Chapter 10
Practical Considerations

10.1 Introduction

As noted at the beginning of this book, plant breeding demands a range of skills, including good management and a multitude of scientific disciplines in combination, to achieve success. Plant breeding operations and evaluation of plant breeding lines will be conducted in laboratories, glasshouses and field situations. This chapter attempts to outline some of the practical realities in a plant breeding programme. Sections covered in this chapter examine: experimental designs, including the types of designs suitable for different parts of a plant breeding programme; glasshouse management and field management; and the applications that can be covered and managed using computers. Finally, this chapter considers some of the practical considerations of the actual cultivar release procedure.

10.2 Experimental design

It has been stressed previously that the basic operations of cultivar development can, for simplicity, be divided into three stages: producing genetic variation, selection among recombinants for desirable new cultivars with specific characteristics, followed by genetic stabilization, multiplication and seed certification. The following few sections are concerned particularly with the middle one of the three processes.

The aim in selection is to identify recombinants that are genetically superior to already existing cultivars. Superiority can be achieved by increased productivity (e.g. increased yield or better end-use quality), by making productivity less variable (e.g. reduced risk of crop failure by introduction of disease, insect or stress tolerance or resistance, earliness, dry down, and so on) or increased profit (e.g. reducing input costs by incorporation of disease resistance).

In each of the cases it will always be necessary to evaluate the performance of breeding lines for qualitative and quantitative characters. In some instances it is possible to select and screen for single gene characters without the complication of undue influences from environmental factors. However, it is widely accepted that virtually all quantitatively inherited characters (most often the ones with greatest commercial value, e.g. yield, quality and many durable disease resistances) are highly modifiable by the environment. Consider the observation of a single plant; the aim is to minimize the non-genetic effect from the equation:

where P is the phenotypic expression, G is the genotypic effect, E is the effect of environmental variables, is the effects attributable to the interaction of the genotype with the environment, and is a random error term associated with a single observation. In the evaluation of breeding material it is only possible to observe the phenotype (a combination of genotypic and environmental effects). The aim is to determine the genetic potential of each breeding line as accurately as possible, and hence it is necessary to either estimate or minimize the environmental and error effects on the phenotypes observed in the field. Achieving this demands careful use of a number of experimental designs.

Running a plant breeding programme is no different from organizing a whole series of scientific experiments, and therefore all aspects of the operation should be treated with the same care and detail that individual experiments require in terms of planning and handling. Good experimental design leads to knowledge of the accuracy of the data upon which evaluation and selection are based. The quality of information collected in a plant breeding programme is the key factor determining the success of the scheme, whether it is one based solely on traditional techniques or whether it incorporates molecular-based technology. The proper use of statistically valid designs can increase the heritability of the trait under selection, that is, it can enhance the ‘genetic signal’ relative to the ‘experimental noise’ ratio and therefore the rates of genetic gain, response to selection and likelihood of success. Unfortunately, often this is overlooked.

Sometimes, the set of trials that a breeding programme uses to gain insight into the performance of advanced breeding lines in a Target Population of Environments (the set of environments or geographical regions a breeder is breeding for) is referred to as multiple environment testing.

It is common to evaluate breeding lines (test entries) in comparison with existing cultivars (controls or checks) within the same trial. In some cases a single cultivar is used, but more often several cultivars are included in the evaluation trials. The choice and number of control entries is largely dependent upon the range and number of cultivars that are currently being grown in the target region for the new cultivars, the type of trial, and the number of evaluations that are to be made. For example, an evaluation trial may contain the highest yielding cultivar available to compare yielding performance, the best quality cultivar to provide a baseline for quality, a cultivar with disease resistance to evaluate response under disease pressure, and so on. When appropriate, it is often a good idea to include cultivars developed by competitors or those that currently are in high demand by the farmers. It is always desirable to include as many control entries as is possible within the restrictions imposed by the extra land used and the effort involved. It should also be noted that evaluation trials can be costly, and that the cost of an evaluation trial is often directly related to the number of total entries that are included. If several thousand breeding lines are to be evaluated then it is unnecessary to include only a few control entries. If, however, only a few lines are under trial, then it would be unwise to include many hundreds of control plots. A simple rule of thumb, which is often useful, is that if between 1 and 200 breeding lines are to be tested, the number of control plots (not always entries) should be about a tenth of the total number of trial entries, while if more than 200 lines are under evaluation, up to 1/20th of the total number in the trial should be controls.

A wide spectrum of possible designs is available, but only a limited number will be detailed here, namely:

• unreplicated designs;
• randomized complete block designs;
• factorial designs;
• split-plot designs.

10.2.1 Unreplicated designs

Unreplicated designs, as their name suggests, are experimental designs where test entries are not replicated and so appear only once at each testing location. There are, however, several (or indeed many) different options even when single replicate designs are used. These include:

• Non-randomized designs without control entries, where genotypes that are to be evaluated are arranged in plots in a systematic order (e.g. numerical order, alphabetical order) (Figure 10.1a). Only genotypes under test are evaluated, and there are no control (check) cultivars grown at the same time. It is not possible to obtain any estimate of error from this type of design or make any direct comparisons with known cultivars.
• Non-randomized designs with control entries, where the test entries are arranged in plots in a systematic order (as above), but control cultivars are interspaced amongst the test entry plots (Figure 10.1b). The control cultivars can be arranged in a systematic order (e.g. every 20 plots), or they can be allocated to plot positions at random. In most cases a number of control cultivars are included. Each control entry may also be replicated more than once in the whole design. Multiple entry plots of control cultivars can often be useful to determine an estimated error variance for the overall field trial.
• Randomized designs without control entries, where the test entries are arranged within the trial at random but no control entries are included (Figure 10.2a).
• Randomized designs with control entries, where the test entries are randomly allocated plot positions within the trial (Figure 10.2b). Control entries can also be randomized throughout the design (and often replicated in more than one plot) or they can be arranged in a systematic order (e.g. every 5th plot), again with the option of having replication only for the control entries.

Figure 10.1 Non-randomized single replicate plot designs (a) without control entries, and (b) with control entries arranged systematically throughout.

Figure 10.2 Randomized single replicate plot designs (a) without control entries, and (b) with control entries arranged systematically throughout.

The efficiency of evaluation trials will always be increased by randomization, and non-randomized trials should be avoided if at all possible and carefully considered before being used. Similarly, it is generally unwise to organize any breeding evaluation trials without including any control entries against which the test lines will be compared. Without these considerations the trials are generally uninformative and can often be misleading.

In the early generations of a plant breeding scheme, there may be many hundreds or thousands of genotypes to be tested, each with only a very limited amount of planting material available. In many breeding programmes, the first ‘actual’ field trials are conducted using head-row plots, where each plot has arisen from a single plant selection in the previous year. Where thousands of lines are to be tested, it may be extremely difficult to completely randomize each individual head-row, but randomization at this early-generation stage can greatly increase efficiency. One option is to utilize nested designs. For example, say that a canola breeding programme has 200 cross combinations to be evaluated and that there are 100 individual single plant selections taken at the stage. Therefore there would be 2,000 head-row plots that would be planted in the field. A randomized complete block (with control entries) would be very large. In addition, from a practical aspect, it is often difficult to examine a single row plot on its own. As an alternative the 200 crosses could be randomized into five replicate blocks, and the 100 single plant selections are grown as rows within cross blocks. Each cross, therefore, would be represented by five sub-blocks (groups) of 20 head-row plots (grown adjacent), and replicated five times throughout the whole trial.

If control entries are arranged in a systematic order it will be possible to make direct comparisons of individual test entries with the nearest control plot, which can have advantages. For example, it makes possible the analysis of the data collected using nearest neighbour techniques, where plot values are adjusted according to the performance of appropriate surrounding test entries. In order to increase the value from unreplicated designs, they can be combined with powerful statistical analyses to allow the handling of any systematic heterogeneity existing among experimental units and/or the spatial autocorrelation that could arise between neighbouring plots. One example is the use of autocorrelation models such as AR1XAR1, a two-dimensional spatial model consisting of separable first-order autoregressive processes along both columns and rows, which deals with autocorrelation patterns running along both dimensions of a trial (columns and rows). The ability to account for spatial correlation patterns in two dimensions generally provides a more efficient analysis of breeding trials. The availability of more effective statistical analyses has recently encouraged a growing number of breeding programmes to include, at least in some stages of testing, unreplicated trials as they allow the testing of a larger number of breeding lines within a fixed budget.

10.2.2 Randomized designs

Although it is possible to obtain an estimate of error variance from unreplicated (single replicate) designs which have multiple entries (i.e. replicates) of chosen control cultivars, it is more common, where possible, to replicate both test lines and control cultivars in order to have a better estimate of the average performance of each entry, along with the variance in its performance, as well as obtaining a better and more representative overall estimate of error variance.

Completely randomized designs

If there is no knowledge of fertility gradients or other environmental variation existing within a test area, many suggest that complete randomization be used to identify superior breeding lines. In such a design, each of the test and control entries are allocated at random to plot positions (Figure 10.3a). Each entry is repeated a number of times according to the required number of replicates. The error variance is estimated from the variance between replicate test entries.

