Selection for sugar content at different times of the year in final stage trials in sugarcane breeding programs

Commercial sugarcane in Australia is harvested and subsequently crushed over a 20- 24 week period, typically between June and November. Increases in cane supply placed a strain on mill capacity, which is expensive to increase. Consequently, extension of the harvesting period was required. Earlier initiation of the harvesting period is the favoured industry option to alleviate mill capacity constraints. However, commercial cane sugar (CCS) levels are usually low very early in the harvesting season, and this reduces profitability at this time of the year. Therefore, low CCS early in the harvesting period and immediately before that reduces profitability and is regarded as an industry wide impediment to an earlier initiation of the harvesting period. Various crop management options could assist in addressing low early CCS, but genotypes are expected to play a pivotal role. Breeding research has identified high levels of genetic variation for early CCS, with measurable gains from targeted recurrent selection realised in early stage selection trials. However, current selection systems in industry breeding programs may not fully exploit these gains, particularly in final stage trials. Data on CCS were collected from different series of final stage trials at various predetermined times over random locations and crop-years (Central region), and predetermined times and N rates and random crop-years (Burdekin region). Additionally, cane yield (TCH) was measured twice per year for each location. These data were used to determine if selection systems focused specifically on early (before or during early-July) and mature (after early-July) periods would maximise genetic gain in each of these two sub-target periods (early and mature), as opposed to selection across both sub-periods for broad adaptation. Further, genotype x environment interaction effects, where environments included combinations of sample time, locations, N rates, crop-years, and crop ages, were quantified and their impact on selection decisions within and across the sub-periods established. Genotypes utilised in this research were considered representative of those undergoing assessment in final stage selection trials, after selection from original seedling populations across two previous selection stages. The experimental design involved numerous repeated measures datasets being collected, with the same plot sequentially sampled without re-randomisation. Analyses on such data in sugarcane research published to date have not evaluated alternative covariance and correlation structures between times in the residual matrix of such data. Thus, novel (to sugarcane) covariance structures to model the residual matrix of repeated measures datasets were examined to ensure application of the most appropriate statistical model. However, the choice of covariance structure to model the residual matrix did not impact on estimated best linear unbiased predictors and variance components or their accompanying standard errors. Further, the most parsimonious covariance structure does not estimate pooled residual variance across environments, which is essential for subsequent estimation of heritability and other key breeding parameters (e.g. response to selection). Therefore, a model following a split-block in time ANOVA, assuming a simple covariance structure to model the residual matrix, was considered satisfactory for CCS data collected within and across crop-years, and yield data collected across crop-years. Genotype x time interaction effects on CCS were small compared with genotype main effects, leading to high genetic correlations between times within each sub-period (early and mature). However, genetic correlations for sample times ≥16 weeks apart were considerably lower, and sample times within locations generally clustered according to sub-period (early vs. mature). However, genotype x sub-period interaction effects on CCS and TCH were not important across random locations and crop-years in the Central region, or fixed N rates and random crop-years in the Burdekin region, because a large portion of total genotype x environment interaction effects were retained within each sub-period. Therefore, subdivision of a selection program into two components, focused on early (pre-July) and mature (July and after) would not significantly enhance genetic gains. However, as genetic variance and response to selection were greatest during the early sub-period, an advantage in overall genetic gain would be realised if final stage trials were assessed early in the harvesting period. For the Central region, the results indicated that final stage trials should be assessed over multiple locations and crop-years, but with only one sample time per crop-year required for measurement of CCS and TCH. In the Burdekin region, significant genotype x N rate interaction effects influenced selection for TCH (especially in ratoon crops), but not CCS. Some of the genotype x N rate interaction effects were attributed to increased lodging, which increased with applied N. The genotype x N rate interaction effects were mainly due to divergent genotype performance for the lowest compared to recommended (or higher) N rates. These results are of practical significance in that they highlight the importance of maintaining adequate levels of N in selection trials, with the medium N rate (corresponding to current industry recommended rates of application) providing optimal growers returns. Direct selection for genotype performance at recommended rates of N fertilizer was found to be optimal. Construction of managed environments by lowering N rates to reduce lodging and possibly decrease error variation would be inappropriate because the genetic correlation between the low and medium N rates was near zero. However, the combination of small genotype x time interaction and unimportant genotype x crop age interaction effects on CCS indicated that effective selection for CCS in final stage trials could occur early in the year at a young crop age, prior to lodging. Overall, the results indicated that final stage trials do not have to be conducted in each of two sub-periods (early and mature) to optimise gains from selection. However, selection gains would be maximised by selection based on CCS measured relatively early in the harvesting season. Conducting final stage trials across multiple locations and crop-years is important because of large genotype interactions with these factors. Recommended rates of N fertilizer should be applied with particular care to avoid sub-optimal levels of N