Potato Research (2017) 60:171–193

DOI 10.1007/s11540-017-9346-z

Review and Analysis of Limitations in Ways

to Improve Conventional Potato Breeding

John E. Bradshaw 1

Received: 2 February 2017 / Accepted: 27 July 2017 /

Published online: 11 September 2017
# European Association for Potato Research 2017

Abstract A number of improvements to conventional potato breeding are possible

but they all have their limitations which need to be appreciated in designing new
breeding programmes. Selection for quantitative traits between crosses before selec-
tion within the most promising ones can solve the problem of the inability of intense
early-generation selection to affect most economically important traits. It also allows
full-sib family selection to be practised but the rate of progress is limited by a low
intensity of selection so that continued recurrent selection is required to accumulate
small improvements into worthwhile ones. The rate of progress from combined
between and within cross selection is limited by the number of vegetative genera-
tions required to complete all necessary phenotypic assessments, as well as by the
intensities of selection. Both problems could be solved by genomic selection
provided adequate selection accuracy can be achieved. However, lack of accurate
large-scale phenotyping may limit the use of genomic selection in the immediate
future. Marker-assisted selection can be used to stack major genes and QTL alleles
of large effect in new cultivars, but the required population sizes will limit the
number of unlinked genes going much beyond eight. Site-directed transformation
should speed the stacking of transgenes in a new cultivar and hence increase the
number of major improvements that can be made by genetic transformation.

Keywords Between cross selection . Clonal selection . Family selection . Genetic

transformation . Genomic selection . Marker-assisted selection

* John E. Bradshaw
johnbradshaw949@btinternet.com

1
James Hutton Institute, Dundee, UK
172 Potato Research (2017) 60:171–193

Introduction

Potato breeding during the twentieth century involved planned artificial hybridizations
followed by multi-stage, multi-trait selection over as many as eight clonal generations,
and from the 1960s, typically started each year with as many as 100,000 seedlings from
200 to 300 crosses (Mackay 2005; Bradshaw 2009a; Bradshaw and Bonierbale 2010;
Lindhout et al. 2011; Eriksson et al. 2016). The two main weaknesses of such breeding
are the number of clonal generations required to select a new cultivar and the inability
of intense early-generation selection to affect most economically important traits, which
are quantitative in nature. This is particularly true for visual selection between seedlings
in a glasshouse and spaced plants at a seed site (Bradshaw and Mackay 1994;
Bradshaw et al. 1998; Haynes et al. 2012). Nevertheless, continued progress worldwide
in adapting potatoes to new environments, farming practices and uses (markets) was
made by cycles of such hybridization and selection, usually among the developing elite
germplasm. Hirsch et al. (2013) made a genome-wide (3763 SNPs) assessment of
historical and contemporary cultivars in North America. They concluded that 150 years
of potato breeding had resulted in selection of alleles in biosynthetic pathways that are
integral to phenotypic traits such as pigmentation (yellow market class) or carbohydrate
amount and composition (processing market class), but not in substantial changes in
heterozygosity and allele dosage.
Sometimes new traits are required, such as resistances to emerging pest and disease
problems, and sometimes broader genetic bases are sought to deal with perceived
plateaus in progress for traits such as yield. Thus, the twentieth century saw the use
of potato landraces and wild relatives in introgression breeding and in base broadening
(Bradshaw et al. 2006). An insight into this breeding history has been provided by Vos
et al. (2015) who used molecular markers (14,530 SNPs) to track the introduction and
selection of novel genetic variants in cultivars released since 1945. A striking feature of
breeding in the twentieth century was the longevity of use of cultivars that became
widely grown. In the UK, for example, Maris Piper was released in 1966 and was still
the number one cultivar in 2016, a position it had held for 40 years. Likewise, the most
popular cultivar in Sweden is still King Edward (VII), which was multiplied from a
seedling in Northumberland, England, in 1902 (Eriksson et al. 2016). It can therefore
be argued that new cultivars did not contribute as much as might have been expected to
the improvement of potato crops worldwide.
Ways to make potato breeding faster, more efficient and more effective have become
available since the 1990s. Progeny tests have been developed for quantitative traits that
allow selection between crosses before starting clonal selection within the most prom-
ising ones (Bradshaw et al. 1995, 2000b). They also allow full-sib family selection to
be practised on a three-year cycle, but their use has not been widely adopted.
Diagnostic molecular markers (Gebhardt 2013; Sharma et al. 2014) have been
developed for use in marker-assisted selection (MAS) and pyramiding of major genes
(Ortega and Lopez-Vizcon 2012) and QTL alleles of large effect (Moloney et al. 2010;
Gebhardt et al. 2011), as well as for marker-assisted introgression (Barone et al. 2001).
Molecular markers allow potato breeders to select for associated desirable genes in
greenhouse-grown plants, having made sure that they are present in the parents of the
cross. However, the few published practical examples have applied MAS to the second
field generation, as recommended by Slater et al. (2013) as cost effective in selecting a
Potato Research (2017) 60:171–193 173

gene for resistance to potato cyst nematodes or to PVY. Furthermore, pyramiding to

date has been restricted to two or three genes.
Slater et al. (2014) also proposed combining the use of MAS with estimated
breeding values (EBVs) for quantitative (polygenic) traits, calculated from pedigree
information, to reduce the breeding cycle from over 10 years to as few as 4 years.
Further progress can be anticipated with the introduction of genomic selection (GS) for
unspecified QTL alleles of small effect, where one wants to combine as many as
possible in a new cultivar (Slater et al. 2016). This has been made possible by SNP
discovery using next-generation sequencing (Hamilton et al. 2011; Uitdewilligen et al.
2013) and the development of SNP arrays such as the 20K one used by Vos et al.
(2015). Again GS could be done on greenhouse-grown plants to allow a short cycle
time in what would in effect be a recurrent selection programme. Genomic selection
could also lead to more radical changes in potato breeding (Slater et al. 2016).
Major improvements to already successful potato cultivars have been achieved by
various types of Agrobacterium-mediated genetic transformation. Cisgenesis has been
used to transfer R-genes, cloned from cross-compatible wild potato species, into late
blight (Phytophthora infestans) susceptible cultivars (Jo et al. 2014; Haverkort et al.
2016). Gene silencing by RNA interference has proved useful for altering starch
composition, processing traits and other quality traits, and has been done using
marker-free transformants and intragenic approaches (Rommens et al. 2006; Chawla
et al. 2012). Finally, transgenesis has been used to introduce traits not found in cross-
compatible wild relatives. The genes used to date have mainly been ones that code for
proteins that are toxic to pests and pathogens and ones whose expression interferes with
virus multiplication in host cells (Davies 2002; Bradshaw and Bonierbale 2010).
However, the period 1995 to 2013 saw a reluctance to accept the first genetically
modified (GM) potatoes in Europe and North America, and the same may be true of the
new GM potatoes now available for human consumption in the USA (Bradshaw 2016).
Combinations of traits can be achieved by stacking genes through co-transformation
with separate constructs, transformation with a multi-gene cassette (four to six is
current limit) or re-transformation with further genes. For example, a new project based
at the Sainsbury Laboratory in Norwich (http://www.tsl.ac.uk/news/new-potato-at-the-
sainsbury-laboratory/) aims to stack eight genes in cultivar Maris Piper. In the future,
the use of site-directed transformation (rather than random insertion) should allow gene
stacking at a single site to create a supergene that will be inherited as a Mendelian gene
(Mahfouz et al. 2016). Gene editing will also be possible by site-directed DNA
sequence modifications, the potato genome sequence having been published in Nature
on 14 July 2011 (Potato Genome Sequencing Consortium 2011). This, for example, has
allowed Van Harsselaar et al. (2017) to annotate the complete set of 77 starch metabolic
genes in potato plants and their genomic localizations.
The current challenge for potato breeders and biotechnologists is therefore to design
breeding programmes that integrate marker-assisted selection of specific alleles, geno-
mic selection of unspecified alleles and phenotypic selection, having decided when a
gene editing or transgenic approach is more appropriate, given consumer acceptability
of the latter. It therefore seems timely to re-examine options for breeding vegetatively
propagated tetraploid cultivars of potato through a review of relevant theory and
experimental results. Both cultivar production (combining alleles in a single genotype)
and longer-term crop improvement (increasing the frequencies of desirable
174 Potato Research (2017) 60:171–193

combinations of alleles over sexual generations) are considered, with breeding exam-
ples drawn mainly from research done at the Scottish Crop Research Institute (SCRI,
now the James Hutton Institute) as I am most familiar with this work. Breeding
potatoes for TPS (true potato seed) propagation, including diploid F1 hybrid seed
potato breeding (Lindhout et al. 2011), is beyond the scope of this paper.

Breeding Programmes, Parents and Crosses

Potatoes are grown worldwide (Hijmans 2001), broadly speaking as a summer crop in
the tropical highlands of Bolivia, Peru and Mexico; all year round in parts of China,
Brazil and the equatorial highlands of South America (e.g. Ecuador and Colombia) and
East Africa (e.g. Kenya and Uganda); as a winter crop in the lowland subtropics (e.g.
northern India and southern China); as spring and autumn crops in the Mediterranean
(e.g. North Africa) and in summer in the lowland temperate regions of the world (North
America, western and eastern Europe and northern China). The growing season can be
as short as 75 days in the lowland subtropics and as long as 180 days in the high Andes.
As well as being a staple food, the potato is grown as a vegetable for table use (with
market classes and consumer preferences), is processed into French fries and crisps
(chips) and is used for dried products and starch production.
The initial key strategic decision for a breeder is therefore the number of
separate breeding programmes required to provide new cultivars for the chosen
range of target environments and end uses. For example, Eriksson et al. (2016)
concluded that breeding can be justified for the Fennoscandian region, with its
relatively short and intense seasons with long day lengths, specific pathogen
pressures and specific consumer preferences. Where the range of targets is not
obvious, an assessment of genotype × environment (including end use) interac-
tions will be required, as explained in detail by Bradshaw (2016). Brown et al.
(1996), for example, found a greater correlation for total marketable yield between
a Scottish ware site, where clones were selected, and sites in England (0.43–0.70)
than with sites in the Mediterranean (0.00–0.67). Although the very best clones
from the Scottish ware site performed reasonably well in the Mediterranean, the
results supported the idea that selection would be optimized by selecting in
environments more similar to those in which the cultivars are to be grown. Indeed,
one could justify separate programmes for northern Europe and the Mediterranean,
particularly as abiotic stresses such as heat and drought and diseases such as
Alternaria and Verticillium are important in the Mediterranean but not in the UK
(Mackay 2005).
Where a programme is targeting export markets, the breeder needs to be aware
of the fact that for quantitative traits, the improvement of performance in one
environment y (CRy, correlated response), as a result of selection in another
environment x, depends on the square roots of the heritabilities of performance
in each environment and the genetic correlation between the two performances
(Falconer and Mackay 1996):

CRy ¼ ix hx hy rG σPy
Potato Research (2017) 60:171–193 175

(i is intensity of selection, h is square root of heritability, rG is genetic correlation and

σPy is square root of phenotypic variance in y). Hence, a low genetic correlation would
limit progress if this was the way the breeding programme operated. The same
considerations apply when considering selection under conventional farming for grow-
ing under organic farming (Tiemens-Hulscher et al. 2012).
In contrast, the response to direct selection in environment y (Ry) is


Ry ¼ iy hy σGy ¼ iy hy 2 σPy

It should be pointed out that when considering selection in clonal generations, one
can use all of the genetic variation (hence broad-sense heritabilities and genetic
correlation), whereas when considering selection between sexual generations, one can
only use the additive genetic variation (hence narrow-sense heritabilities and additive
genetic correlation).
Once the breeder has decided on the objectives of a particular programme, and
hence the important selection criteria, potential parents can be evaluated and
appropriate crosses made. These will be between parents that complement each
other for desirable major genes and QTL alleles of large effect and have high
general combining abilities (GCAs) for quantitative traits. If parental GCAs are
highly correlated with their phenotypes (which approximate to their genotypes if
accurately assessed), mid-parent value should provide a good prediction of mean
performance of the progeny, as explained by Bradshaw and Mackay (1994) and
illustrated with a table of examples. How well mid-parent values predict progeny
means depends upon the narrow-sense heritabilities (h2n) of the traits (in effect,
the amount of variation that can be explained by variation in GCAs); but these
will not be known exactly in advance as they are germplasm dependent (Bradshaw
et al. 2000a, b). Breeders are therefore likely to continue to make a relatively large
number of crosses to ensure that they include the best ones possible from the
available parents. It would be advantageous to predict the amount of variation
within crosses as those combining high variances (transgressive segregation giving
offspring superior to the better parent) with high means would be the best ones
from which to seek new cultivars (Bradshaw 2009b). However, only the parental
means can be recommended for predicting the usefulness of crosses (Bradshaw
et al. 1998).

Early-Generation Selection for Quantitative Traits

The potato breeding programme at SCRI before 1982 was typical of breeding since the
1960s in raising 100,000 seedlings in glasshouses each year, from 200 to 300 crosses
(Mackay 2005; Bradshaw 2009a; Bradshaw and Bonierbale 2010; Lindhout et al. 2011;
Eriksson et al. 2016). It was argued that a large number of seedlings was required
because of the large number of traits (around 20 at SCRI) being considered. However,
intense visual selection based on breeders’ visual preference of the seedlings and
subsequent spaced plants at a seed site was used to reduce this number to 4000 clones
for further evaluation (Bradshaw and Mackay 1994). Although crosses were made for
specific purposes, in practice they were used to generate a population of clones (4000)
176 Potato Research (2017) 60:171–193

from which to seek new cultivars by selection for economically important traits. This
meant that crosses made for a specific objective could fortuitously contribute offspring
to a different objective, an option that still appeals to breeders even if it is not the most
efficient way to go about breeding. Research in the 1980s confirmed that intense early-
generation visual selection for quantitative traits based on breeders’ visual preference
was very ineffective (Bradshaw and Mackay 1994; Bradshaw et al. 1998), although
selection for tuber skin and flesh colour and shape can be done if these are important for
particular consumers. The solution adopted at SCRI to this problem was to practise
selection between crosses before selection within the more promising ones. As this
solution has not been widely adopted, it is worth illustrating with a theoretical example
because it is important to the rest of the breeding strategy being advocated.

Selection between clones versus selection between crosses for quantitative

traits.

Theoretical Example

We do not need to worry about the complexities of tetrasomic inheritance, but simply
accept the arguments put forward by Simmonds (1996) about the amounts of genetic
variation between and within crosses being approximately equal and normally distrib-
uted. These are based on the additive genetic variances between and within full-sib
families being equal in a random mating population in equilibrium. As one cannot
accurately predict or estimate within cross variances, the assumption is made that all
crosses have the same within cross variance and hence there is no correlation between
the cross mean and within cross variance. It is convenient to use the binomial
distribution as an approximation of the normal distribution to construct a population
of seedlings/clones. Thus, 256 clones from each of 256 crosses will give a population
of 65,536 seedlings/clones. Furthermore, if the between crosses deviations are −4, −3,
−2, −1, 0, 1, 2, 3 and 4 in the ratios 1:8:28:56:70:56:28:8:1, and likewise the genetic
variation within each cross, the variance between crosses (value 2) equals the average
variance within crosses (value 2). The total variance between the 65,536 seedlings/
clones has a value of 4 (Fig. 1). The population mean can be set at an arbitrary value,
14000
12000
Number of Clones

10000
8000
6000 Clones

4000 Selected clones

2000
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Clone Value

Fig. 1 Clone values for 65,536 clones (256 clones from each of 256 crosses where variation between crosses
equals variation within crosses; both having a variance of 2) as a histogram, the clone values forming a
binomial distribution (range 2 to 18, mean = 10, variance = 4), which in practice would approximate to a
continuous normal distribution
Potato Research (2017) 60:171–193 177

say 10, for our quantitative trait which, for example, could be breeders’ visual
preference or yield. Finally, for the purposes of this exercise, we will not worry about
the field layout of the crosses and clones other than to assume that the experimental
design is such that there is no environmental variation between crosses additional to
that between clones.
If we select those clones with a value greater than 13 (i.e. ≥ 14), we select 2517
(3.8%), which is a similar percentage to SCRI’s 4000 out of 100,000 (4%) and to
figures of 2–3% quoted by Haynes et al. (2012) for breeding programmes in the USA.
As the standard deviation of the distribution has a value of 2, a yield of 14 is two
standard deviations from the mean. In the absence of environmental variation, the mean
of the selected clones is 14.34 (i.e. 4.34 from the mean of 10). The 2517 clones are
made up of 70, 504, 1036, 744 and 163 from crosses with means of 10, 11, 12, 13 and
14. Likewise Simmonds (1996) demonstrated statistically that the proportion of segre-
gates within crosses superior to standards (controls) falls away quickly with increasing
rank (best to worse) of crosses, where realistic standards (best cultivars currently
available) have means close to the means of the best crosses. Bradshaw (2009b) gave
an example of where this was true for yield in a 6 × 6 half diallel set of 15 crosses, in
which 218 offspring clones (harmonic mean of 14.31 clones per cross) were assessed at
SCRI in Dundee in 1993 and 1994. Hence, my first major conclusion is that below-
average crosses do not contribute any offspring to the selected clones and should be
eliminated at the earliest opportunity. But should one deliberately eliminate the worst
crosses rather than allow this to happen through clonal selection?
If there is a large amount of environmental variation, the selection between clones is
ineffective, and in the extreme case one would end up with a random sample of clones and
no progress. In contrast, the environmental deviations for clones within crosses would tend
to cancel out, so that most of the variation between crosses would be genetic, and in the
extreme case all of the variation would be genetic. Hence, in the extreme case, selection
between crosses would be effective, whereas selection between clones would be ineffec-
tive. For example, let us select the best 9 crosses (cross means ≥ 13) out of the 256 (i.e.
3.5%) (Fig. 2). As the standard deviation of the distribution has a value of 1.414, a cross
mean of 13 is over two standard deviations (SDs) from the mean of all crosses (12.828 is
mean plus 2 SDs). The mean of the 9 crosses is 13.11, and the ratio of effective selection
between crosses to effective selection between clones is 0.717 (3.11/4.34), which is
close to that (0.707) expected from the selection theory (Falconer and Mackay 1996).
80
70
Number of Crosses