Figure 10.3 (a) Completely randomized block design, and (b) randomized complete block design.

Randomized complete block designs

Although there are often merits in choosing a completely randomized design, a more common design (probably the most common design) used by plant breeders is a randomized complete block design, or variations thereof. In these designs, the total area of the test field is divided into units according to the number of required replicates. Each unit is called a block. Each of the test and control entries are randomly assigned plot positions within each block (Figure 10.3b). In the cases where there are distinct fertility gradients or other differences between blocks, then these can be estimated and subtracted from the error variance. It is possible, therefore, to obtain a more accurate estimate of the error variance. Blocking does not necessarily need to be different areas within a field trial. Different blocks in a randomized complete block design could, for example, be different days of testing (where it is not possible to test all replicates in a single day).

Factorial designs

Single replicate designs are often referred to as single dimension designs, and randomized designs are called two-dimension designs. In many cases it is important to simultaneously evaluate a number of breeding lines with regard to their response to different treatments. These types of experimental designs are called multidimensional designs or factorial designs. To illustrate factorial designs, consider the example where there are only four breeding lines to be tested and the performance of each is to be evaluated under three different treatments, or factors . Each genotype entry is grown with each of the different treatments. Overall, there are entries. These are arranged at random as illustrated in Figure 10.4a. In the example, only two replicates are illustrated. In practice more than two plots of each test unit would be grown to ensure the necessary level of replication for a reasonable estimate of the error variation. Replicated factorial designs can be completely randomized, or each replicate can be blocked.

Figure 10.4 (a) Multiple factor factorial design, and (b) split-plot design.

Analysis of factorial designs allows estimates of differences between test entries and between treatments compared with an estimated error. These designs also allow evaluation of any interaction that may exist between test entries and treatments (in other words, variation in the response of the lines to the treatment imposed). To illustrate this, consider the performance of two test lines (A and B) each evaluated under low- and high-nitrogen conditions. If the yield performance of the two lines follows the pattern where entry A is higher-yielding than entry B at both nitrogen levels, it is said that there is no interaction. If, however, entry A is highest yielding at high nitrogen levels but entry B is highest-yielding at low nitrogen levels, then there is said to be interaction between genotypes and nitrogen levels. The significance of the interaction is tested in the analysis of variance. It should be noted that in testing for interactions, if there are no changes in genotype ranking then although formally an interaction may be detected, the implications for plant breeding are minimal, unless other treatments (maybe outside the range tested) are envisaged as being likely.

Split-plot designs

In some cases a breeder is interested in estimating the difference between the effect of one factor and the interaction of that factor with a second factor, while having lesser interest in variation within the second factor in its own right. In plant breeding, for example, it is well established that higher nitrogen (within limits) applied to cereal crops will result in higher yield. studies are routinely carried out, not to determine whether there is an average yield increase with increased nitrogen (as this has already been established). The primary goal is to determine the differences between genotypes and the interaction between genotypes and nitrogen levels. In these cases a special type of factorial design called a split-plot design is commonly used.

Split-plot designs divide the total area of a test into main blocks, sub-blocks and sub-sub-blocks. Firstly the main blocks are arranged at random, and then factors within sub-blocks are arranged at random within the main blocks, factors of sub-sub-blocks are arranged at random within sub-blocks, and so on. A simple split-plot design is illustrated in Figure 10.4b. The numbers 1, 2, 3 and 4 are main blocks and the letters a, b, c and d are sub-blocks. Only a single replicate is shown in the figure, but as before, other replicates would be similar in structure and would always be blocked.

The analysis of variance produces two errors for use in F-tests. The error #1 is estimated by the sub-blocks within main effect, while the interaction effects would be tested against the interaction between main .

10.2.3 The increasing role of linear mixed model approaches to analyse breeding data

Most breeding datasets represent not only fixed but also random sources of variation, and conventional analysis of variance approaches in software programs do not formally take such differences into account. Fortunately, the development of software enabling linear mixed model-based analyses such as ASReml, Genstat, SAS and S-Plus, among others, and the increasing computing power at the disposal of breeders, enable the use of linear mixed model platforms to analyze breeding datasets. This box provides only a coarse overview of linear mixed models and their use in breeding, but hopefully encourages interested readers to seek further support from more specialised books or from biometricians.

The basic linear mixed model includes both fixed and random factors as follows:

where:

1. = vector of observations
2. = associated matrix design
3. = vector of fixed effects
4. = associated matrix design
5. = vector of random effects
6. = random residual vector.

A factor is said to be random when it can be thought of as a random sample from a population (for instance, environments, genotypes, sampling units), or a randomization unit such as a plot. Otherwise factors are considered to be fixed, for instance the treatments it has been decided specifically to apply. When the purpose of a breeding trial is selecting superior individuals the genotypes are often considered random, as this enables its estimation as BLUPs (best linear unbiased predictors). BLUPs maximize the correlation between true genotypic values and predicted genotypic values, which is typically the purpose of breeders.

The discussion of whether genotypes are to be declared fixed or random factors is beyond the scope of this book, but suffice to say here that any set of genotypes managed by a given breeding programme could be considered as a random sample of the entire germplasm collection available for a given species that could have arisen as a result of the selection process. This in turn allows the assessment of phenotypic performance as random effects through BLUPs.

Using BLUPs increases the amount of information used to assess phenotypic performance. It also enables a more effective allowance for spatial variation as it can be easily combined with statistical approaches resulting in a higher effectiveness in selecting ‘truly’ superior individuals. They also increase the predicting ability of future performance.

The seminal paper of Arthur Gilmour and colleagues published in 1997 demonstrated how linear mixed models could address frequent sources of variation existing in breeding datasets: large-scale variation across fields, natural variation or local trends existing within individual trials, and also extraneous variation sometimes induced by experimental procedures such as planting and harvesting of experiments or trimming of research plots. These sources of variation would go unnoticed in a regular analysis of variance, not only adding unwanted noise to datasets but also potentially impacting the ability to pinpoint truly superior individuals and therefore the ability to make meaningful genetic progress.

Linear mixed models also present a significant enhancement when it comes to analysing datasets from multi-environment trials where the performance of advanced breeding lines is evaluated across the target population of environments that a breeding programme focuses upon. Most breeders are unaware of the default variance/covariance structure available in many commercially available software packages, named compound symmetry, assuming an equal genetic variance over locations and the same genetic correlation/genetic covariance between any pair of locations. Seasoned plant breeders are aware this is rarely the case, and that often there are large differences among the set of locations they use to test their breeding materials. This can represent a significant loss of efficiency and accuracy to any breeding programme, particularly if some of its locations frequently induce biotic or abiotic stresses. Linear mixed models enable plant breeders (or biometricians) to analyse datasets with increasingly more realistic assumptions, such as a fully unstructured variance/covariance structure where genetic variances are specific for each location, whereas genetic correlations are typical for each pair of locations.

Currently, linear mixed model platforms can be organized and automated so that hundreds of experiments can be analysed at the speed needed by breeding programmes in order to make advancement selections on time. We do not necessarily advocate every plant breeder becomes a user of linear mixed model-based analyses; nevertheless, their deployment represents an effective approach to increase the efficiency of selection, to gain deeper insight in return for the large effort spent on multilocation trials, and to increase the likelihood of commercial success.

The use of linear mixed model approaches is strongly advised when molecular breeding or genomic selection approaches are pursued. Unless the best phenotype is used to establish marker–trait associations, the theoretical strength of molecular breeding or genomic selection might not translate into the improved cultivars that farmers, end-customers and humankind require.

Until recently, deploying linear mixed models was constrained by the lack of computing power, limited availability of user-friendly software, and the very small number of practitioners within the plant breeding community. Now it is feasible for any breeding programme located anywhere in the world to analyse its data through mixed models approaches as these three constraints have generally been overcome.

10.3 Greenhouse management

A large proportion of the tasks that are necessary in the early parts of a plant breeding programme can often be carried out in a greenhouse. An integrated greenhouse system is not essential for a successful varietal development programme. However, many of the operations can be carried out more effectively if a greenhouse system is conveniently available. This is particularly true for private breeding programmes where the speed of cultivar development is of critical importance.

Greenhouses come in many different shapes and sizes and can be constructed from many different materials, including wood, aluminium, glass, plastic and polythene. The actual design of these systems will not, however, be covered here. A greenhouse can simply be considered as a relatively large area where there is some control of environmental conditions such as soil type, irrigation management, nutrient management, lighting and temperature.

The operations that are often carried out in a greenhouse with regard to a plant breeding programme include:

• artificial hybridization;
• seed increases of breeding lines, including progressing to homozygosity;
• evaluation of traits that are difficult to control under field conditions.

10.3.1 Artificial hybridization

It is possible to carry out artificial hybridization under field conditions; however, many breeding schemes use greenhouse facilities for this task because it is easier to achieve the conditions necessary to ensure controlled pollination between chosen parents (Figure 10.5). Also, it can often be possible to achieve controlled pollination out of season in greenhouses, and usually it is easier to prevent unwanted illegitimate cross-pollination.

