stalks—two stalks placed side by side in the planting furrow. Because plot length and gap varied, we buffered each main-plot treatment with similar-sized buffer plots and gaps using a commercial cultivar. All plots were on single rows spaced 1.8 m apart, which is the standard commercial row spacing in Louisiana.

Because Tropical Storm Francis severely lodged the cane at the USDA-ARS site, we abandoned the test in 1998. The number of millable stalks per plot was recorded and 10-stalk samples were harvested from the middle of the plots in December 1998 and 1999. Sugar concentration (theoretical recoverable sugar) (g sugar kg^–1 cane) was determined from the samples using the Brix and saccharimeter readings (pol) as reported by Gravois and Milligan (1992). Cane yield (Mg ha^–1) was estimated from the product of stalk weight and stalk number, and sugar yield (Mg ha^–1) was calculated as the product of cane yield and sugar content.

We excluded the commercial check cultivars in all analyses. Using SAS Proc Mixed (SAS Institute, 2002), observations (Y) were first analyzed using the randomized block split-plot model with all terms considered random:

[1]

where T_i, R_j, and G_l refer to the test i, rep j, and genotype l effects. We then ran a simple one-way ANOVA for the plot size effect k on the residuals resulting from the analysis of Eq. [1] (Levene, 1960). This analysis tested plot size–effect residuals for heterogeneity of variance.

A pooled model including the plot size treatment effect and its interaction with the other effects was also analyzed:

[2]

where T_i, R_j, P_k, and G_l refer to the test i, rep j, plot size k, and genotype l effects. Plot size was considered fixed and all other terms were considered random. For this analysis, we assumed a heterogeneous residual variance (SAS Proc Mixed; SAS Institute, 2002; repeated/group = plot size). Individual analysis for each plot size was also performed, using a reduced model that excluded plot size–related terms to estimate genotype (_G²), genotype by test (_GT²), and residual (_E²) variance components.

Broad-sense heritabilities (h²) for each plot size k were determined as:

[3]

where _Gk² was the genetic variance of plot size k, _Pk² was the phenotypic variance of plot size k, _GTk² was the genotype by test variance of plot size k, and _Ek²/r was the residual variance of plot size k with r replications. The number of replications was assumed r = 1 to mimic typical selection conditions. Standard errors of the heritabilities were estimated according to Dickerson (1969) Genetic correlations (r_g) were obtained using SAS Proc Mixed (Holland, 2006).

A useful way to determine the effect of plot size is to assume a trait measured in the different plot sizes as different traits. We were interested in how well selection for a trait in the smaller plot sizes (xsm) affected a response for the same trait in the larger plot sizes (ylg). Correlated response to selection for the same trait in different plot sizes was estimated as (Falconer, 1981):

[4]

where CR_ylg is the correlated response to selection of trait y in a large plot by selecting for the trait in a small plot, i_xsm is the standardized selection intensity assumed as 33% for a large population (Becker, 1984) and applied in trait x in small plots, h_xsm is the square root of the heritability in the smaller plot, and _Pylg is the phenotypic standard deviation of the larger plot size.

To characterize the effect of plot size on selection decisions, we performed two additional analyses. In the first analysis, we set the type 1 error rate at ( = 0.05). The value of the 99th percentile individual (P₉₉) was predicted as: P₉₉ = µ_k + Z_{0.99 G,k} for each plot size k, where Z_0.99 was the 99th percentile for a normal distribution. We then calculated the critical point (CP_k) for each plot size k (k = 1.82, 3.35, 4.88 m) to retain a genotype in the high yielding subset:

[5]

where t_{(0.95, dfe)} is the t value for 1 – significance and error degrees of freedom (dfe). We assumed the experimental conditions for the error estimate, but assumed one replication to mimic typical first clonal information (dfe = 26, r = 1); _Ek² is the error variance component for plot size k. This procedure gives the smallest subset of genotypes one needs to retain and have a probability = 0.95 (P = 1 – ) of including the best 1% of the genotypes. We then calculated the proportion of the population (P_k) to retain with this confidence for each plot size k (dfe = 26), or P_k = [Z > (CP_k – µ_k)/_Gk].

The second characterization translated the results into more meaningful information for the breeder. Normally about one-third of the genotypes in the first clonal trial are advanced to the second clonal trial. We wanted to know the relative confidence we might expect using the different plot sizes at this selection rate. We equated the 66th percentile value as P₆₆ = µ + Z_{0.66 Gk} = P₉₉ – t_{(Pthird k, 26)}(2_Ek²/1)^1/2. The probability of retaining the highest 1% of the population by selecting the top third of the population using three plot sizes (P_{third k}) was associated with the t-value, or t_{(Pthird k, 26)} = (P₆₆– P₉₉)/(2_Ek²/1)^1/2.