60
50
40
Crosses
30
Selected crosses
20
10
0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Cross Mean

Fig. 2 Cross means for 256 crosses (256 clones within each cross) as a histogram, the cross means forming a
binomial distribution (range 6 to 14, mean = 10, variance = 2), which in practice would approximate to a
continuous normal distribution
178 Potato Research (2017) 60:171–193

700
600

Number of Clones
500
400
Random clones
300
200 Random clones from
selected crosses
100
0
2 3 4 5 6 7 8 9 101112131415161718
Clone Value

Fig. 3 Variation between the 2304 clones from the 9 best crosses compared with random sample of similar
size (2517) from a population of 65,536 clones

The 9 crosses provide 2304 clones which also have a mean of 13.11 (i.e. 3.11 units from
the mean) as no selection was practised among the 256 clones within each cross. We
can compare the distribution of these clones with a random sample of 2517 from the
population of 65,536 clones to see the effect of selecting the 9 best crosses (Fig. 3).
In summary then, if all of the variation is genetic, the mean of the best 2517 clones
selected from the population of 65,536 clones is 14.34 compared with 13.11 for the
2304 clones from the selection of the best 9 out of 256 crosses. However, if the
genetic variation is swamped by environmental variation, the mean of the 2304
clones from the best crosses will still be 13.11 whereas the mean of the 2517 clones
will be little better than the population mean of 10. So how much environmental
variation needs to occur before selection between crosses is superior to selection
between clones? The answer can be derived from the theory for comparing family
selection with individual selection given by Falconer and Mackay (1996). Selection
between crosses is superior to selection between individual seedlings/clones when,
for the latter, the environmental variation is greater than the genetic variation (broad-
sense heritability less than one half). In the paper on early-generation selection by
Bradshaw et al. (1998) mentioned earlier, selection for visual preference between
crosses was superior to selection between individuals (clones) in the seedling and
first clonal generation (spaced plants at seed site) despite more genetic variation
within than between crosses by a factor of 1.87. Hence, my second major conclusion
is that we should actively select the best crosses on the basis of cross means rather
than through indirect selection of the best individual clones.

Practical Example

Mackay (2005) gave an example from SCRI of discarding whole progenies

(crosses) in a successfully targeted and accelerated breeding programme for
processing potatoes which started with just 3100 seedlings from 43 crosses.
Seventeen progenies (crosses) were selected out of 43 grown at a seed site and
assessed for fry colour after storage at 4 and 10 °C. Superior clones were then
identified within these progenies on the basis of tuber traits, fry tests, dry matter
and trialling as ware at two sites proximal to processing plants in the third and
subsequent field generations. Three cultivars (Golden Millennium, Harborough
Harvest and Montrose) with enhanced (low temperature) storage characteristics
Potato Research (2017) 60:171–193 179

were placed on the UK National List in 1999 just 8 years after being raised as
seedlings in a glasshouse.

Selection Between Crosses for Quantitative Traits

Once the need to first practise selection between crosses has been accepted, a breeder is
not restricted to visual selection. Progenies (crosses) can be evaluated as pots of
seedlings for disease and pest resistance and plots of tuber progenies at normal spacing
in ware trials, without attempting to identify individual clones within crosses. The latter
can still be done in parallel on seedling progenies in the glasshouse and tuber progenies
of spaced plants at the seed site. When progenies (full-sib families) are assessed in
plots, the relevant environmental variation is between plots, but the effect of this
variation on progeny means can be reduced by replication with a consequent increase
in broad-sense heritability (h2b). Selection between crosses also allows longer term crop
improvement by cycles of full-sib family selection. Thus, Bradshaw et al. (1995,
2000b) assessed up to 143 crosses in seedling progeny tests for disease and pest
resistance and in replicated yield trials for genetic studies as part of a potato breeding
programme, the 143 coming from a 15 × 15 diallel set of crosses including selfs and
some reciprocal crosses. Furthermore, in a breeding programme operated at SCRI from
1991 to 2009, four cycles of selection were practised between 120 to 145 crosses, with
20% on average selected (Bradshaw et al. 2009).
The selection intensity could be increased by assessing more and selecting fewer
crosses (families). However, the latter will be limited by the minimum numbers of
families that need to be selected to avoid loss of favourable alleles and inbreeding
depression, say around 20. At present, the maximum number of families that can be
assessed in progeny tests is limited by resources to around 200. Hence, 20 out of 200 is
the best selection intensity (i = 1.742; Falconer and Mackay 1996) that can currently be
achieved. In our simple example, a higher selection intensity of 9 out of 256 crosses
was used to achieve a percentage of clones (3.5%) similar to that from selection of
individual seedlings/clones (4%), which was that used in the SCRI breeding pro-
gramme (4%). If half the crosses were eliminated for each of eight traits, one would
be left with one cross out of the 256 whereas to achieve 20 out of 200 crosses from
equal selection for eight traits would mean eliminating only one quarter for each trait
(i = 0.424 for each trait).
As more traits are selected, less progress will be made with each one, emphasizing
the need for continued recurrent selection with the length of each cycle as short as
possible so that small improvements accumulate into worthwhile ones (Bradshaw et al.
2003). For example, Bradshaw (2009b) did further analyses of the yield data from his
15 × 15 diallel set of crosses (Bradshaw et al. 2000b). If 10 out of the 105 full-sib
families were selected for yield and crossed to form the next generation, the predicted
response to selection was a 10% increase on the mean of all 105 F1’s (i = 1.749,
h2n = 0.425). However, with selection for other traits as well, that for yield is likely to
be for above average yields (i = 0.798), and this reduces the predicted response to 4.5%
per cycle, or 1.5% per annum at best with a cycle length of 3 years (these figures would
be down to 2.4% per cycle and 0.8% per annum if i was only 0.424, as above with eight
traits selected). A longer cycle time of 4 or 5 years would be required if the ware trials
180 Potato Research (2017) 60:171–193

were done for 1 or 2 years after the tuber progeny test with spaced plants. Indeed, there
is a strong argument for doing the ware trials in the target environment(s) to reduce the
effects of genotype × environment interactions. My third major conclusion is that the
intensity of selection will therefore limit progress from selection between crosses and
hence continued recurrent selection will be required to accumulate small improvements
into worthwhile ones (Bradshaw et al. 2003; Bradshaw 2009b). A low proportion of
additive to total genetic and phenotypic variance would reduce progress from recurrent
selection, so this needs to be assessed as done by Bradshaw et al. (2000b) who, for
example, found a narrow-sense heritability of 0.425 for yield, but a much higher one of
0.90 for fry colour. In conclusion, it would seem difficult to make much progress with
more than eight key traits, and ideally fewer should be targeted.