Figure 10.5 Artificial hybridization in (a) canola breeding and (b) wheat breeding. Note that pollination bags cover racemes and ears that have been pollinated so as to avoid unwanted crosses.

The method used for controlled pollination will be dependent upon the crop species involved, whether the crop is out-crossing or self-pollinating. The major goal in artificial hybridization is to ensure that the seed produced is in fact from the particular, desired parent combination. Therefore steps must be taken to ensure that seed has not resulted from an unwanted self-pollination or from an accidental cross-pollination that is not intended by the breeder.

Artificial pollination therefore demands that naturally inbreeding lines (or lines which are self-compatible) are emasculated to avoid self-pollination. Emasculation in most crop species can be achieved by manually removing the male plant parts (i.e. anthers) before they are mature and pollen is dehisced. In some cases it is possible to use chemical emasculation where specific chemicals applied at the critical growth stage will render the plants male-sterile. Chemical emasculation is, however, not widely used in routine breeding, and mechanical emasculation is used almost exclusively as a means of avoiding selfing in crossing programmes.

Once the chosen plants have been emasculated, within a few days pollen from the male parents can usually be applied to the receptive female stigma. Pollen can be transferred manually, often using a small paintbrush, or by removing dehisced male parent anthers and brushing pollen onto the female stigma. Cross-pollination can also be achieved simply by having emasculated females grown in close proximity to male flowers and allowing pollen to pass naturally from male to female. When this is to be done it is common to place emasculated female flowers and pollen-fertile male flowers together within a pollination bag to ensure that the desired hybridization occurs and to avoid the female being pollinated by stray pollen which may, for example, be blown in the air or carried by insects. If necessary, within these bags, suitable pollinating insects can also be placed to help pollination efficiency.

In a number of crop species there are self-incompatibility systems, which have evolved naturally to encourage cross- (rather than self-) pollination and hence maximize heterozygosity of plants within the species (e.g. as exists in many Brassica species). Similarly, many crop species have male sterility systems (either nuclear or cytoplasmic in inheritance), which can be utilized in cross-pollination systems. In both these cases it is not necessary for the female parents to be emasculated to guarantee cross-pollination.

Regardless of the breeding system, it is common to place pollination bags over flowers either prior to pollination (to avoid unwanted crosses) and/or after pollination (to ensure that no further pollination takes place) (Figure 10.5). It should be remembered that bagging crosses is time-consuming, and that if not done carefully can have an adverse effect on the potential success of the artificial hybridization and the amount of seed produced. Physical damage can be caused during the bagging operation, or the bag may create an environment unsuitable for seed production. It is also important to label carefully at each stage, otherwise the origin of any seed produced may be in doubt.

With many crop species, particularly crops that are clonally propagated and where the end product does not involve the botanical seed (e.g. potato, banana, sugarcane), it is not always easy to have parental lines develop sexually reproductive parts. In a number of instances flower induction can be achieved by manipulation of environmental conditions, by adding or reducing nutrient levels, manipulation of day length, or by artificially controlling the natural source–sink relationship.

For example, in potato, many past breeders have specifically selected breeding lines that rarely produce flowers, with the idea that energy put into sexual reproduction would detract from tuber yield. Flowering can be induced in some genotypes by planting tubers under long day conditions and having plants develop to maturity in shorter days. Enhanced flowering in potato can also be achieved by ‘growing on a brick’, where parent tubers are planted on building bricks and covered with soil. At the stage when tuber initiation occurs the soil is washed from the mother tuber and newly initiated tubers are removed, hence offering greater resources for flower development. A similar effect can be achieved by grafting potato shoots onto tomato seedling rootstocks. Applying high levels of nitrogen at particular growth stages can also sometimes increase the duration of the flowering period.

In other crops (and sometimes also in potato) the opposite is true, and reduced levels of nutrients cause stress to parental plants thus inducing the plant to flower, which would otherwise not occur under optimum conditions.

Finally, irrespective of crop or breeding system, it is always desirable to have multiple and sequential plantings of parents that are to be used in crossing designs. Genetically different parents will flower and dehisce pollen at different times, and multiple plantings will increase the possibility of achieving all the hybrid combinations planned.

10.3.2 Seed and generation increases

If hybridization is carried out between two homozygous parents, then the plants will be heterozygous at all the loci by which those parent lines differ, and all plants will be genetically identical. It is therefore common practice to go from the populations to under glasshouse conditions. This tends to maximize the use of seed because of the high levels of germination and survival that can be achieved. If populations are grown under field conditions it generally requires greater quantities of hybrid seed. This is disadvantageous since the cost of producing seed is usually high, because it involves emasculation followed by hand-pollination, as opposed to simply bagging the to allow selfing in order to produce the .

With many annual (and some biennial) crops it is possible to grow more than a single generation each year by utilizing greenhouses, so reducing generation times and hence increasing the speed to homozygosity. Single seed descent used in spring barley, where plants are grown at high density and with low nutrition, can be used to increase populations to populations within a single year (i.e. three generations in 12 months).

At the advanced stages of a plant breeding scheme, greenhouse growth can also be utilized to increase advanced selections under controlled conditions prior to producing breeders' seed. This can be particularly useful in crops that are grown as true-breeding, inbred lines, but in which a relatively high frequency of natural out-crossing occurs (e.g. Brassica napus).

Tissue culture techniques are becoming a routine part of many plant breeding schemes. Plants rarely can be transferred directly from in vitro growth to field conditions without involving an intermediate greenhouse stage. Here the greenhouse stage could involve an intermediate operation where plants are weaned from in vitro to in vivo sterile soil mix, allowed to develop and are later transplanted to the field. Alternatively the greenhouse can be used to produce seed (or tubers) from plants that have previously been grown in vitro, and hence provided with a more easily protected environment free of pests and diseases.

10.3.3 Evaluation of breeding lines

One advantage of growing plants under greenhouse conditions, rather than field conditions, is related to environmental control. Control of the environment can be critical to guarantee epidemics of pests or disease, or to evaluate stress factors, to allow resistance screening. There have been a number of studies that have resulted in protocols suitable for evaluating plants under glasshouse conditions.

Disease and pest testing involves subjecting segregating breeding populations to a disease or insect and selecting those plants that show resistance. Examples include spraying barley seedlings with a suspension of mildew spores and screening for resistant lines, or spraying potato seedlings with a spore suspension of late blight or early blight and recovering the seedlings that are not killed. These tests are often more effective if there is good environmental control, such as is provided in a greenhouse. This helps to guarantee that the results are repeatable and the particular pathogens are allowed to increase and indeed infect the plants. It also allows control of the disease when it is time to stop further infection

Screening breeding lines for abiotic stresses can also be achieved under greenhouse conditions if the environment can be controlled in a repeatable and relevant manner. Stress screening has been shown to be reliable for such factors as tolerance to nutrient deficiency, drought, salinity and heat, where it is not always possible or easy to control the relevant environmental factors involved under natural conditions in the field.

It should, however, be noted that evaluations designed to be carried out under greenhouse conditions must first be compared with results that would have been achieved under natural field condition. There have been numerous cases where selection has been carried out under controlled conditions and later found to bear little, if any, relationship to what subsequently is experienced under field conditions.

10.3.4 Environmental control

Artificial lighting (fluorescent and/or incandescent) is nearly always necessary to achieve maximum use of greenhouse space. Lighting is, however, expensive both to install and to maintain, particularly if different lighting regimes are required. When, however, lighting is available, it usually allows the greenhouse to be utilized throughout the whole year.

If plants are to be propagated in the greenhouse throughout the year it will also be necessary to have a suitable heating and/or cooling system. A range of different types of systems is available and these cannot be adequately covered here. However, it should be noted that all the types involve a relatively high cost to install and operate. Thus it is usual to expect to have to justify the costs in terms of likely returns of, for instance, increased numbers of generations, effectives of tests, and so on. Good control of temperature is of course important if healthy plants are to be propagated. A particular example in which temperature control is often needed is in biennial crops where plants require vernalization (chill treatment) before they will flower. Plants grown under greenhouse conditions can be vernalized outside the greenhouse (e.g. in a growth chamber or cold room) but this involves moving plants between facilities, which can be time-consuming and expensive if the number of plants involved is large.

Growth within greenhouses requires artificial irrigation. Irrigation can be by hand, which allows for some flexibility but does not usually allow for complex management systems. Automatic irrigation is usually preferable and can be of three forms:

• Above-plant irrigation (or misting) where plants are sprinkle or mist irrigated from above. This can be relatively inexpensive but can cause problems if plants are tall. Above-plant irrigation can also increase the risk of plants becoming infected by fungal diseases where leaf moisture is necessary for infection to take place. It can also be a problem by generating leaf scorch in strong sunlight.
• Below-plant irrigation where plants are irrigated by capillary action from having moist or wet material below the plant pots. Below-plant irrigation avoids the above-ground plant parts becoming wet, although it can be difficult to establish young plants and maintain very large plants with such a system alone, and it is sometimes necessary to hand-water as a supplement to the system.
• Drip irrigation where each plant pot is individually irrigated directly into the soil by a drip line. There are several different forms of drip irrigation and this system offers greatest flexibility over all others. This system is, however, the most expensive to install and is not always available in all plant breeding greenhouses. Since the system requires that individual drip lines are located in each plant pot, there can be some restriction on the number of plant units that can be grown, so this needs to be carefully considered when setting the system up.

All methods of irrigation offer the possibility of applying nutrients along with the water and so they can be provided ‘continuously’, thus enabling more optimized growth of plants over other methods of nutrient/fertilizer application.

10.3.5 Disease control

Unless disease is to be deliberately imposed as a selection pressure, as in the case of a resistance screening scheme, it is desirable to avoid as many diseases and pests as possible in a breeding greenhouse. The best results are invariably achieved when the plants are as healthy and disease-free as possible. Crop failure in a greenhouse as a result of plant pests (mainly insects) or disease can carry a high cost and should, of course, be avoided if at all possible.

Disease and pest control can be achieved by adopting good management practices, including sensible breaks in production along with appropriate sterilization strategies. However, the application of chemical insecticides and fungicides (and sometimes nematicides) is also a frequently needed practice. Application can be by spraying plants or pests or by fumigating a whole area within the greenhouse. The main advantage of chemical controls of disease and pests are that they can be applied in anticipation of a problem appearing. Therefore they offer preventative disease and pest control. The disadvantage is that many of these chemicals are indeed harmful both to humans and other plants as well as insect life, and it is therefore always desirable to minimize their use.

There are now many types of biological controls that can be used to control insect pests within a greenhouse. A well-known example is the release of ladybugs (ladybirds), which are natural predators of aphids, into greenhouses. There are many other predator insects available that can offer effective control of other insect pests. A sample of specific predator types available and the pests they attack include: Amblysieus cucumeris against thrips; Aphidoletes aphidimyza against aphids; and Encarsia formosa against whitefly; while ladybugs and green lacewings are used as general insect predators. The major drawback to biological predatory control relates to the fact that the pest must in fact be present, even if at a low level, before the predators are released (otherwise how will they be able to survive!). It is therefore difficult to avoid some insect damage and almost impossible to achieve complete preventative control.

The risk of soil-borne diseases can be avoided (or at least substantially reduced) by using only sterile soil, or soil mixes, in the greenhouse. However, unless an inert, synthetic soil substitute is used (e.g. ‘Perlite’ or sand/gravel/Perlite) the possibility that disease will occur as a result of infected soil cannot be entirely avoided. Often the sterilization procedure fails to remove all disease or fails to kill microbial spores or weed seeds. In addition if peat moss is used in soil mixes it is almost impossible to ensure the mix is free from insect pests that have a reproductive cycle in the peat moss.

Achieving good disease and pest control in greenhouses also relies upon other factors. For example, good insect-proofing throughout the house will reduce the risk that insects will enter the greenhouse. However, it should be borne in mind that people are very effective spreaders of plant diseases in greenhouses. Personnel from the breeding programme are likely to be in contact with plants outside the greenhouse (i.e. will visit field plots) and so there is a great risk that these staff will transmit disease or carry in insect pests prevalent to the crop with them while visiting the glasshouse. Simple rules, such as carrying out any greenhouse operations first thing each day with other field tasks being done later, can help in reducing disease incidence and spread. Having a mat soaked with a disinfectant agent at each door of the greenhouse is a cheap yet effective barrier to keeping many pathogens out of a greenhouse.

Plant viruses can cause particularly serious problems in plant breeding schemes since many virus diseases are transmitted through the planting material (e.g. seed viruses in cereals and tuber-borne viruses in clonal crops). Many viruses can be eliminated by avoiding the virus vectors, which are often insects (particularly aphids). Workers in the breeding programme can also be responsible for carrying insect vectors into greenhouses on their hands or clothes. Again, the risks of infection can be reduced by applying simple rules (e.g. protective clothing, sterile gloves, etc.).

10.3.6 Economics

Despite the attraction of greenhouses as an integral part of any plant breeding programme, there is no doubt that this facility can be responsible for a high proportion of the overall cost of operating a breeding system. In addition, due to the high cost of building and maintaining greenhouse facilities, the actual space available will be limited. In the practical world (the one in which we unfortunately all live), economic use of greenhouse space is an ever-present major factor.

Plants in greenhouses are grown either in pots (or some other individual unit) or in beds (where many plants are propagated together). The size of pot used (or the plant density in seedling beds) will have a large influence on the number of plants that can be grown in a unit area. It is therefore necessary to choose a pot size/density that will allow good plant health and growth. If small pots are used, then more plants can be propagated at lower cost. If they are, however, too small, then plant health, root development or reproductive efficiency can be affected.

It is necessary to allow access to plants grown in greenhouses. Increased efficiency of greenhouse space can be achieved by using rolling benches, where plants are grown on benches that can be easily moved to allow access, but minimizes the greenhouse space that is allocated to walkways. Rolling benches can, however, cause problems in cases where plants are tall and require staking or tying to prevent them falling over and being damaged. In addition, rolling benches can increase the need for uniform lighting over the whole greenhouse area rather than only over designated growth areas or static benches.

10.3.7 Experimental design in the glasshouse

One final note on the use of greenhouses and plant breeding relates to experimental design. Many believe that the conditions in greenhouses are such that there is uniformity in soil type, lighting, irrigation, and so on. In comparison with conditions that may prevail in the field, there may indeed be less environmental variation in a greenhouse. Despite this, it should be noted that there will be differences nonetheless between, say, plants next to the glass and those in the centre of the house. Therefore all experiments grown in greenhouses should be treated with a clear understanding of the fact that variability in environmental conditions still exists, and therefore good experimental design, replication and randomization will be as important in greenhouse experiments as in other situations.

10.4 Field plot techniques

A large proportion of the work in a plant breeding programme is carried out using field trials. The aim of plant breeding is to develop superior cultivars that are genetically more adapted than the cultivars that are already available. New and old cultivars are grown within agricultural systems on a large scale. For example, wheat grown in the Pacific Northwest of the US is grown in fields that cover many hundreds of acres. Obviously it is not possible to evaluate the many thousands of potential new cultivars in a plant breeding scheme on the large field areas in which they will eventually be grown if successful. The aim, therefore, of field trialling is to predict how each genotype would perform if they were grown on a large acreage basis.

In order to grow accurate and representative field plot trials, it is first necessary to determine the way that the crop is grown in agriculture and to try to use this as a basis for the practices used in the plot trials. Factors that need to be determined include:

• land preparations and agronomic management;
• seeding rate, final plant density and depth of planting;
• nutrient levels and when nutrients are available (pre-plant and/or post-emergence);
• irrigation management;
• timing of operations such as planting and harvest windows;
• chemicals available, for example, which insecticides, fungicides or herbicides are registered for use on the crop, at what rates they are applied, and what the seed treatments are;
• regions where the new cultivars will be targeted.

Do not forget that the major aim of field trials is to mimic what would happen in commercial agriculture. Therefore field trials should usually be planted at the same time that the crop is normally planted. Planting depth, plant density, nutrient management, weed control, disease control, harvest time and method, and post-harvest treatment should all match commercial production as far as this can be achieved within the restraints of small plot management.

10.4.1 Choice of land

In order to choose a good area of land for field plots, it is necessary to identify the factors that magnify soil differences and to reduce, if possible, soil heterogeneity.

Fertility gradients are generally more common in sloping land. Soil nutrients are soluble in water and tend to settle in the lower land areas. Therefore these lower soils tend to be more fertile than the higher areas. An ideal experimental site will be on flat land but this is not always possible – how many farmers' fields have you seen that are as flat as a football pitch?

If the land has previously been used for plot experiments, then this can lead to increased soil heterogeneity. Therefore areas of land that have previously been planted to different crops, different and varied fertility regimes, or subjected to varying cultural practices, should be avoided if possible. In cases where this has occurred, then the area should be planted with a uniform crop, with uniform management and fertilization for at least two years before it is reused for plot experiments. A second source of soil heterogeneity is related to unplanted alleys or roadways from previous experiments. If possible, unplanted alleys from previous research should be marked and avoided.

Grading (ground levelling) usually removes soil from elevated areas and redistributes it to the lower areas. This operation, which is designed to reduce slopes, results in uneven depths of topsoil and often exposes unfertile subsoil. These differences can prevail for many years and should be avoided unless soil heterogeneity trials determine that the grading effect is minimal.

Large trees and other structures can cause shade, which will affect plant performance, and also their roots spread further than their canopies and so will influence plant growth. Areas near buildings may be affected by soil movement and heterogeneity caused by the building operation. Plots adjacent to trees or wooded areas can also carry a greater risk of damage by birds or mammals.

The evaluation of soil heterogeneity requires growing uniformity trials. These involve growing a single cultivar (or a number of cultivars) in plots with very high levels of replication. Uniformity trials highlight soil fertility gradients and identify particularly productive or non-productive areas in fields. Uniformity trials can be used to produce contour maps of productivity. Statistical procedures such as serial correlation studies or least mean squares between rows, columns and diagonals can be applied to determine the significance of soil heterogeneity. It is also possible now to use various spectral imaging techniques to assess the crop's growth and performance.

