Once a library of genetic variants has been generated, the genes must be introduced
into an organism so that they can be expressed and the activity of the variants
estimated. Ideally, variants should be evaluated in the final desired host because
heterologous expression can result in changes in measured activity. For instance, if
proteinaceous cofactors are necessary for function, their interactions can be optimized
leading to improved activity. Other factors that can contribute to improved
activity include improvements in mRNA and protein stability, reduction in protease sensitivity, optimization for host temperature pH or osmotic conditions,
interaction with available chaperone proteins, etc. However, while direct expression
and evaluation of variants in plants is desirable, it should be recognized that
such experiments are inherently problematic. First, plant generation times are
upwards of several months, making experimental cycles long if stable expression
is to be employed. However, it may be possible to reduce this time for seed
phenotypes using a fluorescence-based screen (Stuitje et al.
, 2003). Second, and
perhaps more problematic, insertion of a gene encoding particular activity into
the plant genome via Agrobacterium-mediated transformation, yields a wide spectrum
of expression levels, and consequently, enzyme activity depending on the
integration site of the T-DNA (Nowak et al.
, 2001). This is particularly problematic
for identifying variants with improved activities because it is difficult to determine
whether changes in activity are the result of changes in the enzyme or alterations in
expression between independent transformed plants. If the screen is for qualitative
differences, such as the occurrence of a novel product, this problem may not be
prohibitive. Transient expression in systems such as tobacco suspension cultures or
soybean embryos may offer a partial solution to this problem (Cahoon et al.
Whether whole plant or transient expression system is employed, a major problem
is attaining sufficiently high numbers of transformants to provide a reasonable
probability of identifying a substantially improved activity. Typically, directed
evolution experiments require the generation of 104–105 per cycle of improvement.
On the other hand, microbial systems offer generation times in hours to days
(rather than months for whole plants), and it is relatively straightforward to
produce sufficiently large numbers of transformants for analysis. However, in
heterologous expression, often improvements in performance can be attributed
to improvements in codon usage specific for the heterologous host. Such changes,
while they improve the property being measured in the heterologous host, do
not translate into improvements when expressed in the desired host; indeed
mutations to improve expression of a plant gene in Escherichia coli would likely
result in decreased expression when the ‘‘improved’’ gene is reintroduced into
plants. This example underscores the genetic maxim that ‘‘you always get what
you select for’’ and reinforces the notion that creating a screen that achieves the
goals of any particular project without producing unwanted results is one of the
biggest challenges facing protein engineers.
In summary, the best screens are conducted in the desired host; however, one
must weigh the constraints of time and transferability when designing a strategy for
improving a particular enzyme. A useful compromise for assessing plant enzymes
and variants is heterologous expression in yeast (Broadwater et al.
, 2002; Covello and
Reed, 1996). Being a single-celled eukaryotic system, it has the short generation times
of microbes along with the subcellular organization of eukaryotes.