Selection Within the Most Promising Crosses for Quantitative Traits

Once the best crosses have been identified, the search can start within them for the best
clones for release as new cultivars and also for use as new parents, if one decides it is
worth increasing the length of the recurrent selection cycle to combine between and
within full-sib family selection. Selection for quantitative traits should begin as soon as
it is effective. Bradshaw et al. (1998), for example, found that visual selection of clones
at the unreplicated small plot stage produced correlated responses for faster emergence,
earlier maturity, higher yield and greater regularity of tuber shape. They therefore
recommended raising 200 seedlings from each of the best 20 crosses in sufficiently
large pots to progress 4000 clones to the field as small plots. This was the maximum
that had been handled at that stage in the conventional breeding programme at SCRI
prior to 1982 (Bradshaw and Mackay 1994). Likewise, 1000 out of the 4000 clones had
been selected for entry into the first year of replicated year trials and subsequently
reduced to 20 over three more years of trials (clonal generations).
Extensive assessment of offspring clones from the cross between processing clone
12601ab1 and table cultivar Stirling confirmed that the broad-sense heritabilities (h2b)
of clonal means (variation in clone means over years and replicates due to genetic
differences between clones) for 20 traits were all moderately high (> 0.54) to high
(> 0.75). The 20 traits included yield (h2b = 0.90) and agronomic performance, external
and internal defects, cooking and processing characteristics and disease and pest
resistance (Bradshaw et al. 2008; Hackett et al. 2014). Some of the lower values of
heritabilities were due to the presence of clones × years interactions, a reminder that
genotype × environment interactions can limit progress in potato breeding. Haynes
et al. (2012) concluded from their research on early-generation selection in the eastern
USA that evaluation in multiple locations should be done as early as possible to
identify clones with the broadest adaptation and hence the greatest likelihood of going
on to become new cultivars. They could achieve this in their programme by reducing
numbers assessed in the first field generation (16,500 to 11,000) but increasing them in
the second generation (350 to 1650). Mild selection in the second generation would
then allow a substantial number of clones (> 350) to be assessed at a number of
locations.
An overall selection intensity of 20 out of 4000 clones (i = 2.892) is larger than
that for between cross selection (i = 1.742 for 20 out of 200 crosses). Hence, more
Potato Research (2017) 60:171–193 181

progress is expected from practising selection between clones within the best
crosses than from the selection between crosses, assuming similar amounts of
genetic variation and similar broad-sense heritabilities, as seems reasonable from
our theory and experimental results. However, when considering longer-term crop
improvement, the duration of the recurrent selection cycle becomes important.
Bradshaw et al. (2009) took 3 years for selection between clones within their most
promising crosses, using independent culling levels, but more generations were
required in the conventional breeding programme at SCRI because more traits
were assessed (Bradshaw and Mackay 1994). With the intensities of selection just
considered, between and within cross selection intensity per year (i/T where T is
time in years) are equal (0.58) for 3 and 5 years, respectively. Hence, between
cross selection taking 3 years followed by within cross selection over 3 years is
superior to two cycles of between cross selection. Bradshaw et al. (2009) con-
cluded that selection between clones for disease and pest resistance was worth-
while in their breeding programme and recommended combined between and
within cross selection on a five-year cycle. My fourth major conclusion is that
the rates of progress in both cultivar production and continued crop improvement
are limited by the number of vegetative generations required to complete all of the
necessary assessments, as well as by the intensities of selection. One way to both
increase i and reduce T is through genomic selection, provided there is adequate
selection accuracy (r) as the response to selection per unit of time is proportional
to ir/T (Bassi et al. 2016; Slater et al. 2016).

Integrating Genomic Selection for Quantitative Traits

Genomic selection (GS) is now well established in commercial animal breeding

(Hayes et al. 2013) and the principles are conceptually straightforward. A refer-
ence (training) population of individuals is accurately phenotyped and genotyped
with molecular (genetic) markers to generate an equation for predicting an indi-
vidual’s phenotype from its marker content. This involves statistical methods such
as ridge regression that treat markers as random effects, because there are more
markers than individuals, and also cross-validation to avoid over-fitting the data.
The validation population is usually a sample of the training population not used
to generate the prediction equation. Other individuals (breeding population) can
then be genotyped and selected based on their predicted trait values (phenotypes)
alone. Methods for GS in plant breeding programmes are still being developed
with issues to resolve (Jonas and de Koning 2013; Bhat et al. 2016), although the
first theoretical paper on the use of GS in potato breeding has now been published
by Slater et al. (2016). In this section, therefore, untested ideas are advanced for
the use of GS in potato breeding based on recently published work in maize and
wheat (Crossa et al. 2014; Hickey et al. 2014; Mackay et al. 2015; Bassi et al.
2016), and apples (Kumar et al. 2012) and table grapes (Viana et al. 2016), as well
as the paper by Slater et al. (2016). These ideas need to be tested by computer
simulations that take into account knowledge of recently introgressed SNPs (Vos
et al. 2015) and the consequences of tetrasomic inheritance in potatoes (Bourke
et al. 2015). They also need to be tested in practice.
182 Potato Research (2017) 60:171–193

Genomic Selection Within the Most Promising Crosses

Genomic selection could be incorporated into selection for quantitative traits within the
most promising crosses (progenies). Let us consider the 4000 clones which could come
from raising 200 seedlings of each of the best 20 crosses in sufficiently large pots to
progress them straight to small plots in the field the following year. All 4000 could be
maintained as small plots at the seed site and genotyped, possibly with as few as 200–
500 markers for prediction within crosses where close relatives share long haplotypes
because they are separated by a small number of meioses (Hickey et al. 2014). In
addition, 50 from each cross (1000 in total) could be advanced into replicated yield
trials and assessed for a number of key traits. Ideally, these yield trials should be done in
the target environment(s) to prevent genotype × environment interactions reducing the
responses to selection. The phenotypic data for these key traits could be combined into
a selection index. In the absence of economic weights, desired relative improvements
over industry standards for each trait can be used. The index will then take account of
differences in heritabilities and correlations between traits, and hence prove superior to
independent culling levels for overall merit of a clone (Bradshaw et al. 2003; Bradshaw
2016). Genomic selection could then be practised on the remaining 150 clones from
each progeny to provide more potential cultivars without loss of time. In other words,
they would be assessed alongside those that had undergone phenotypic selection. This
would reduce the amount of phenotypic evaluation required to identify potential
cultivars, particularly in special tests and costly field trials conducted over seasons
and locations, but it would be dependent on the precision with which yield and other
traits could be estimated for 1000 clones in large trials with small plots and few
replicates. In other words, can 20 out of 4000 clones be selected on the basis of 1 year’s
phenotypic and genotypic data? Furthermore, will the economic cost of genomic
selection be cheaper than phenotypic evaluation in terms of the achieved responses to
selection (outcomes)?
Clones could also be selected for use as parents in the next round of crossing
but they would come from different crosses and hence be less closely related.
Subsequent rounds of inter-crossing would further reduce any relationships. Prog-
ress in crop improvement would be much faster if the prediction equations derived
from the total set of 1000 (20 crosses × 50 clones) phenotypes and marker
genotypes in one generation were accurate enough for use in the next generation
(i.e. to practise selection in the glasshouse on the offspring from the next round of
crosses). This would allow an additional breeding cycle of effective selection
between clones in 2 years. The theoretical results from Slater et al. (2016) are
encouraging and suggest that more molecular markers would be required but not
necessarily larger population sizes. Their estimates of the number of SNP markers
required were 8000 based on decay of linkage disequilibrium and genome map
length, and 14,789 based on effective population size of their breeding population.
Their estimates of selection accuracy ranged from 0.19, for a narrow-sense
heritability of 0.23 and a reference population size of 500, to 0.77 for a narrow-
sense heritability of 0.86 and a reference population size of 5000. Even with the
low selection accuracy, GS was worthwhile compared with phenotypic selection
due to the reduction in cycle time from one generation to the next. Their selection
intensity was 10% for parents for crossing, but the effectiveness of much higher
Potato Research (2017) 60:171–193 183

intensities (20 out of 4000 clones is 0.5%) would be worth exploring. If higher
intensities could be used with GS than with phenotypic selection, this would be an
advantage in addition to the shorter cycle time. However, my fifth major conclu-
sion, albeit tentative, is that successful implementation of GS may depend upon
the development of appropriate high-throughput phenotyping for use in the field
and on harvested material, as reviewed by Araus and Cairns (2014). Prashar et al.
(2013) have demonstrated that infrared thermography can be used as an easy,
rapid and non-destructive screening method for field evaluation of potato trials,
and have found a negative association between canopy temperature and final tuber
yield in a diploid mapping population grown under ample moisture supply.
Bernhard et al. (2016) have demonstrated that near-infrared reflection spectrosco-
py (NIRS) could be used to predict crude protein and dry matter content in fresh
and dried potato tuber samples. However, there are not yet enough examples to
say if large-scale phenotyping is going to be the limiting factor in genomic
selection.