Although uniformity trials have their place in field experimentation, they usually have little to offer a plant breeder. Uniformity trials indicate the response of specific genotypes to a given area in a given season. When these trials are repeated with different genotypes or in different years, then different results are often obtained (not surprisingly). In plant breeding evaluation trials, the number and diversity of genotypes under test are usually far greater than what can be considered in uniformity trials. Also it should be noted that often there is little choice of what land can or cannot be used for plot trials.

A plant breeding programme usually, at least in the advanced stages of selection, uses a number of different locations, sometimes spread over a large geographical region or ‘target population of environments’. One main location may be identified where the majority of material is evaluated in the early and intermediate selection stages or where seed is increased. A number of different locations will be used (dispersed throughout the region where the new cultivars will be targeted) where advanced lines are tested for adaptability. Where many locations are used it is common to use farmers' fields for test plot evaluation. Some of the distinct differences between a farmer's field and the conditions prevailing at, say, an experimental research station, would include:

• Lack of experimental equipment or lack of small plot machinery. This can often be easily overcome by taking planting, spraying and harvest machinery from the research farm.
• Lack of experimental facilities such as precise irrigation control and pest or disease control, weather stations, and so on.
• Lack of post-harvest storage or assessment facilities. Therefore harvested produce needs to be transported to a central testing laboratory for post-harvest quality assessment.
• Large variation between farms and fields within farms. This is often not a major problem as the majority of trials on these farms are to select for such adaptability over a range of environments.
• The farm sites are usually further away from the base research laboratory and sometimes long trips are necessary to visit the plots. Therefore visits are usually limited and it can be difficult to identify potential problems as they arise and hence avoid their worst effects.

Despite all the potential difficulties with off-station or farm trials, it is possible to achieve very good results. Best results are usually obtained when the ‘better’ farmers are chosen for the tests and when these farmers are specifically interested in the results from the trials. Finally, when trials are to be carried out on farmers' land it is always advisable to keep the experiments simple and to have relatively large plot units, to make them more robust. It is always good advice to spend some time and provide farmers with at least a brief explanation about the trials underway, rather than just appear to plant, collect notes and harvest. This not only shows respect to the farmer, but also increases their interest and commitment to the trials, reducing the likelihood of trial losses or less than ideal agronomic management.

10.4.2 Plot size and replication

Historically, it has always been assumed that larger plots are more efficient and more representative than small plots in yield and other assessment trials. Nevertheless, this is not necessarily always the case. Research carried out at CIMMYT (International Maize and Wheat Improvement Center, or Centro Internacional de Mejoramiento de Maíz y Trigo) compared limited versus normal nitrogen soil supply using different plot sizes, and was able to conclude that the residual variance of progenies grown under nitrogen-limited conditions in small, non-bordered plots was usually less than when grown in larger, self-bordered plots, suggesting that soil heterogeneity was likely to be the main source of residual variance. Similarly there is no doubt that greater replication levels are always more desirable than fewer replicates. However, when the breeding field trials are being carried out under drought, low nitrogen or other abiotic stress conditions, which amplify the natural variation existing in agricultural soils, a larger number of replications might increase the likelihood of encountering unexpected soil variation patterns. In such circumstances, rather than using a larger number of replicates, it might be more effective to use appropriate statistical designs able to accommodate such variation patterns as well as a linear mixed models-based statistical analysis, as previously discussed. In other words, standard plot size and number of replications do not fit all requirements, and need to be adapted to the needs and resources available.

The difficulty of organizing efficient field trials is often related to some compromise in plot size and replication which will allow large numbers of test lines to be evaluated at low cost and on as small an area of land as may be available.

Land availability may not be the limiting factor in determining plot size or replication level. It would be pointless to organize more field plots than could be effectively managed by the staff available. Similarly, data need to be collected from effective field trials, and if too many unit plots are grown then it may not be possible to effectively evaluate all the plants or the produce from the trials. Finally, some crop species produce products that are bulky or perishable. It may be necessary to store the produce (or at least a sample of produce) from each plot, so the storage space available would then be a major determining factor.

In the early selection stages the amount of planting material available is often limited, and this puts practical constraints on the field trialling that is possible. For example, if only 2 g of seed are available for evaluations, and commercial seeding rates are 4 kg per acre, then only small plots with limited replication will be possible.

Increasing replication will be more efficient than increasing plot size in the majority of cases. Therefore if 200 plants were to be grown for evaluation purposes, then the most statistically efficient design would involve 200 replicates of randomized single plants. From a practical standpoint this may not, however, be the most effective or practical method or provide the most representative outcome. For example, there may not be the necessary machinery available that would allow for mechanized planting of completely randomized single plants. Therefore the dimensions of machinery available can be a determining factor when setting plot dimensions. If the only plot-seeder available plants six rows, then all plots are likely to be a factor of six rows wide. Similarly if a small combine harvester is available that has a cut of 1.5 metres, then plots are likely to match this harvesting capability. In addition, single plant evaluation may take greater land areas than would be available. Finally, single plants, if completely randomized, need to be spaced distinctly apart to differentiate one from another. The phenotypic performance of some crop species is markedly different when grown at wide spacing (wider than would be normal for commercial production) than if grown at narrow spacing.

Different plots in field trials invariably contain different genotypes. The performance of these genotypes can, in some cases, be affected by competition from the adjacent plots. For example, if a short genotype is grown next to a tall vigorous genotype, then the performance of the short type may be reduced compared with a single stand of the short stature plants. To a large extent these effects can be reduced by good experimental design and replication where the probability that adverse or advantageous competition occurring in all replicates is reduced with increasing replication. Sometimes nesting based on pedigree might alleviate such competition effects, since half or full sibs are likely to have a similar plant height, architecture or rooting patterns than sibs from unrelated pedigrees.

Some researchers suggest growing larger plots and harvesting or evaluating only the centre rows (i.e. that portion that is completely surrounded by plants of like type). It should be noted, however, that this would require greater amounts of planting material and larger land areas. It should also be noted that genotypes can suffer as much (or greater) competition by being grown by itself, and ripple effects can occur. To examine ripple effects, consider a five-row plot (rows A, B, C, D and E) where row A is grown adjacent to a different, tall and very competitive genotype. In this case then the A row may contain small stunted plants due to the competition from the tall genotype, and hence will result in lower yield. Row B, however, is likely to be affected by competition because although grown next to a like genotype, the like genotype (A row) is stunted and low-yielding. Therefore row B will be taller and more productive due to the lack of competition from row A. In a similar manner, row C will have to compete with the larger, more competitive B row plants and have reduced yield. The competition effects will be reduced, however, with increased distance from the tall different genotype and hence the term ‘ripple effect’.

Breeders need to be aware that by harvesting only a portion of the total plot the error variance will be increased, as the error variance of the mean (average of all plants in the plot) is the error variance of a single plant divided by the number of plants.

It should be remembered that the value of field plot trials is to make comparisons and not to estimate definitive yield performance. Therefore field trials are used to compare the relative performance of different test lines in comparison to control entries. In this case increased or decreased yield as a result of competition will only become a factor if there is interaction between edge effects and genotypes.

10.4.3 Guard rows and discard rows

It is common practice to surround trials (and sometimes even individual plots) with guard or discard rows. These are areas planted to a specific cultivar or genotype, which is not part of the evaluation test. Guard rows are used for several reasons, including:

• If any mechanical damage occurs (e.g. a tractor spray unit accidentally runs over a plot), it is likely to happen to the edge plots. If these are to be discarded, then this damage is less likely to affect the performance of any of the test or control entries.
• Phenotypic performance can be greatly increased by avoiding differential edge effects. Therefore plots that are grown on the edge of a trial will not have any competition on one side, while all other test entries will be affected by competition from adjacent plots.
• In multifactor field trials, guard rows can be used to separate different treatment factors that may be difficult to apply to specific areas without having some effect on the immediately adjacent plot.
• Sometimes, guard rows avoid or reduce the theft of edible organs such as corn ears or fruits from experimental entries, as trespassers are tempted to steal edible plant organs from guard rows rather than from experimental breeding lines. Although this situation is not always considered in developed countries, in some developing countries it represents a significant issue that must be dealt with properly, as otherwise years of hard work and research investments might disappear overnight.

Guard rows are usually the same species as that under evaluation, but this is not always a necessity.

10.4.4 Machinery

Over past decades there has been an increase in the availability of small-scale machinery suitable for field plot trials. Most of the machines are designed as miniature versions of what is used in larger-scale agriculture. Tasks that can now be mechanically orientated include:

• Planting
• Weed, disease and pest control
• Harvesting

It is nearly always desirable to plant field trials mechanically as this is likely to result in more uniform plots than can be achieved by hand planting. This is almost always true for relatively small, seeded crops (e.g. barley, wheat and rapeseed). When the planting material is larger (e.g. potato tubers), hand planting can produce as good, or better, results compared with mechanical planting. The need for automatic planting will be dependent, therefore, on the size of seed to be planted, the density of seed sown, and the time that can be saved by automatic planting.

The most common small-seed plot planters are cone planters (Figure 10.6). This type of seeder can be used very successfully to plant either very small plots or much larger plots. A measured or counted quantity of seed is poured over a cone such that the seed is evenly distributed around the base of the cone that operates the seeder. During planting the cone revolves and seed passes through a hole to be subsequently dropped via disk or tube coulters into the soil at the required depth. It is often possible to plant several rows from the same unit seed lot. In this case, after the seeds drop from the cone they are evenly distributed to a number of tubes, which will each plant a single row. Cone planters are usually designed so that a range of plot lengths are possible. This is achieved by gearing the rate that the cone revolves. After one complete revolution then all the seed from one lot will have passed down the open hole. Planting can be done with continuous movement, with each plot being dropped onto the revolving cone at a designated trip point.

Figure 10.6 Planting yield assessment trials using a single cone planter.

Cone planters are available where the seeds for each plot/row are loaded into a cassette or magazine. This is then mounted above a seeder unit with several revolving cones. With this system it is possible to plant several sets of rows simultaneously with each set being a different genotype. Cone seeders are particularly useful as they can be used with small seed lots, and all seed loaded is planted to completion. Therefore there is no need to maintain a seed reservoir, which would need to be emptied between different plots/genotypes. Cone planters are also self-cleaning.

With small seed, cone planters can result in an even distribution of planted seed, but it is sometimes desirable to have a more precise placement of seed. If this is necessary then precision planters can be used. With these machines it is possible to obtain spaced plants at relatively even density. Precision planters are in general of three types:

• Belt planters, where a reservoir of seed is maintained over a revolving belt. The belt has holes cut which are of precise size and shape so that only a single seed will fit through the hole. The density of planting is achieved by the number of holes in the belt and the rate at which the belt revolves (e.g. Stanhay seeders).
• Vacuum planters, where suction is applied to a revolving plate that has holes drilled to allow only a single seed to be sucked to the plate at each hole position. As the plant revolves, the vacuum is turned off at a specific place in the plate's rotation. At this point the single seed is dropped into the soil.
• Cup planters, where a series of rotating cups are dipped into a reservoir of seed. The size of each cup is such that only a single seed is scooped up as the cups revolve through the seed. At a specific point on the rotation the cups are tipped and the single seed is dropped into the soil.

The major limitation of precision planters, when used by plant breeders, is that they usually require a volume of seed in the reservoir in order to operate effectively. Therefore they have only limited use when small amounts of seed are available.

In some crops, transplanting is common, even on a commercial scale (e.g. fresh tomatoes). Small-scale transplanters are available that allow automatic transplanting of field plots. Seedlings are grown in ‘seedling flats’. At transplanting time the seedlings are removed from the flats by hand and placed into the transplanter. Systems have also been developed where the seedling flat fits onto the transplanting machine and the whole operation is automated. In this latter case it is usually possible only to transplant large plots.

The areas between different plots in field trials are usually left unplanted. There is very little competition in these areas and weeds can be a major problem. Weed control in field plot trials can be carried out mechanically or chemically. Mechanical weed control can be by hand hoeing (a task often enjoyed by many summer student helpers!). Automatic mechanical devices such as rota-tillers and harrow cultivators can achieve inter-row and inter-plot weeding. Often it requires a combination of chemical herbicide application, rota-tilling, harrowing and hand-hoeing to ensure that plots remain weed-free.

Evaluation of disease and pest resistance is an important factor in field testing. Test lines and controls will be grown in regions or areas where specific diseases or pests are common. In these trials, disease is often encouraged by including particularly susceptible genotypes as spreaders and by artificial inoculation of these spreader lines.

In other field studies it is not desirable to have disease or pest epidemics, and so these need to be controlled. Control is usually by chemical application, although some biological control of insects may be available. It should be remembered that many diseases are spread (and most are not helped) by having poor weed control.

A variety of harvesting machinery is available including small-plot combine-harvesters (e.g. Hege or Wintersteiger) that will cut, thrash and partially clean seed samples (Figure 10.7), and harvesters that will dig root crops such as potatoes. Often very small plots (or individual plants) need to be harvested separately. In some cases this can only be achieved by hand-harvest (e.g. pulling single plants or hand-digging individual produce). In the case of grain crops the small plots can be hand-harvested but the seed is removed from the selected plants by small-scale mechanical thrashing.

Figure 10.7 Small-plot combine-harvesting canola trials.

10.5 Use of computers in plant breeding

Many routines in a plant breeding programme follow a cyclic annual operation. Therefore the same tasks (or similar operations) are carried out on a seasonal basis. In general terms a simple breeding scheme may involve:

• Deciding which breeding lines are to be tested in which environments. Which characters are to be evaluated from each trial? What control or check genotypes will be included in each trial for comparisons?
• Designing experiments or evaluation trials. What types of experimental design will be used for each trial (unreplicated designs, randomized complete block designs, lattice designs and split-plot designs)? Having decided on an appropriate design, then field plot plans need to be produced, planting material organized and arranged in order for planting.
• A number of clerical tasks will be required, such as producing plot labels or harvest labels (Figure 10.8), genotype lists and perhaps score sheets.
• After field (or other) trials have been planted, then data will be collected throughout the growing season, at harvest and post-harvest.
• As data are collected, each variate assessed needs to be analysed. More detailed analysis of over-site trials and to examine relationships between variates will be carried out.
• Statistical analysis is only one step in data interpretation. Further data examination techniques of scatter diagrams, histograms or bar charts can be used to obtain a better understanding of data.
• Selection will be applied based on information collected. Breeding lines will be divided into various categories (e.g. definitely select and advance to next stage, not quite sure so repeat in smaller trials, discard from the breeding scheme).
• Once the ‘best’ genotypes have been identified, the operation will begin at stage one again the following year.

Figure 10.8 Assortment of planting and harvest labels used to organize breeding material.

This cyclic operation will continue over a number of years, starting in the early stages where perhaps many thousands of lines will be tested at limited sites and with few characters recorded, and moving to an intermediate and advanced stage, until after several selection rounds only one or two potential new cultivars have survived for varietal introduction.

Computers can be of great benefit to plant breeding in carrying out all of the tasks above, and perhaps even in others not mentioned. Often it is necessary to use different software packages for different aspects of the programme, although there are a few packages that have been specifically designed for managing plant breeding programmes or for field experimentation studies.

Computer software packages that are available are all roughly of the same form. There is a central data storage (database) that can be accessed by a number of routines. Each routine will perform specific operations. Various routines may add information to the database while others will take information from the database and carry out a specific task.

The following section will examine a number of the options or routines that are available and explain how they may be used to increase the efficiency of a plant breeding programme.

10.5.1 Data storage and retrieval

All plant breeding schemes will generate vast bodies of data. If these data are to be used effectively for selection of the most desirable genotypes, then reliable storage and retrieval of information is essential. Computers offer the option of storing data in such a manner that datasets can be tabulated for inspection in a number of different ways using database management systems.

In simple terms there are two types of databases used in plant breeding, called breeding line databases and germplasm databases. There are a number of differences between the database structures depending on the two types.

Plant breeding databases will store information on assessment trials. Therefore, a large proportion of the entries in these databases will be discarded after each selection stage. Early generations will have thousands of records (where one record is associated with information from a single genotype) with only a few data scores on each. Conversely, a genotype that survives to the advanced stages will have been assessed over several years, and in many of these years, assessment will have been carried out at a number of different locations. Therefore the amount of data storage space needed for each record will depend upon the stage that has been reached in the breeding scheme.

Germplasm databases, on the other hand, hold information on a wide range of different genotypes. However, unlike a breeding database, new accessions (or records) are added but very rarely are records deleted. Information stored in a germplasm database will have been collected over years and sites but not all accessions will have been assessed in a common environment. It is therefore necessary to rate accessions (e.g. on a 1 to 9 scale or A, B, C, etc.) so that comparisons can be made. It is very rare that actual yield data (e.g. t/ha) are stored on a germplasm database; it is more likely that a particular accession will be rated by a particular ‘score’ for yield.

Irrespective of the type of database, each record (test entry or accession) will be assessed for a number of characters or traits of interest. Variates can be of two forms, numerical (e.g. disease rating of 2, yield of 25.32 t/ha) or character (e.g. alpha-numeric character string like ‘Yellow flowers’). In addition the different database types will hold other information not related to simple assessment, for example, an alpha-numeric string to identify the particular genotype (e.g. PI.23451 or 89.BW.11.2.34) and parentage of the line. Germplasm databases may store information not usually stored on a database, for example, species name, ploidy level, source of origin of seed, age of seed, amount of seed available, and so on.

When a particular genotype entry is introduced into a plant breeding programme it is usually identified by an alpha/numeric code. For example, cultivars will have specific names ‘Jack's Wonder’. Genotypes, which have derived from other germplasm collections, will have an accession number. For example USA plant introduction lines all have PI numbers (e.g. PI.12342).