	RESULTS AND DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
The magnitude of sugar yield, cane yield, and stalk number decreased as plot size increased while sugar content and stalk weight means were not affected by plot size (Table 1 ). What is of real interest is the variation of the traits. We observed significant residual variance differences among plot sizes for all traits except sugar concentration and stalk weight. This observation was the basis to assume a heterogeneous residual variance in the pooled model (Table 2 ). The analysis of residuals to test for a heterogeneous residual variance is a robust test (Levene, 1960). However, the observed differences in the estimated residual variances and the lack of overlap of the standard errors of the variance components suggests that the observed differences among plot sizes in residual variances for stalk weight are real. Stalks near the end of sugarcane plots are commonly shorter than the inside stalks and this likely explains the differences in variation. Even though stalk samples were collected from the inside of the plot, the 1.82-m plot limited the distance from the end one could sample.

The substantial inflation of the genetic variance in the smallest plots compared with the intermediate and larger plots suggested that the estimated genetic variance of the small plots was confounded with a substantial amount of genotype by end-row competition variance (Table 2). It could be construed that genotype by competition variance is a type of genotype by environment interaction variance; the environment being the plot size. However, the genotype by plot size interaction variance (_GP²), estimated in the pooled model, was not significant or substantial for any trait. Despite large apparent competition effects in the smallest plot size, similar heritability values among the plot sizes contraindicated a negative effect of using a small plot size.

Sugarcane breeders commonly assume selection decisions for cane yield-related traits (sucrose yield, stalk number), based on very small plots, such as the 1.82-m-long plot, are more erroneous than those based on the larger plot sizes. These selection errors would arise from both competition and using a plot size insufficient to obtain a reliable stalk number for a genotype. The observed differences attributed to plot size may be due to confounding of both end-row competition and the plot-length effects. Lateral competition was the same for all plot sizes. By design, the gap between plots within the row increased with plot length, but the proportion of the plot affected by its proximity to the end decreased with larger plot sizes. So smaller plots were not only closer to different genotypes on the plot ends, but also exposed a larger proportion to end-row effects. Sugarcane clones are generally shorter toward the ends of plots. Visual observation of sugarcane height in plots suggests that end-row effects extend about 1 m into the plot. The confounding of end-plot gap length and plot size prevents distinguishing their relative effects on trait evaluation, but it appears that most of the negative effects vanished with respect to the residual variances when the plot length and gap equaled or were greater than the intermediate plot length and gaps.

We observed high genetic correlations for all traits between plot sizes with the exception of stalk number between the two smaller plot sizes with the largest plot size (Table 3 ). The correlations for stalk number in the smaller plot sizes (1.82 and 3.35 m) with that in the 4.88-m plot size ranged from r_G = 0.64 to 0.70. Other than this observation, there appeared to be no obvious trend with plot size. This observation suggests some chance experimental irregularity with stalk number in the 4.88-m plot size treatment. The predicted correlated responses to selection confirmed the correlation observation that there seemed little or no negative effect of selecting in smaller plots.

To further characterize the confidence the breeder may have in selecting among genotypes grown in the different sized plots, we calculated the proportion of a population needed to retain with 95% confidence (P = 0.95) the highest 1% of population using three plot sizes. The proportion needed was quite high (79–92%) for sugar and cane yields, whereas the predicted proportion was substantially less (26–43%) for sugar concentration and stalk weight (Table 4 ). This calculation is generally proportional to the broad-sense heritability (Table 2) but gives a better sense of the confidence of selection. Surprisingly, the calculated proportions for cane yield increased with plot size and decreased with sugar content. Cane yield is stalk weight multiplied by stalk number. There was a small increase in the calculated proportion of stalk weight and stalk number needed to retain the top 1% for these traits. Cane yield being the product of these two traits apparently multiplied this observation.

When the selection percentage was assumed as the top one-third, the likelihood of retaining genotypes with the heaviest stalks or highest sugar content or stalk number was greater than 80% for most of the plot sizes (Table 5 ). We did not expect to observe a similar level of confidence among the three yield components. We expected to have more confidence with the sugar content–based selections than the stalk number–based selections because most reports of heritability for these yield components report a lower heritability for stalk number than sugar concentration or stalk weight (Hogarth, 1971; Kang et al., 1983). Cane and sugar yield confidence was about 60% when the top third of the population was selected. There was no apparent plot size effect on these confidence estimates.

The probabilities and confidences (Tables 4 and 5) are at the liberal end of the confidence spectrum (i.e., the proportions needed to keep the top 1% are probably larger than estimated in this study). We used t values in the predictions that reflect single comparison-type calculations. A more accurate approach would be to use a Dunnett's type comparison (control versus all clones), which would substantially lower the predicted confidences of selection. The selection rate at this stage is logistically fixed at about one-third of the population. It is expected that the relative confidence would not change with a more conservative t value.