Integrating Marker-Assisted Selection for Major Genes and QTL Alleles

of Large Effect

As genetic knowledge of the potato accumulates, it will become easier to choose

parents known to possess desired major genes and large-effect QTLs, and to select
genotypically for these in their offspring in a greenhouse. The desired allele may have
an easily recognized phenotype or be its own molecular marker, or be extremely tightly
linked to a diagnostic molecular marker, or lie between closely linked flanking markers.
Major genes have already been mapped for flesh, skin and flower colour, for tuber
shape and eye depth and for resistances to late blight, cyst nematodes, root-knot
nematodes, viruses (PVY, PVA, PVX, PVM, PVS and PLRV) and wart (for
references see Bradshaw and Bonierbale 2010; Gebhardt 2013; Slater et al. 2014).
Large-effect QTLs have also already been mapped for total glycoalkaloid content
(TGA), maturity, canopy height and resistances to late blight, Verticillium wilt, cyst
nematodes and PLRV (Bradshaw and Bonierbale 2010; Gebhardt 2013; Slater et al.
2014). QTLs for agro-morphological and quality traits have been both confirmed and
extended through marker-trait association analysis of diverse sets of genotypes from
collections of interest to breeders (D’hoop et al. 2014). Some quantitative traits such as
yield will still best be viewed as complex polygenic traits and some genotypic variation
will remain unexplained by QTL alleles of large effect. We therefore need to consider
integrating marker-assisted selection with selection for complex polygenic traits.
Potato breeders will need to consider how many major genes and QTL alleles
of large effect they can combine in a single cultivar, both from a single cross and
longer term over cycles of crossing and selection. The answer is determined by the
frequencies of the desired combinations of alleles (genotypes) relative to the
population size of the breeding programme. We are therefore now going to assume
that our chosen crosses (for example, the 20 of previous sections) are each also
segregating for desired major genes and QTL alleles of large effect, and that we
have diagnostic markers for them so that we can practise marker-assisted selec-
tion. We are going to consider selecting clones that have all of the desired
184 Potato Research (2017) 60:171–193

markers, thus eliminating any that lack one or more markers. Although a single
copy of a gene/allele may be enough to achieve the desired phenotype, unless the
allele is eventually fixed in one’s parental material, it will continue to segregate in
crosses and hence limit the number of alleles from different loci that can be
combined (stacked).

Theory for Marker-Assisted Selection

Let us start by considering one locus and all initial crosses of type Aaaa × aaaa so
that in the absence of double reduction, the offspring are ½Aaaa and ½aaaa. Let
us discard all offspring lacking an A allele and randomly inter-cross the remaining
genotypes (Aaaa × Aaaa) so that in the absence of double reduction, the off-
springs are ¼AAaa, ½Aaaa and ¼aaaa. Once again, let us discard all offspring
lacking an A allele and randomly inter-cross the remaining genotypes (1/3 AAaa
and 2/3 Aaaa), and keep repeating the process. Each generation gametes AA, Aa
and aa in the ratio r: s: t are randomly mated to produce genotypes AAAA, AAAa,
AAaa, Aaaa and aaaa in the ratio r2, 2rs, 2rt + s2, 2st and t2, and genotype aaaa is
discarded. The results for seven generations are shown in column two of Table 1.
The frequency of A— increases from 0.5000 to 0.9524, but even at this high
frequency, that of genotype AAAA is only 0.0527. This is because genotype AAAA
approaches fixation slowly as the frequency of allele A increases (frequency of A
is 0.506 in generation seven).
The frequencies for two unlinked loci are simply the squares of those for one
locus, and likewise as more unlinked loci are added. If just two loci were
segregating, one could screen 4000 seedlings from the initial crosses, in the
glasshouse with molecular markers, and expect to find about 1000 with both
markers to take to the field for phenotypic evaluation of other traits. If four loci
were segregating, one would need to screen 16,000 seedlings in the glasshouse
to find about 1000 with all four markers. Alternatively, one could screen 4000
seedlings in the glasshouse and random mate the 250 or so with all four markers,
and then raise another seedling generation in the glasshouse. This time, one
would expect over 1250 out of 4000 seedlings to have all four markers. If eight

Table 1 Marker-assisted selection for presence of at least one copy of desirable allele (A) at from 1 to 16
unlinked loci: frequencies of desired genotypes

Generation Loci

1 2 3 4 8 12 16
1 0.5000 0.2500 0.1250 0.0625 0.003906 0.000244 0.000015
2 0.7500 0.5625 0.4219 0.3164 0.1001 0.03168 0.01002
3 0.8488 0.7205 0.6115 0.5191 0.2694 0.1399 0.07259
4 0.8953 0.8016 0.7176 0.6426 0.4128 0.2652 0.1704
5 0.9226 0.8512 0.7853 0.7245 0.5249 0.3803 0.2756
6 0.9403 0.8842 0.8314 0.7818 0.6112 0.4778 0.3736
7 0.9524 0.9070 0.8638 0.8227 0.6769 0.5569 0.4581
Potato Research (2017) 60:171–193 185

loci were segregating, one would need to screen some 256,000 seedlings in the
glasshouse to find about 1000 with all eight markers. However, one could screen
4000 seedlings in the glasshouse and random mate the 16 or so with all eight
markers, and then raise another seedling generation in the glasshouse. This time,
one would expect about 400 out of 4000 seedlings to have all eight markers.
Another cycle of crossing and selecting would result in over 1000 out of 4000
seedlings having all eight markers. If there were 16 loci segregating, only 1 in
65,536 seedlings would have all 16 markers. Thus, above about 8 loci segregat-
ing, marker-assisted selection soon becomes impractical to get started for all the
desired alleles. However, once one is making crosses between clones that all
have every marker, a steady increase in the frequencies of such clones occurs
over generations simply through elimination of offspring clones not possessing
all of them (Table 1), but even after seven generations, only about 11 in 4000
seedlings will have reached fixation at two loci.
These theoretical results provide an explanation for the lack of potato cultivars
in which more than a few major genes and QTL alleles of large effect have been
stacked. In the SCRI potato breeding programme (Bradshaw 2009a), introgression
of resistance genes from the wild and cultivated potato species of Latin America
started for late blight in 1932, for viruses in 1941 and for potato cyst nematodes in
1952. After over 75 years of breeding, no cultivars have been produced with the
number of major genes (8) we have been discussing in this section. Deliberate
parental line breeding in 5-year cycles was done at the Plant Breeding and
Acclimatization Institute (IHAR)-National Research Institute in Poland from the
early 1950s, with the aim of producing parents with multiple resistances to six
viruses, late blight and soft rot (Zimnoch-Guzowska et al. 2013). From 1968 to
1993 (25 years), the number of major genes for virus resistance in parental lines
was increased from two to five, and recently, a molecular marker (CAPS
GP122718) has been made available to commercial breeders for selecting for the
Ry-fsto gene for complete resistance to all known strains of PVY. In India, Sharma
et al. (2014) used diagnostic molecular markers to screen a diverse collection of
126 potato genotypes in the National Germplasm Repository for the presence of
three genes for resistance to late blight, two for resistance to PVY and three for
resistance to potato cyst nematodes. They found 8 out of the 126 genotypes
possessed a maximum of four or five genes. More generally, Vos et al. (2015)
found that since 1945 introgression breeding had introduced 24% of the 14,530
SNPs they examined, but by 2005 the highest frequencies of introduced alleles
were 16 and 18% for two SNPs introduced before 1962. Returning to the theo-
retical results, my sixth major conclusion is that with a determined effort, breeders
could use marker-assisted selection to stack up to eight unlinked genes in new
cultivars, but would struggle to get much beyond eight.

Stacking More than Eight Genes

One way of stacking more than eight genes would be to fix the desired allele at
each locus, but this is more easily said than done. Let us explore marker-assisted
selection with the ability to recognize and select for allele dosage. Breeding to
produce multiplex parents for disease and pest resistance genes was first
186 Potato Research (2017) 60:171–193

discussed by Cadman (1942) with respect to the Nx gene for race-specific

resistance to PVX, and explained in detail by Toxopeus (1953) and Bradshaw
and Mackay (1994). The scheme is straightforward for a single locus, but takes
three generations:


Aaaa  Aaaa→1 4 AAaa

AAaa  AAaa→1 4 ðAAAA; AAAa in ratio 1 : 8Þ

AAAa  AAAa→1 4 AAAA

The frequencies for two unlinked loci are simply the squares of those for one
locus, and likewise as more unlinked loci are added. If four loci were segregating,
1 in 256 seedlings each (glasshouse) generation would have the desired genotype,
so a population of 4096 seedlings would on average provide 16 for the next round
of crossing and selection. However, with six unlinked loci, just 1 in 4096
seedlings would have the desired genotype. In order to fix eight loci with similar
population sizes, one would need additional generations with crosses of type
AAaa × Aaaa, AAAa × AAaa and AAAA × AAAa. Hence, fixing four loci followed
by another four, or fixing all eight simultaneously, would take six generations in
the glasshouse without selection for any other traits. This is clearly a lot of work
so breeders may be content to practise marker-assisted selection with elimination
of clones not possessing the desired markers. This would allow them to combine
marker-assisted selection with selection for quantitative traits. Assessment of 4000
seedlings in the glasshouse and selection of ones with all of the markers can result
in 1000 clones being advanced to the field and grown in unreplicated small-plots,
provided one or two cycles of crossing and further selection are done, depending
on the number of loci segregating. The 1000 clones could then be phenotyped and
genotyped to enable genomic selection to be practised on the offspring of the ones
selected for use as parents. The seedling population size would need to be large
enough to combine the genomic selection (GS) with marker-assisted selection
(MAS). If the loci used in MAS can be matched to markers generated in GS
genotyping, then a single genotyping platform can be used for MAS and GS.

