THE PRACTICAL IMPLEMENTATION OF NGG FOR ACCELERATING BIOPROCESS QBD
One such project that the authors have been involved with is a collaborative partnership between a major pharmaceutical company
and ArrayXpress, a contract genomics services company. While the specific organism and target compound are confidential, the
tools, techniques and processes utilized provide a great example for demonstrating the benefits of the NGG systems biology
approach. The primary objective of the project under discussion was to increase production titers of an essential target compound
used in the manufacturing process of a current large revenue generating commercial product. The secondary objective was to
build knowledge that will allow for faster and more efficient manufacturing of other products using the same organism/expression
platform. As with all bioprocesses, the organism itself is only a single variable influencing productivity, with many environmentally
tunable variables making up the remainder. We have CPPs that influence production, but due to prior technology limitations,
the manufacturing engineers did not know their full impact on the metabolic processes and production efficiencies of the cells.
Therefore, these parameters had previously simply been lumped together as an unknown called "process variability" or "biological
variability", and as such, their manufacturing was completely at the mercy of the process itself, with limited process stability
and dramatic product titer variability.
By bringing together the cells and the CPPs in a systems model, we can now see the entire equation. The cells are the primary
production machinery; therefore our approach was to evaluate the physiological condition and the state of the cells during
the various media and fermentation development stages. We first generated a working hypotheses by developing a fishbone diagram
that showed all the confirmed and putative CPPs associated with the overall target compound production process (see Figure
1). This allowed for the identification of critical areas to be characterized in more detail, which was subsequently experimentally
Figure 1: Fishbone diagram showing all the confirmed and putative CPPs associated with the overall target compound production
Our approach was to design highly focused and statistically sound microarray experiments with complementing standard analytical
chemistry tests. We wish to emphasize the importance of having a very well thought out experimental design and analysis strategy
prior to project initiation. This approach made it possible to identify key genes and their associated molecular pathways
that were differentially affected due to changes of various CPPs in the overall production process. The use of DNA microarrays
provides a detailed qualitative snapshot of the state of the transcriptome at the time of sampling, somewhat like a molecular
fingerprint, that can reveal subtle process variations in great detail. This approach is especially useful in time course
experiments like the ones we faced, to determine whole transcriptome changes associated with different CPPs, monitored across
different growth phases (different time points) of the cells during the media and fermentation optimization stages.
Strong bioinformatics, both in statistical design and data analysis and mining, are the next key to success. A particularly
important aspect of statistical inference in high throughput problems, such as microarray experiments, is the assessment of
statistical significance exhibited by the data in the presence of a tremendous multiplicity of hypotheses. A single experiment
can involve tens of thousands of hypothesis tests. This assessment requires efficient estimation of experimental error and
careful control of false discovery rates. We applied two interconnected analysis–of–variance models: A normalization model
that accounts for experiment–wide systematic effects that could bias inferences made from the data on individual genes, and
a gene model that is fit to the normalized data from each gene, allowing inferences to be made using separate estimates of
variability. Expression differences are then parameterized as factorial effects in linear mixed effects models appropriate
to the experimental design. These effects can be estimated efficiently using statistical softwaresuch as JMP Genomics or
SAS PROC MIXED. Resulting least square estimates are then mapped onto their associated metabolic pathways using KEGG metabolic
pathway maps (www.genome.jp/kegg/pathway.html) in combination with proprietary software mapping tools (3).
The ability to map differentially expressed genes onto their associated biochemical pathways provides the opportunity to "zoom
in" on each of the metabolic pathways associated with protein production. Key metabolites that are either depleted or produced
are relatively easy to identify, but true process understanding comes from identifying how the compounds are used in the metabolic
machinery. Amino acids, for example, could be depleted by translation, interconversion to other amino acids, or detoxification
by the cell. Each of these routes has dramatically different impacts on cell health and productivity. With the application
of NGG techniques you do not have to wait until the end of the project to begin seeing results. Each individual experiment
contributes to the "systems" knowledge but in the short run provides specific information on variables that can be tuned for
performance. Over the past three years we have completed numerous microarray experiments as part of our primary media and
fermentation optimization objectives. A few examples will be highlighted here that will demonstrate the power of microarray
technology to improve bioprocess stability and production yields as part of a larger NGG initiative.
In the manufacturing process of the target compound of interest, the original growth medium components were not well defined.
As a result, different medium lots varied dramatically in protein yield and product titers. One of the primary objectives
was to develop a chemically defined medium that would yield consistent titers. In our experiments, we evaluated whether stress
response mechanisms of the production cells caused a reduction in titer during phase transition, and how media and fermentation
conditions impacted these stress responses. We carefully designed time course experiments to cover transition through growth
phases with trial versions of different defined media. Complimentary to this, we completed analytical chemistry tests to assign
putative roles to transcription regulators that might be involved in stress response. By ultimately correlating differentially
expressed genes of sigma factors with their associated biochemical pathways, we were able to optimize and change certain media
components that led to improved protein production.