Different genetic lines derived from a breeding scheme will generally have similar identifying codes. Genotype codes can be assigned in numerical order (e.g. line 1, line 2, etc.). It is more useful to assign an identifying code that provides some information regarding the background of a specific genotype. For example, in a specific rapeseed/canola breeding group all crosses made are assigned a code identifier that includes the year of crossing, a two-letter code of the purpose behind the cross, and a numerical number. A cross identified by 93.WI.123 would indicate the 123rd cross made in 1993, with the purpose of developing a winter industrial (WI) type.

Specific individual genotypic selections from the cross would have different trivial numbers (e.g. 93.WI.123.23, 93.WI.123.69, etc.). If some form of pedigree selection scheme is used, then additional trivial numbers can be added to indicate the number of within-population selections made.

In setting up a suitable database the user must decide on a suitable database structure. This will determine the number of entries to be tested in each trial, the number of locations where evaluations will be carried out at each stage, and the number (and type) of data that are to be stored.

Irrespective of the type of database or form of data storage the primary aim is the same: to make information available for inspection in a clear and concise form.

10.5.2 Field plan design

Field trials and experiments are of major importance in a successful plant breeding programme. The ability to use computers for randomization has been realized for many years, and most programmes use some form of computer generation of field trials. These packages use entered information such as type of design, experiment title, number of entries, number of replicates, and so on, and produces a randomization along with a map representation of how the plots will appear in the field.

Once a computer has generated a field design it is possible to store all the trial details, number of entries, entry codes and actual randomization on a database system. This information can be retrieved later for analysis of data or producing plot labels.

10.5.3 Clerical operations

Despite advances made in database management, the ability to carry out complex selection strategies or analysis using computers, the simplest and most useful task a computer can do for a plant breeding programme is to perform as many of the routine clerical operations as possible. Several years ago, all plant breeding schemes produced all field maps, genotype lists, seed packet labels and score books by hand. Even in cases where highly methodical and dedicated staff are used, there are inevitable transcription errors, and more importantly it is time-consuming.

Computer systems can easily be used for:

• printing field plot plans (mentioned above);
• printing a variety of labels that can be used to organize experiments, store seed or planting material, or harvest experiments (Figure 10.8). Organization of large randomized experiments can be achieved with ease if computer labels are printed with the genotype identifier along with a field position (e.g. either a plot number or a two-dimensional array number);
• printing score books, which show simply a number of boxes in which data can be handwritten for a number of traits;
• genotype lists of lines, which have attributes in common (e.g. all bread-quality wheat, all industrial rapeseed, etc.);
• data summary tabulations of information stored in the database;
• keeping track of exactly where each test entry is (e.g. what stage of the breeding scheme, at which sites each entry is being evaluated, exactly where in the field each test line can be found).

10.5.4 Data collection

A breeding scheme is only as effective as the data collected on how the different genetic lines perform. The breeder must collect data on performance of different traits from appropriate assessment trials. Data collection and data management are areas that have received very little attention in a plant breeding context. However, the information-gathering stage is of great importance.

Data management is of three types:

• Collection of information into a form suitable for computer entry.
• Validation of data to ensure that errors can be corrected before data analysis.
• Sorting data into a form suitable for entry either into analytical software or for storage on a database.

Data collected from experiments can be hand-recorded onto score sheets and then entered (i.e. key to disk) at a later stage. This form of data collection and entry may appear inefficient compared with more direct systems (below). However, there are one or two advantages of hand-recording and later entry. There is always a hard copy of the information collected that can be referred to at a later date. Hand-recording of visually assessed data can often be achieved quickly compared with other means (although the data still need to be typed later). Therefore a combination of experienced assessor and experienced typist/data recorder may be as quick and efficient as directly logging data.

Information can be logged directly into a computer system. Data logging can be of two forms:

• Information is entered into a hand-held data logger. When combined with plot identification through bar codes, the use of data loggers can greatly speed up data collection in the field.
• Information is transferred direct from an analogue/digital machine (e.g. electronic balance or moisture meter).

In each case, the data are usually later transferred to the main computer system. Data validation (e.g. that the numbers are reasonable or within a certain range) can often be achieved during, or as a part of, the transfer operation. Alternatively, data may be validated or checked as entered into a hand-held unit.

Automatic transfer of data from analytical machinery is always an advantage as it reduces time and effort to achieve results. More important, however, is that this form of data collection usually avoids any additional transcription errors (e.g. writing down or hand-typing the wrong number).

Hand-held data loggers are rapidly becoming smaller, more sophisticated and cheaper. However, it can often take longer to enter data (particularly alpha-numeric character information) into a hand-held unit than to simply write the information. In addition, some expertise is required in the use of hand-held units, particularly accurate typing skills. Finally, if data are collected in a handheld data logger it is always best to have some form of hard-copy printout of the data as recorded. This would indeed apply to any data recording.

10.5.5 Data analysis

One primary consideration for the analysis of assessment trials is the ease and speed of processing. Often the most important traits (e.g. yield and quality) are not recorded until late in the season. A rapid throughput of analyses can therefore be critical to allow selection decisions to be made and new trails organized before planting time. This can best be achieved if all the genotype identification codes, experimental design details and randomization information have previously been stored in a database. In order to carry out an analysis of variance for a single variate assessed at one location, and produce an easily understandable but comprehensive output, it should only be necessary to enter parameters to identify which trial is to be analysed and the variate name.

As data are collected throughout the growing season, analysis of individual traits can be carried out soon after data collection. Inspection of de-randomized data and genotype averages can often serve as a good check that there are no major errors in the data. It is important that each variate be analysed to determine the variability within the genotypes for particular characters. Most database systems will automatically store means and statistics as analyses are performed.

The mode of data entry will, to a large extent, be determined by the method used to collect data (i.e. automatic logging, data logging, or pencil and paper). Irrespective of how the data are collected, eventually the data to be analysed will be available for entry into an analysis and storage scheme. The order in which numbers are entered can differ from one of no pattern (not a good idea), field plot order (either going across the trial or up the trial), or in standard order (e.g. genotype 1 replicate 1; genotype 1 replicate 2; genotype 1 replicate 3; genotype 2 replicate 1; etc.). It is important that data be entered in the order expected by the software package.

Other features that will facilitate a rapid and efficient turnover of analysing individual traits and storing information will include:

• the ability to estimate missing values;
• the analysis of a subset of the total number of replicates (e.g. in a four-replicate trial, data for some traits may only collected from replicates three and four);
• to be able to transform (e.g. ARCSIN transformation for percentage data) or convert (e.g. convert dates to days after sowing, or to convert plot yield into t/ha) data before analysis;
• to derive variates from single or multiple datasets before analysis (e.g. if grain yield is recorded along with straw weight, total above-ground biomass can be derived by adding the two recorded characters).

If multiple environments are used (say at the advanced trial stage), then over-sites analysis (simple analysis of variance or joint regression analysis) can be performed using stored means from individual site analyses. If an assessment trial is grown at two (or more) locations, and yield per plot is recorded from multiple replicates at each site, the following procedure can be used to obtain an analysis of variance of yield over sites:

• Analyse data from each location separately and store the genotype means on a database, along with the error variance from the analysis.
• When each location has been analysed separately, then an analysis of variance with source terms: genotypes, locations, and an error term can be produced quickly and easily. The error term is obtained by simply pooling the error terms from each of the individual analyses.
• When conducting across-locations or across-years analyses, it is sometimes overlooked that the default structure for the genetic variances and covariances in most statistical packages is that known as ‘compound symmetry’. In many cases these assumptions do not hold true, and therefore analysis results would not be correct. Based on the particular nature of the multi-environment testing effort of breeding programmes, breeders have to assess whether they should use variance/covariance structures more appropriate to their specific conditions. This is sometimes not considered because of fears that the statistical analysis will become more complex and/or that it will somehow slow down the speed of the breeding programme. Unfortunately, the consequence of this is that selection decisions departing from the correct ones are being made, which can decrease the rate of genetic gains being made and, more importantly, the ability of new cultivars to address the increasing needs of humankind.

To interpret data from assessment trials and provide indications of possible selection strategies, then joint regression analysis, over-site analysis, simple and multiple regressions and correlation analysis can all offer an insight into the variability of characters and also the relationship between traits. In addition, visual inspection of histograms and scatter diagrams can help in decision-making. Multivariate transformations (canonical analysis, principal components analysis, etc.) have been suggested as possible aids to plant breeders by reducing the dimensions of selection problems. If these transformations are readily and easily applied to breeding datasets, perhaps plant breeders will more readily use them.

Alongside complex analysis, it should be possible to carry out simple calculations, to include addition, subtraction, multiplication and division. Other calculations that may be helpful would include expressing data as a percentage of either the trial mean or the average performance of one or more control lines.

10.5.6 Selection

If many hundreds of lines are to be considered for selection, then computer simulation (by selecting a subset and comparing that subset to those lines rejected) can be a big help in either setting culling levels for different characters, or in setting weights in an index scheme.