Gene Editing

Gene editing using systems like CRISPR/Cas9 (Bortesi and Fischer 2015) offers
a way of creating and fixing novel alleles. For example, Andersson et al. (2017)
demonstrated that transient application of CRISPR-Cas9-mediated genome
editing in protoplasts of tetraploid potato (Solanum tuberosum) yielded mutations
in all four alleles of the gene encoding granule-bound starch synthase (GBSS) in
up to 2% of regenerated lines from a single transfection. Full knockout of GBSS
enzyme activity was confirmed in four-allele mutated lines by phenotypic studies
of starch. One remaining wild-type allele was shown sufficient to maintain
enough GBSS enzyme activity to produce significant amounts of amylose.
Hence, the CRISPR-Cas9 system is an efficient way to fix a recessive mutant
by site-directed mutagenesis (gene knockouts), but more research is required to
see if it can be used for gene editing to create and fix novel alleles in potatoes.
Potato Research (2017) 60:171–193 187

Genetic Transformation of Potatoes

We have seen in the previous sections that improvements can be made to current
breeding practice. However, the various methods proposed still have limitations, and
this means that progress in potato improvement will be faster but not dramatic in the
short term. In other words, a number of cycles of hybridization and selection will be
required for substantial progress. Hence, when planning potato improvement
programmes, the possibility of major improvements by genetic transformation should
first be given serious consideration. It would seem sensible to see if currently widely
grown cultivars have limitations which could be overcome by genetic transformation,
including cisgenesis (late blight resistance: Jo et al. 2014; Haverkort et al. 2016), gene
silencing by RNA interference (reducing sugars and cold sweetening: Rommens et al.
2006; asparagine and acrylamide: Chawla et al. 2012; enzymic browning and black-
spot bruising: Chi et al. 2014; late blight resistance: Sun et al. 2016) and transgenesis
(Colorado potato beetle and viruses PLRV and PVY: Duncan et al. 2002; potato tuber
moth: Douches and Grafius 2005; protein content and amino acid balance: Chakraborty
et al. 2000; improved nutritional value by inulin production: Hellwege et al. 2000;
carotenoid content: Ducreux et al. 2005; Morris et al. 2006). This list of traits and
references is representative rather than complete, but certain types of traits are currently
missing, namely yield, photosynthetic efficiency and efficiencies of nutrient and water
use. Some crop ecologists have expressed doubts about biotechnology providing major
improvements to these traits because any simple (major gene) solutions would already
have been found by millions of years of natural selection (Denison 2012). Hence, it
seems premature to provide a definitive list of potato traits for improvement by genetic
transformation, but the number of potential targets is steadily increasing.
If it is concluded that some trait improvements are easier to achieve by genetic
transformation than by conventional breeding, the breeder can make the conscious
decision to improve other traits by conventional breeding before doing the genetic
modification. However, delays in making these improvements to new cultivars just
before or after their release could occur for two reasons. Firstly, the need to select the
best transformed potato clones for commercialization requires a lot of work, namely
seeking the best constructs and promoters and eliminating any undesirable
transformants or somaclonal variants (Haverkort et al. 2016) that arise from the
transformation per se and in vitro regeneration, respectively. Davies (2002), for exam-
ple, recommended the production of several hundred independently transformed lines
followed by field selection as normally done by plant breeders in their programmes.
Secondly, the required combinations of traits may require stacking genes through co-
transformation with separate constructs, transformation with a multi-gene cassette (Jo
et al. 2014) (four to six appears to be current limit), or re-transformation with further
genes. Re-transformation was used in both the DuRPh project at Wageningen Univer-
sity and Research Centre (Haverkort et al. 2016) and in the production of Simplot’s
Innate™ potatoes (James 2015). Simplot separately incorporated two cassettes into
processing cultivars using intragenic approaches, each cassette silencing two genes (for
Asn1 and Ppo5, and PhL and R1), making four in total (Rommens et al. 2006, 2008).
The DuRPh project planned to insert a stack of multiple R genes for resistance to late
blight, from crossable wild potato species, into established cultivars using marker-free
Agrobacterium tumefaciens-mediated transformations (i.e. by cisgenesis). A cisgenic
188 Potato Research (2017) 60:171–193

cultivar with three R genes was achieved in two stages. Cultivar Atlantic was trans-
formed with a combination of two R-genes in one T-DNA and then re-transformed with
a third R-gene. Eventually, transformation with three genes in one T-DNA was
achieved.

Integration of Genetic Transformation Into Conventional Programme

Given the potential delays in modifying new cultivars just before or after their
release, we need to consider if the genetic transformation could be incorporated
at an earlier stage into the breeding programme. As widely grown cultivars will
be used as parents in the conventional breeding, one can envisage genetically
modified ones being used as parents, followed by selection of offspring contain-
ing the introduced gene(s) for further assessment. Where selection for quantita-
tive traits is first practised between crosses, followed by selection within the
most promising ones, the following can be done. The untransformed cultivars
can be used in crosses to determine the best ones, and then the transformed
versions used in repeats of the best crosses to provide more seed on which to
practise within progeny selection. Field trialling of the former would not be
governed by any GM regulations, and selection within the latter could initially
be done in a greenhouse. However, the same problem would occur as with
marker-assisted selection for major genes and QTL alleles of large effect. One
would be limited by the required population sizes to the segregation of around
eight genes. However, in the future, the use of site-directed transformation
(rather than random insertion) should reduce the number of undesirable
transformants and allow gene stacking at a single site to create a supergene that
will be inherited as a Mendelian gene (Mahfouz et al. 2016). It should then be
possible to stack genes more quickly at relatively few sites. The segregation of
such units will not pose any major logistical problems in a potato improvement
programme combining genetic transformation with conventional breeding that
includes marker-assisted and genomic selection.

Summary

1) Potato breeding programmes commonly start with 100,000 seedlings because

improvement in a large number of traits (at least 20) is being attempted. However,
intense visual selection based on breeders’ visual preference of the seedlings and
subsequent spaced plants at a seed site is then used to reduce this number to around
4000 clones for the start of assessment for economically important traits. Further-
more, low heritability of breeders’ visual preference negates any potential advan-
tage of high selection intensity. Overall, such early-generation selection is
ineffective.
2) Selection both between progenies (crosses/full-sib families) and then within the
most promising ones can be done for economically important quantitative traits.
Replication of progenies and clones can achieve high broad-sense heritabilities and
assessment in target environments can ensure that the responses to selection are not
reduced by genotype × environment interactions. However, the intensities of
Potato Research (2017) 60:171–193 189

selection will be limited by the numbers of progenies (200) and clones (4000) that
can be assessed.
3) As more quantitative traits are selected, less progress will be made with each one,
emphasizing the need for continued recurrent selection with the length of each
cycle as short as possible so that small improvements accumulate into worthwhile
ones. A low proportion of additive to total genetic and phenotypic variance will
reduce the rate of progress from recurrent selection. It would seem difficult to
make much progress with more than eight key traits, and ideally fewer should be
targeted.
4) The rates of progress in both cultivar production and continued crop improvement
by recurrent selection are limited by the number of vegetative generations, and
hence years (T), required to complete all of the necessary assessments, as well as
by the intensities of selection (i). The biggest reduction in T could come from
genomic selection of quantitative traits on seedlings in the glasshouse, and this
would translate into faster progress provided there is adequate selection accuracy
(r) from reference population (genotyped and phenotyped) to breeding population
(genotyped), as the response to selection per unit of time is proportional to ir/T.
The response would be even greater if higher selection intensities could be used.
The size of reference population required for calibrating the breeding material
being used (possible range 1000 to 4000) will determine if there is a need to
develop high-throughput field phenotyping. The number of molecular markers
(order of 10,000) should not be a problem.
5) As more and more desirable major genes and QTL alleles of large effect are
identified, breeders will need to consider how they can be combined in a single
cultivar, both from a single cross and longer term over cycles of crossing and
selection (on glasshouse grown seedlings). As the number of genes/alleles in-
creases, the frequency of the desired combination of alleles (genotypes) decreases,
necessitating larger and larger initial population sizes and/or an increasing number
of generations over which to increase the frequency of the desired combination.
The number of clones possessing the desired combination will need to be large
enough to allow worthwhile selection for other quantitative traits. Theoretical
results indicate that breeders could use marker-assisted selection to stack up to
eight unlinked genes in new cultivars, but would struggle to get much beyond
eight.
6) A number of cycles of crossing and selection will be required for worthwhile
improvements in (polygenic) quantitative traits as well as achieving desirable
combinations of major genes and QTL alleles of large effect. Hence, when
planning potato improvement programmes, the possibility of major improvements
to existing and potential cultivars by genetic transformation should first be given
serious consideration. Improvements to some traits (e.g. disease and pest resistance
and quality traits) are proving easier than others (e.g. yield and water and nutrient
use efficiencies). Stacking eight transgenes is the most being considered at present,
but site-directed transformation should allow genes to be stacked more quickly at
relatively few sites, as well as reducing the number of undesirable transformants.
7) The challenge for potato breeders and biotechnologists is to design breeding
programmes that integrate marker-assisted selection of specific alleles, genomic
selection of unspecified alleles and phenotypic selection, having decided when a
190 Potato Research (2017) 60:171–193

gene editing or transgenic approach is more appropriate, given consumer accept-

ability of the latter. In doing so, they need to appreciate both the opportunities and
limitations afforded by the new methods.