In summary, small plot sizes inflated genetic and residual variances but seemed to have little effect on heritability or genetic correlations among the plot sizes examined except for stalk number. The lower correlations for stalk number between small and larger plot sizes compared with the other yield components suggested that the small plot sizes might not always identify the best genotypes with respect to stalk number but would suffice for stalk weight and sugar concentration. The first clonal stage cane yield is subjectively assessed in a manner that considers the variable nature of the small plot size for number; that is, if only one stool is present, its absolute stalk number may be low, but the vigor of the stool will be considered and the breeder may ignore the observation that the clones only produced one stool. The moderate probability (59–66%) of retaining the top 1% of the population for yield when selecting the conventional top one-third of the population, suggests that there is room for improvement in the selection process. The results reported herein, however, do not suggest that increasing the plot size, at least to the sizes studied here, will improve the likelihood of identifying the highest yield genotypes. A significant caveat to the study is that comparison to a well-bordered plot was not made. Comparisons to such a plot configuration needs be made to know if conclusions from this study with respect to plot size fully relate to the commercial target environment of a pure stand.

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
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.

Received for publication December 20, 2006.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 

• Arceneaux, G. 1929. Plot arrangement and some results of variety tests in Louisiana. Proc. Int. Soc. Sugar Cane Technol. 3:338–348.
• Arceneaux, G., I.E. Stokes, B.A. Belcher, and R.T. Gibbens, Jr. 1939. Border competition in sugarcane variety tests under Louisiana and Georgia conditions. Proc. Int. Soc. Sugar Cane Technol. 6:403–420.
• Becker, W.A. 1984. Manual of quantitative genetics. Academic Enterprises, Pullman, WA.
• Dickerson, G.E. 1969. Techniques for research in quantitative animal genetics. p. 36–79. In A.B. Chapman (ed.) Techniques and procedures in animal science research. American Society of Animal Science, New York.
• Duncan, W.G. 1969. Cultural manipulation for higher yields. p. 327–342. In J.D. Eastin et al. (ed.) Physiological aspects of crop yield. ASA, Madison, WI.
• Falconer, D.S. 1981. Introduction to quantitative genetics. 2nd ed. Longman, New York.
• Gravois, K.A., and S.B. Milligan. 1992. Genetic relationships between fiber and sugarcane yield components. Crop Sci. 32:62–66.[Abstract/Free Full Text]
• Hogarth, D.M. 1971. Quantitative inheritance studies in sugarcane: I. Estimation of variance components. Aust. J. Agric. Res. 22:93–102.[CrossRef][Web of Science]
• Holland, J.B. 2006. Estimating genotypic correlations and their standard errors using multivariate restricted maximum likelihood estimation with SAS Proc Mixed. Crop Sci. 46:642–654.[Abstract/Free Full Text]
• Jackson, P., and T.A. McRae. 2001. Selection of sugarcane clones in small plots: Effects of plot size and selection criteria. Crop Sci. 41:315–322.[Abstract/Free Full Text]
• Kang, M.S., J.D. Miller, and P.Y.P. Tai. 1983. Genetic and phenotypic path analysis and heritability in sugarcane. Crop Sci. 23:643–647.[Abstract/Free Full Text]
• Levene, H. 1960. Robust test for the equality of variances. p. 278–292. In I. Olkin (ed.) Contributions to probability and statistics. Stanford Univ. Press, Palo Alto, CA.
• SAS Institute. 2002. SAS/STAT release 9.1. SAS Inst., Cary, NC.
• Schexnayder, H.P., Jr., S.B. Milligan, and F.A. Martin. 1993. Photoperiod and crossing in the Louisiana "L" sugarcane variety development program. p. 4–24. In 1993 sugarcane research annual progress report. Louisiana State Univ. Agricultural Center, Baton Rouge, LA.
• Skinner, J.C. 1961. Sugar cane selection experiments: 2. Competition between varieties. Technical Communications, year 1961, no. 1. Sugar Experiment Stations Board, Brisbane, QLD.
• Skinner, J.C., and D.M. Hogarth. 1978. Efficiency of border rows in replicated sugar cane variety trials. Euphytica 27:629–643.[CrossRef][Web of Science]
• Stickler, F.C. 1960. Estimates of optimum plot size from grain sorghum uniformity trial data. Tech. Bull. 109. Kansas State Univ. of Agric. and Appl. Sci., Lawrence, KS.
• Tovey, D.A., K.T. Glasziou, R.H. Farquhar, and T.A. Bull. 1973. Variability in radiation received by small plots of sugarcane due to differences in canopy heights. Crop Sci. 13:240–242.[Abstract/Free Full Text]
• Wassom, C.E., and R.R. Kalton. 1953. Estimation of optimum plot size using data from bromegrass uniformity trials. Bull. 396. Iowa State College Agric. Exp. Stn., Ames, IA.

This Article Abstract Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Milligan, S. B. Articles by Bischoff, K. P. Search for Related Content PubMed Articles by Milligan, S. B. Articles by Bischoff, K. P. Agricola Articles by Milligan, S. B. Articles by Bischoff, K. P. Related Collections Biometrics Field evaluation techniques Sugarcane