The speed with which different selection strategies can be compared using computers offers the potential of investigating a number of different selection options within a narrow time schedule between final assessment of genotypes and preparations for planting the following stage trials.

10.5.7 Data transfer

The amount of data collected on individual genotypes in a plant breeding programme is directly proportional to the stage of selection. By the most advanced stages, data from surviving lines will have been collected over several years and locations. If plant breeding database systems are to be of a useable size and if all information available is to be stored together in a common database, then each season either:

• records for discarded breeding lines must be deleted from the database and new data storage allocated to those selected lines; or
• selected lines must be transferred, along with any data collected, to a new database where additional storage is available.

Either of the above options can be used and both are equally efficient. If the first option is chosen, the old database can first be copied before the unwanted records are deleted. This allows access to data from discarded lines, which can often be used in the future, for example to gain an indication of particular defects of specific parents. It should be stressed that any breeding programme must keep updated, secure back-ups of its datasets, pedigrees and related information at all times. If for whatever reason the datasets of a breeding programme are lost and a properly updated back-up is not available, many years of work and significant funding might be lost as well, and in extreme cases the affected breeding programme might be forced to either start from zero or shut down. Fortunately, the rapidly decreasing cost of computers and servers makes easier and cheaper the secure storage of the massive amount of data a breeding programme accumulates over the years.

10.5.8 Statistical consultation

It is essential that agricultural experiments have clear objectives and that they are well organized and are designed based on sound statistical reasoning. Consultation with qualified statisticians should be done before, and also throughout, the experimentation period. Plant breeding assessment trials are no exceptions. There has been some concern that statisticians will not be consulted if breeders are capable of easily generating a number of different experimental designs and performing complex analyses of data from these experiments. Care must be taken to ensure that the appropriate design is chosen to answer the questions required. Most plant breeders' trials are, however, of a standard form where a number of test genotypes are compared in performance with a number of standard or control cultivars. Although the majority of plant breeders are more than capable of using the appropriate experimental design and making the correct interpretation of a standard analysis, it should be noted that many analysis types (e.g. multivariate analysis) are now readily available to non-qualified workers but that interpretation of these results often requires an experienced person. The point is, therefore, that statisticians should not be ignored and where possible they should be consulted and encouraged to contribute ideas in data interpretation. The support provided by statisticians is often much more valuable when they are engaged before experiments are conducted and designed, rather than once they have been harvested.

10.5.9 Ease of use

One feature of computers and computer software that has not been discussed is the ease of operating the system. Many software packages are user-friendly, which means that they can be used by relatively inexperienced staff. This does not, however, imply that these database systems can be used without computer training. There will be at least minimal training required if a database scheme is to be integrated into a breeding programme.

Most user-friendly computer packages give clear and precise instructions in the form of prompted messages, to which the user replies with one or more operations or data entries. In many cases these prompted instructions can partially eliminate the need for ‘user manuals’. It is, however, general experience that a combination of prompted commands along with a fully documented and concise user's manual will normally be required.

10.6 Release of new cultivars

The ultimate goal of any plant breeding programme is to develop superior genotypes and to release these into agriculture as new cultivars that better serve the increasing needs of humankind. The final stage of a breeding scheme therefore involves the process of release, perhaps protection, and distribution of planting material to the seed industry and/or the farming community.

The first part of this process is when the breeder decides that a particular genotype has merit as a potential new cultivar. This decision will have been made by observing the performance of the potential new cultivar as it passes through all the stages of the breeding programme. This would entail a number of years and, in the more advanced stages, a range of different locations. It cannot be stressed too strongly that if there is any doubt regarding the worth of a potential cultivar, then these doubts must be addressed before deciding to release it for growing. The general agricultural community does not generally take kindly to being sold seed of a cultivar which proves to be of little, or no, use. In the seed market a good reputation is difficult to obtain but easy to lose!

In most cases the decision to ‘release’ or ‘launch’ a cultivar is not the exclusive decision of the breeder. Where a breeder is working for a commercial company, then the final decision to take the first steps towards commercialization is unlikely to be made only by the breeders. Others within the company, the board of directors, financial or marketing staff, will all contribute to the decision concerning the potential commercial impact that the cultivar may have, and more importantly, the potential profits that can be expected to the company if release is successful. If the new cultivar has been developed in a University department or other public research organization, then the final decision on release may involve heads of department and deans of the college or experimental station. Irrespective of whether public or private investment has financed the development of the line, then there is logic in the breeder having a major input into the final decision.

In this decision-making process, the requirements (often statutory) that are made of a new cultivar must be borne very clearly in mind. If the cultivar fails to meet the stipulated criteria, then it will not be possible to commercialize it and all the effort will have been wasted.

10.6.1 Information needed prior to cultivar release

Before a breeding line can be considered for release it must be shown to be distinct from other cultivars that already exist. Distinctness can be for morphological characters (e.g. flower colour) or a quality trait (e.g. low linoleic acid content in the seed oil). It is sometimes possible to say that a new cultivar is distinct for a quantitative trait such as high yield, but in this case the new cultivar must always express the high yield character if release is granted, and in practical terms this is not an easy way to proceed. More recently, breeders are using molecular techniques to distinguish new releases from already existing cultivars.

The new cultivar must also be stable and uniform (i.e. stable over several rounds of increase) so the genotype must always appear the same, irrespective of where it is grown. Therefore if a new cultivar is released which is described as having uniform white flowers, then all individual plants grown must have white flowers.

Careful attention to the final stages of seed increase and meticulous care in producing breeders' seed can be of great benefit in ensuring the uniformity and stability of the new variety.

10.6.2 Value in release

Prior to releasing a cultivar, breeders must demonstrate (from data collected from evaluation trials) that there is indeed merit in releasing the new cultivar. This will involve presenting data from several years testing and from a number of locations, but the exact requirements and procedures will vary from country to country.

In many countries, government authorities carry out official independent testing of all new cultivars before release is allowed. These trials are carried out over two or three years and at a number of locations throughout the target region. The aim of these trials (National List Trials) is to ensure that new cultivars are suitably adapted for the region. If breeding lines show sufficient value for cultivation and use (VCU) then they will be added to the National Variety List of the particular country. In the case of EU countries, when any new cultivar is placed on the National Variety List of any EU country, then it is automatically entered onto the EU Common Catalogue and can hence be increased and sold in other EU states.

In other countries, such as the US, there are no official National List Trials in which each new cultivar is evaluated. However, each US state has an appointed body of people who review performance data for all new cultivars and determine whether they merit release within the particular state. Breeders can submit data for release in more than a single state simultaneously.

10.6.3 Cultivar names

Any cultivar to be sold commercially must be given a unique name (or identifying code) prior to commercial release. Within any given crop species there should only be one with that particular name. So a wheat cultivar and a potato cultivar can both have the same name, say, ‘Sunrise’, but two barley cultivars cannot have a common name, say, ‘Maltster’. Also, unless there is some unfortunate problem such as a cultivar mistakenly being allowed a duplicate name, it is difficult to change the name after release.

Hybrid cultivars are often given a code number rather than a recognizable name. In such cases the number code has a prefix that identifies the company responsible for its development.

In choosing names it is useful to select ones that are easy to remember, and convey, if possible, the right image (e.g. ‘Star’, ‘Golden Supreme’ or ‘Bountiful’). Equally it is wise to avoid names that are obviously inappropriate, such as ‘Usually Dies’ or ‘No Profit’.

Another point to bear in mind is, if a cultivar is to be marketed in a foreign country, or in an area where a second language is common, then it is important to check that the cultivar name does not have an unfortunate meaning in the other language, that it does have a desired image in that language, and that it be easily pronounced. For example, there is no ‘w’ in the Spanish alphabet, so it would be unwise to call the cultivar ‘Wally's Wonder’ if it is to be commercialized in a Spanish-speaking country, such as Spain or Mexico. The same would apply to the letter ‘ñ’ from Spanish, which is non-existent in the English language.

Think questions

1. You are employed as a barley breeder for a commercial company. At present the company does not have any greenhouse facilities. You are trying to convince your peers that the breeding programme would benefit from having a greenhouse. List five uses you would have for a greenhouse in your breeding programme, if one were available.

   Great job (we hope!), you have convinced the Board of Directors to proceed with purchasing a new greenhouse. List five features you would like to request to have in the new greenhouse facility.

2. Briefly describe one problem of using a greenhouse in a plant breeding programme.
3. What is the primary goal of conducting field evaluation trials in a plant breeding programme.
4. List things to avoid when organizing plant breeding field plots.
5. Why is experimental design important in plant breeding evaluation trials?
6. List four problems that may be encountered in organizing and carrying out ‘off-station’ trials.
7. You have been employed as Senior Breeder at the Hass Bean Breeding Co. for several years and have had to carry out your breeding operation with very little funding. Due to a recent takeover of the company by the Human Bean Seed Association, you have been given a vast sum of additional money. You have decided to spend this money upgrading the computer facilities used by your breeding group. To achieve this goal you have appointed a computer programmer to develop a computer-based plant breeding management system. List the five main features the computer package needs to perform that would help to increase the efficiency of breeding future new cultivars.