Acknowledgements Discussions with Professor Ian Mackay and Dr. Keith Gardner of The John Bingham
Laboratory, NIAB, Cambridge, on genomic selection in wheat are much appreciated. I would also like to
thank the editor and referees for drawing my attention to additional references as well as other constructive
suggestions.

References

Andersson M, Turesson H, Nicolia A, Fält A-S, Samuelsson M, Hofvander P (2017) Efficient targeted
multiallelic mutagenesis in tetraploid potato (Solanum tuberosum) by transient CRISPR-Cas9 expression
in protoplasts. Plant Cell Rep 36:117–128
Araus JL, Cairns JE (2014) Field high-throughput phenotyping: the new crop breeding frontier. Trends Plant
Sci 19:52–61. doi:10.1016/j.tplants.2013.09.008
Barone A, Sebastiano A, Carputo D, della Rocca F, Frusciante L (2001) Molecular marker-assisted introgres-
sion of the wild Solanum commersonii genome into the cultivated S. tuberosum gene pool. Theor Appl
Genet 102:900–907
Bassi FM, Bentley AR, Charmet G, Ortiz R, Crossa J (2016) Breeding schemes for the implementation of
genomic selection in wheat (Triticum spp.) Plant Sci 242:23–36. doi:10.1016/j.plantsci.2015.08.021
Bernhard T, Truberg B, Friedt W, Snowdon R, Wittkop B (2016) Development of near-infrared reflection
spectroscopy calibrations for crude protein and dry matter content in fresh and dried potato tuber samples.
Potato Res 59:149–165
Bhat JA, Ali S, Salgotra RK, Mir ZA, Dutta S, Jadon V, Tyagi A, Mushtaq M, Jain N, Singh PK, Singh GP,
Prabhu KV (2016) Genomic selection in the era of next generation sequencing for complex traits in plant
breeding. Front Genet. doi:10.3389/fgene.2016.00221
Bortesi L, Fischer R (2015) The CRISPR/Cas9 system for plant genome editing and beyond. Biotechnol Adv
33:41–52
Bourke PM, Voorrips RE, Visser RGF, Maliepaard C (2015) The double-reduction landscape in tetraploid
potato as revealed by a high-density linkage map. Genetics 201:853–863. doi:10.1534/genetics.115.181008
Bradshaw JE (2009a) Potato breeding at the Scottish Plant Breeding Station and the Scottish Crop Research
Institute: 1920-2008. Potato Res 52:141-172
Bradshaw JE (2009b) A genetic perspective on yield plateau in potato. Potato J 36:79–94
Bradshaw JE (2016) Plant breeding: past, present and future. Springer, Switzerland
Bradshaw JE, Bonierbale M (2010) Potatoes. In: Bradshaw JE (ed) Root and tuber crops, handbook of plant
breeding 7. Springer, New York, pp 1–52
Bradshaw JE, Mackay GR (1994) Breeding strategies for clonally propagated potatoes. In: Bradshaw JE,
Mackay GR (eds) Potato genetics. CAB International, Wallingford, pp 467–497
Bradshaw JE, Stewart HE, Wastie RL, Dale MFB, Phillips MS (1995) Use of seedling progeny tests for
genetical studies as part of a potato (Solanum tuberosum subsp. tuberosum) breeding programme. Theor
Appl Genet 90:899–905
Bradshaw JE, Dale MFB, Swan GEL, Todd D, Wilson RN (1998) Early-generation selection between and
within pair crosses in a potato (Solanum tuberosum subsp. tuberosum) breeding programme. Theor Appl
Genet 97:1331–1339
Bradshaw JE, Lees AK, Stewart HE (2000a) How to breed potatoes for resistance to fungal and bacterial
diseases. Plant Breed Seed Sci 44:3–20
Bradshaw JE, Todd D, Wilson RN (2000b) Use of tuber progeny tests for genetical studies as part of a potato
(Solanum tuberosum subsp. tuberosum) breeding programme. Theor Appl Genet 100:772–781
Bradshaw JE, Dale MFB, Mackay GR (2003) Use of mid-parent values and progeny tests to increase the
efficiency of potato breeding for combined processing quality and disease and pest resistance. Theor Appl
Genet 107:36–42
Potato Research (2017) 60:171–193 191

Bradshaw JE, Bryan GJ, Ramsay G (2006) Genetic resources (including wild and cultivated Solanum species)
and progress in their utilisation in potato breeding. Potato Res 49:49–65
Bradshaw JE, Hackett CA, Pande B, Waugh R, Bryan GJ (2008) QTL mapping of yield, agronomic and
quality traits in tetraploid potato (Solanum tuberosum subsp. tuberosum). Theor Appl Genet 116:193–211
Bradshaw JE, Dale MFB, Mackay GR (2009) Improving the yield, processing quality and disease and pest
resistance of potatoes by genotypic recurrent selection. Euphytica 170:215–227
Brown J, Dale MFB, Mackay GR (1996) General adaptability of potato genotypes selected in the UK for the
Mediterranean region. J Agric Sci Camb 126:441–448
Cadman CH (1942) Autotetraploid inheritance in the potato: some new evidence. J Genet 44:33–52
Chakraborty S, Chakraborty N, Datta A (2000) Increased nutritive value of transgenic potato by expressing a
non-allergenic seed albumin gene from Amaranthus hypochondriacus. Proc Nat Acad Sci 97:3724–3929
Chawla R, Shakya R, Rommens CM (2012) Tuber-specific silencing of asparagine synthetase-1 reduces the
acrylamide-forming potential of potatoes grown in the field without affecting tuber shape and yield. Plant
Biotechnol J 10:913–924
Chi M, Bhagwat B, Lane WD, Tang G, Su Y, Sun R, Oomah BD, Wiersma PA, Xiang Y (2014) Reduced
polyphenol oxidase gene expression and enzymatic browning in potato (Solanum tuberosum L.) with
artificial microRNAs. BMC Plant Biology 14:62. doi:10.1186/1471-2229-14-62
Crossa J, Pérez P, Hickey J, Burgueño J, Ornella L, Cerón-Rojas J, Zhang X, Dreisigacker S, Babu R, Li Y,
Bonnett D, Mathews K (2014) Genomic prediction in CIMMYT maize and wheat breeding programs.
Heredity 112:48–60. doi:10.1038/hdy.2013.16
Davies HV (2002) Commercial developments with transgenic potato. In: Valpuesta V (ed) Fruit and vegetable
biotechnology. Woodhead Publishing Limited, Cambridge, pp 222–249
Denison RF (2012) Darwinian agriculture. Princeton University Press, Princeton
D'hoop BB, Keizer PL, Paulo MJ, Visser RG, van Eeuwijk FA, van Eck HJ (2014) Identification of
agronomically important QTL in tetraploid potato cultivars using a marker-trait association analysis.
Theor Appl Genet 127:731–748. doi:10.1007/s00122-013-2254-y
Douches DS, Grafius EJ (2005) Transformation for insect resistance. In: Razdan MK, Mattoo AK (eds)
Genetic improvement of Solanaceous crops volume I: potato. Science Publishers, Enfield, pp 235–266
Ducreux LJM, Morris WL, Hedley PE, Shepherd T, Davies HV, Millam S, Taylor MA (2005) Metabolic
engineering of high carotenoid potato tubers containing enhanced levels of β-carotene and lutein. J Exp
Bot 56:81–89
Duncan DR, Hammond D, Zalewski J, Cudnohufsky J, Kaniewski W, Thornton M, Bookout JT, Lavrik P,
Rogan GJ, Feldman-Riebe J (2002) Field performance of transgenic potato, with resistance to Colorado
potato beetle and viruses. Hortscience 37:275–276
Eriksson D, Carlson-Nilsson U, Ortíz R, Andreasson E (2016) Overview and breeding strategies of table
potato production in Sweden and the Fennoscandian region. Potato Res 59:279–294
Falconer DS, Mackay TFC (1996) Introduction to quantitative genetics, 4th edn. Longman, Harlow
Gebhardt C (2013) Bridging the gap between genome analysis and precision breeding in potato. Trends Genet
29:248–256
Gebhardt C, Urbany C, Li L, Stich B, Paulo J, Draffehn A, Ballvora A (2011) Molecular diagnostics for
complex pest and disease resistance and tuber quality traits: concept, achievements and perspectives.
Potato Res 54:313–318
Hackett CA, Bradshaw JE, Bryan GJ (2014) QTL mapping in autotetraploids using SNP dosage information.
Theor Appl Genet 127:1885–1904
Hamilton JP, Hansey CN, Whitty BR, Buell CR (2011) Single nucleotide polymorphism discovery in elite
North American potato germplasm. BMC Genomics 12(1):302. doi:10.1186/1471-2164-12-302
Haverkort AJ, Boonekamp PM, Hutten R, Jacobsen E, Lotz LAP, Kessel GJT, Vossen JH, Visser RGF (2016)
Durable late blight resistance in potato through dynamic varieties obtained by cisgenesis: scientific and
societal advances in the DuRPh project. Potato Res 59:35–66
Hayes BJ, Lewin HA, Goddard ME (2013) The future of livestock breeding: genomic selection for efficiency,
reduced emissions intensity, and adaptation. Trends Genet 29:206–214
Haynes KG, Gergela DM, Hutchinson CM, Yencho GC, Clough ME, Henninger MR, Halseth DE, Sandsted
E, Porter GA, Ocaya PC (2012) Early generation selection at multiple locations may identify potato
parents that produce more widely adapted progeny. Euphytica 186:573–583. doi:10.1007/s10681-012-
0685-1
Hellwege EM, Czapla S, Jahnke A, Willmitzer L, Heyer AG (2000) Transgenic potato (Solanum tuberosum)
tubers synthesise the full spectrum of inulin molecules naturally occurring in globe artichoke (Cynara
scolymus). Proc Nat Acad Sci 97:8699–8704
192 Potato Research (2017) 60:171–193

