Antibodies with a sweet tooth.

One of the challenges in antibody manufacturing is the glycosylation pattern.  In a recent publication from the James lab in the UK, they demonstrated the how variations to the cell culture additives could be used to drive the glycosylation.

They generated two CHO cell lines expressing their antibody and then varied manganese, uridine, and galactose.  The viable cell density, titer, and galactosylation were measured.  They used a nice response surface approach to get some predictive models out of the data.  The data were presented in graphical format so I had to pull it out of the paper using "Paul Plot".  Fortunately, my results are close enough that I can show you the output and derive some useful knowledge in the process.

The IVCC appears to be only dependent upon the uridine concentration.
ANOVA Results for IVCC

Source	DF	Sum of Squares	Mean Square	F Ratio
Model	1	2.0070400	2.00704	60.5653
Error	13	0.4308000	0.03314	Prob > F
C. Total	14	2.4378400		<.0001*

The titer was driven by uridine, galactose and their pairwise interaction
ANOVA Results for Titer

Source	DF	Sum of Squares	Mean Square	F Ratio
Model	3	37278.925	12426.3	31.8069
Error	11	4297.475	390.7	Prob > F
C. Total	14	41576.400		<.0001*

Beta-1,4-Galactosylation was measured relative to a control and was found to be dependent upon the uridine, galactose, and square of the uridine.
ANOVA Results for Galactosylation

Source	DF	Sum of Squares	Mean Square	F Ratio
Model	3	2.1632500	0.721083	17.5972
Error	11	0.4507500	0.040977	Prob > F
C. Total	14	2.6140000		0.0002*

In the paper, Mn2+ was found to have a binary effect (on/off) but this didn't show up in my models (probably a function of "Paul Plot").

When the profilers are linked (see previous post), the plots point a path towards optimization of the media.  Notice how the confidence interval for the galactosylation increases near the inflection point of the plot?  That's probably an artifact of my data extraction program.  This does illustrate an important point - effect magnitude and its impact on the analysis.  As the magnitude of the change decreases relative to the input change, the predictive ability of the model is reduced.  The only way to improve that situation is with more data.

Profilers for IVCC, Titer, and Galactosylation

I also note that an increase in the titer comes at the expense of the galactosylation.  The result will require a close collaboration with the non-clinical folks to establish the role of galactosylation on the function of the antibody as well as providing a potential control strategy for the quality attribute and a discussion point related to manufacturability.  The last thing Management wants to deal with is a low-producing clone that has a difficult to control profile.  The results from a multivariate design can be used to advocate for a particular path forward for a project.

The authors complete the analysis with the second clone and I'll encourage folks to have a look for themselves at the data.  All in all,  the work presents a nice contribution to my lit library.