The ideal membrane would retain 100% GCMP with R = 1 and the theoretical yield would be 100%. If the UF membrane allowed GCMP to leak through, then the yield would drop drastically.
The expected recovery can be calculated from the above equation using a hypothetical retention coefficient (Table 1). For
example, a 10% leak of GCMP into the permeate would lower the recovery to 37%. Based on this analysis, it seemed likely that
the 100 kDa UF membrane retention coefficient had changed; causing the low yield in the production lots 6–12.
Table 1. Retention coefficient and recovery of desired component
The retention behavior of a UF membrane can be characterized by its R90 value. R90 is not included in the certificate of analysis
provided by the vendor for each lot of UF membrane but can be obtained by request. For 100 kDa NMWCO membranes, the vendor
reported an R90 range of 160–485 kDa.
R90 was determined by the vendor as part of the quality control procedure using the method described by Tkacik and Michaels.10 A mixture of dextran fractions was allowed to equilibrate in a UF system. An HPLC sizing column was used to measure the
concentration of dextran in the permeate and retentate. In a plot of the rejection coefficient versus the MW of the dextran,
R90 is the MW of that dextran with 90% retention. A higher R90 implies a membrane with a higher porosity. UF membranes are
produced in large rolls, which are cut and assembled into UF cassettes. When new membrane rolls are produced, R90 is tested
in two locations on the roll and the average is reported. When cassettes are produced, the R90 reported is the R90 of the
membrane rolls used to make the cassette.
Table 2. R90, yield, and reuse for ultrafiltration membranes used in development and production
After the R90 values obtained from the vendor for the membrane lots used in the NeisVac-C development were compared to those
used in early production lots, the reason for the yield problem became clear (Table 2). From development through the 5th production
lot, low value R90 membranes were used. Starting with production lot 6, the membranes used had much higher R90 and these lots
of membrane could no longer retain GCMP completely and a low yield was observed. The membrane reuse also was tracked. There
was no significant impact of membrane reuse on the yield.
The corrective action from the investigation was to change from a 100 kDa NMWCO membrane to 50 kDa NMWCO membrane for this
UF processing step. The vendor reported R90 values of 95–160 kDa for a 50 kDa UF membrane. With the change, GCMP yield returned
to levels seen during development (consistently close to 100%) but purity had to be evaluated.
To evaluate purity, the production process was scaled down 400 fold. A representative process stream was spiked with impurities
(autoclaved whole cells), treated with NaOH, and then the tightest 50 kDa membrane obtainable (R90 of 93 kDa) was used to
diafilter GCMP. As shown in Table 3, the removal of impurities was satisfactory at the small scale using typical starting
material and spiked starting material. For the spiked starting material, the protein level dropped from 260% to <1% and the
nucleic acid level dropped from 226% to 3% after NaOH treatment and 50 kDa diafiltration.
Table 3. Small-scale purity data before and after NaOH treatment and filtration with a 50 kDa membrane