Hickey JM, Dreisigacker S, Crossa J, Hearne S, Babu R, Prasanna BM, Grondona M, Zambelli A,
Windhausen VS, Mathews K, Gorjanc G (2014) Evaluation of genomic selection training population
designs and genotyping strategies in plant breeding programs using simulation. Crop Sci 54:1476–1488
Hijmans RJ (2001) Global distribution of the potato crop. Am J Potato Res 78:403–412
Hirsch CN, Hirsch CD, Felcher K, Coombs J, Zarka D, Van Deynze A, De Jong W, Veilleux RE, Jansky S,
Bethke P, Douches DS, Buell CR (2013) Retrospective view of North American potato (Solanum
tuberosum L.) breeding in the 20th and 21st centuries. G3 3:1003–1013
James C (2015) Global status of commercialized biotech/GM crops: 2015. ISAAA Brief No. 51. ISAAA:
Ithaca, NY
Jo K-R, Kim C-J, Kim S-J, Kim T-Y, Bergervoet M, Jongsma MA, Visser RGF, Jacobsen E, Vossen JH (2014)
Development of late blight resistant potatoes by cisgene stacking. BMC Biotechnol 14: 50. http://www.
biomedcentral.com/1472-6750/14/50
Jonas E, de Koning D-J (2013) Does genomic selection have a future in plant breeding? Trends Biotechnol 31:
497–504
Kumar S, Bink MCAM, Volz RK, Bus VGM, Chagné D (2012) Towards genomic selection in apple (Malus ×
domestica Borkh.) breeding programmes: prospects, challenges and strategies. Tree Genet Genomes 8:1–
14. doi:10.1007/s11295-011-0425-z
Lindhout P, Meijer D, Schotte T, Hutten RCB, Visser RGF, van Eck HJ (2011) Towards F1 hybrid seed potato
breeding. Potato Res 54:301–312
Mackay GR (2005) Propagation by traditional breeding methods. In: Razdan MK, Mattoo AK (eds) Genetic
improvement of Solanaceous crops volume I: potato. Science Publishers, Enfield, pp 65–81
Mackay I, Ober E, Hickey J (2015) GplusE: beyond genomic selection. Food Energy Secur 4:25–35.
doi:10.1002/fes3.52
Mahfouz MM, Cardi T, Stewart CN (2016) Next-generation precision genome engineering and plant
biotechnology. Plant Cell Rep 35:1397–1399. doi:10.1007/s00299-016-2009-8
Moloney C, Griffin D, Jones PW, Bryan GJ, McLean K, Bradshaw JE, Milbourne D (2010) Development of
diagnostic markers for use in breeding potatoes resistant to Globodera pallida pathotype Pa2/3 using
germplasm derived from Solanum tuberosum ssp. andigena CPC 2802. Theor Appl Genet 120:679–689.
doi:10.1007/s00122-009-1185-0
Morris WL, Ducreux LJM, Fraser PD, Millam S, Taylor MA (2006) Engineering ketocarotenoid biosynthesis
in potato tubers. Metab Eng 8:253–263
Ortega F, Lopez-Vizcon C (2012) Application of molecular marker-assisted selection (MAS) for disease
resistance in a practical potato breeding programme. Potato Res 55:1–13
Potato Genome Sequencing Consortium (2011) Genome sequence and analysis of the tuber crop potato.
Nature 475:189–195
Prashar A, Yildiz J, McNicol JW, Bryan GJ, Jones HG (2013) Infra-red thermography for high throughput
field phenotyping in Solanum tuberosum. PLoS One 8(6):e65816
Rommens CM, Ye J, Richael C, Swords K (2006) Improving potato storage and processing characteristics
through all-native DNA transformation. J Agric Food Chem 54:9882–9887
Rommens C M, Yan H, Swords K, Richael C, Ye J (2008) Low-acrylamide French fries and potato chips.
Plant Biotechnol J 6(8):843-853
Sharma R, Bhardwaj V, Dalamu D, Kaushik SK, Singh BP, Sharma S, Umamaheshwari R, Baswaraj R,
Kumar V, Gebhardt C (2014) Identification of elite potato genotypes possessing multiple disease
resistance genes through molecular approaches. Sci Hortic 179:204–211
Simmonds NW (1996) Family selection in plant breeding. Euphytica 90:201–208
Slater AT, Cogan NOI, Forster JW (2013) Cost analysis of the application of marker-assisted selection in
potato breeding. Mol Breed 32:299–310
Slater AT, Cogan NOI, Hayes BJ, Schultz L, Dale MFB, Bryan GJ, Forster JW (2014) Improving breeding
efficiency in potato using molecular and quantitative genetics. Theor Appl Genet 127:2279–2292
Slater AT, Cogan NOI, Forster JW, Hayes BJ, Daetwyler HD (2016) Improving genetic gain with genomic
selection in autotetraploid potato. Plant Genome 9. doi:10.3835/plantgenome2016.02.0021
Sun K, Wolters A-MA, Loonen AEHM, Huibers RP, van der Vlugt R, Goverse A, Jacobsen E, Visser RGF,
Bai Y (2016) Down-regulation of Arabidopsis DND1 orthologs in potato and tomato leads to broad-
spectrum resistance to late blight and powdery mildew. Transgenic Res 25:123–138. doi:10.1007/s11248-
015-9921-5
Tiemens-Hulscher M, Lammerts van Bueren ET, Hutten RCB (2012) Potato: perspectives to breed for an
organic crop ideotype. In: Lammerts van Bueren ET, Myers JR (eds) Organic crop breeding. Wiley-
Blackwell, Chichester, pp 227–237
Potato Research (2017) 60:171–193 193

Toxopeus HJ (1953) On the significance of multiplex parental material in breeding for resistance to some
diseases in the potato. Euphytica 2:139–146
Uitdewilligen JGAML, Wolters A-MA, D’hoop BB, Borm TJA, Visser RGF, van Eck HJ (2013) A next-
generation sequencing method for genotyping-by-sequencing of highly heterozygous autotetraploid
potato. PLoS One 8(5):e62355. doi:10.1371/journal.pone.0062355
Van Harsselaar JK, Lorenz J, Senning M, Sonnewald U, Sonnewald S (2017) Genome-wide analysis of starch
metabolism genes in potato (Solanum tuberosum L.) BMC Genomics 18:37. doi:10.1186/s12864-016-
3381-z
Viana AP, Vilela de Resende MD, Riaz S, Walker MA (2016) Genome selection in fruit breeding: application
to table grapes. Sci agric 73. doi:10.1590/0103-9016-2014-0323
Vos PG, Uitdewilligen JGAML, Voorrips RE, Visser RGF, van Eck HJ (2015) Development and analysis of a
20K SNP array for potato (Solanum tuberosum): an insight into the breeding history. Theor and Appl
Genet 128:2387–2401
Zimnoch-Guzowska E, Yin Z, Chrzanowska M, Flis B (2013) Sources and effectiveness of potato PVY
resistance in IHAR’s breeding research. Am J Potato Res 90:21–27. doi:10.1007/s12230-012-9289-5