Typical situation slide (please comment if you have other use cases)
|
Industry
|
Biotech, Pharmaceuticals
|
|
Job Title
|
Upstream/Fermentation Scientist, Downstream Scientist
|
|
Department or Role
|
R+D
|
|
Industry
|
Pharmaceutical, Biotechnology
|
|
Critical Business Issue
|
Reduce level of impurities in downstream caused by upstream process settings that are impossible to remove
|
|
Specific Capabilities
|
Identify upstream conditions effecting impurity levels in downstream
|
|
Business Win
|
Improved product purity, faster time to market, improved quality, reduced cost
|
|
EBI/Business Metrics
|
Risk Mitigation, Product Pipeline, Cost Reduction, Time to Market
|
|
User Win
|
Better communication between teams, novel skillset, perception, promotability
|
|
Problems/Reasons
|
Upstream and Downstream are siloed, poor understanding of each others issues, upstream may target conditions for large productivity, rather than purity
|
Terms to look out for: Bioreactor, Chromatography, Upstream, Downstream, Purification, AAV, Proteins, Filtration, Isolation
Questions to ask: Do you ever have issues removing impurities from your products during development? Do you understand how the upstream process is affecting your separations downstream? Would you like to better understand the effect of your process steps on one another?
Context
The demo is based off the paper from Lonza at Discovery Europe 2024 – I’ve recreated the results here using the tool MultiDLC HPLC simulator and created a DoE around the height of peaks.
The paper focuses on the typical set-up of a Biotech/Pharma process, material is created upstream (by chemical reaction/fermentation) which then has to be isolated and purified by the downstream team. The challenge is that upstream processes typically optimise for high productivity levels, which can also lead to the creation of impurities in the product passed over to downstream. Some impurities can be removed by optimisation of the downstream separation process, but some may be too costly or impossible to remove. Impurities can lead to problems in efficacy, regulatory approval and customer perception of the final product, so it is important it is minimised.
The impurity levels can be monitored ‘small scale’ with a chromatogram of the downstream process (see below) – the less impurities, the better!
USP and DSP typically don’t communicate well between one another, so Lonza took an innovative approach to get the two groups to work together to understand if they could alter the upstream process conditions to reduce the impurities in the downstream process. This was performed with a DoE of the upstream process, where the output measured was the downstream functional chromatogram data.
This is a really nice case study from a real life situation from a big pharmaceutical organisation, which should help convince prospects of the value.
Last things first
I’ve created a dashboard of the final functional DoE result, and simplified the view to show only the chromatogram (left) and to show the inputs as a custom slider (right) to keep the focus on the value of the FDE.
We can go through and show key findings:
- Ammonium only affects the first peak (Product)
- Initial VCD seems to only affect the 3rd peak
- pH only affects the impurities, making it a good candidate to change.
We can take it a step further by optimising via a target (more details below in article) – this provides the conditions that will remove the impurities from the downstream process, avoiding additional processing steps, costly process optimisation or abandoning the product process.
The Process
The process is similar to that described in this HPLC demo. In brief:
- Create the DOE using Custom Design
- Run and execute the design and collect the time series data, creating a row that identifies the experiment number
- Join the data-tables with the ‘Join’ tool
- Create a target function (see below)
- Run a Wavelet model in FDE of time and Ct. with the upstream process settings as supplementary
- Optimize the process settings according to the target.
Extra – Target creation
To create the target, you can take an example data-set from the experiments and alter the values to create an ‘ideal’ peak profile.
Raw example
Post processing to create target
I changed the values of the time values where the impurities were to 0, then altered the spline fit to make the line fit to the points as above. Then you can ‘Save Formula’ for the spline, and bring the data back into the original data table as a Target. This is what they look like together:
