Technical transfer and commercialisation of lyophilised biopharmaceuticals — application of lyophiliser characterisation and comparability – AAPS Open

The following sections presents the acquired K_v and equipment limitation data across the three commercial lyophilisers and the Lyostar II.

Figure 4A provides the overall mean K_v results for the centre location population of vials commercial scale data. It is the average of the three shelves assessed. K_v increases non-linearly with increased P_c. The non-linear relationship has been described previously, and the behaviour observed in the study is consistent with previous published data T (Kawasaki et al. 2019; Hibler et al. 2012). K_v was measured at SP2 for lyophiliser 03 only. The data shows an equivalent K_v profile for lyophilisers 01 and 02. The Lyostar II data demonstrates equivalency at 133 µBar (SP2), however exhibits a higher K_v below 133 µBar and lower K_v above 133 µBar when compared with the commercial units. Therefore, at the specific P_c set point of 133 µBar (100 mT), K_v measured was deemed to be equivalent across all our lyophilisers evaluated. Differences in profile curves are likely a function of lyophiliser design and the relevant equipment specific dependency on conductive and radiative heat input as convection changes with change in P_c. A similar trend has been shared previously by Tchessalov who compared K_v profiles as a function of P_c across multiple lyophilisers (Tchessalov 2016).

K_v comparability data for commercial lyophilisers (Lyo) 01, 02, and 03 and laboratory scale (LS) lyophiliser. N = 18 data points for commercial scale lyophiliser (with the exception of SP1 Lyo 01), N = 6 data points for laboratory scale lyophiliser. A Centre vials K_v. B Comparison of centre vials K_v across shelves for SP2 (133 μBar)

Figure 4B shows the K_v data by shelf for SP2. The graph illustrates not only the equivalence between the four lyophilisers but also the consistency within each individual lyophiliser for K_v from the top, middle and bottom shelves.

As outlined earlier in the manuscript and captured by Eq. 1, K_v is directly proportional to T_p. Once K_v is established, coupled with other inputs such as R_p and lyophiliser recipe set points, the T_p profile can be predicted. Based on the K_v values at 133 µBar, with an equivalent R_p, it would be expected that the resulting T_p would be equivalent for a given formulation across all four lyophilisers for main centre vial area. Where the primary drying P_c set point is set above or below 133 µBar, then T_s or P_c may need to be adjusted at commercial scale to generate an equivalent T_p profile in a given lyophiliser depending on K_v.

Table 6 provides the a, b and c coefficients as per the non-linear regression fitting formula provided by Eq. 3 for lyophilisers 01, 02 and the Lyostar II. Results are not displayed for lyophiliser 03 as only SP2 was conducted on this equipment. These coefficients can be utilised in primary drying prediction tools such as the one provided by SP scientific (Scientific 2016) or any other customised heat-mass transfer model.

Shelf temperature input for K_v calculation

Supplementary T_s data from the shelf fluid inlet, outlet and the shelf surface at the outlet and data collected from RTDs embedded in brass pucks within the Ellab shuttle located in the vial pack sitting in contact with the shelf surface is shown on Fig. 5. Figure 5A captures T_s data from lyophiliser 03, SP2. As described in the “Materials and methods” section, T_s surface at the inlet measured by RTD was used to calculate K_v. T_s surface at the inlet was chosen as it accounts for not only the shelf fluid but also the shelf stainless steel construction.

A Example of K_v lyo cycle trace for shelf 4, Lyo 03 at SP2 (133 µBar). B K_v for lyophilisers (Lyo) 02 and 03 calculated using various T_s inputs (T_s inlet, T_s inlet surface, T_s outlet and T_s surface measured using the brass puck)

For the K_v recipes provided in Table 5, the T_s set point was 0 °C. The T_s surface measured at the outlet was typically around 1 °C lower than at the inlet. This may be explained as due to heat energy flowing from the thermal fluid to the product to facilitate sublimation resulting in a colder outlet temperature. This was not so evident in the shelf fluid inlet and outlet data in this case as these readings were recorded from the main thermal fluid path manifold, and only 3 of the 13 shelves contained product.

T_s surface data collected from the brass pucks within the vial pack was 10 to 15 °C lower than the T_s set point. This may be attributed to the sublimation cooling impact of the vial pack impacting the shelf surface. Also, with this apparatus, the puck temperature is likely an average of the shelf temperature the bottom surface is resting on and the vapour temperature impacting the upper face that is not isolated. For this reason, we recommend using the T_s surface inlet for K_v calculations until further suitability of data collection via the brass puck method is established.

Figure 5B shows K_v data calculated using various T_s measurement inputs for lyophilisers 02 and 03. The data shows that whether using the T_s thermal fluid data or T_s surface data collected by the RTDs, the resulting K_v calculations are comparable. However, T_s surface recorded from the brass puck, which is arguably located more appropriately in proximity with the vials, results in double the K_v values following the calculation. Previously published K_v data has been calculated using the T_s inlet and is consistent with our data shown in Fig. 5 calculated using the T_s surface inlet (Tchessalov et al. 2021).

K_v edge vs centre

A distribution of K_v across the shelves was generated by normalising the weight loss recorded for vials at the edge locations against the mean weight loss from the central location. It has been shown in the literature (Pikal et al. 2016; Rambhatla 2003) the major factor for a higher K_v at the edge relative to the centre is the radiative heat due to proximity with product chamber walls. Materials of construction are also a consideration,for all laboratory and commercial scale equipment used in this study, stainless steel walls and doors were used.

Figure 6 shows K_v heat maps generated for the Lyostar II (A), lyophiliser 02 (B) and lyophiliser 03 (C). The edge effect is more prominent in the commercial units in comparison with the Lyostar II. K_v at the edge for lyophilisers 02 and 03 is up to 2 times that of the centre but only up to 1.5 times in the Lyostar II. For the commercial units at 3 vials deep from the edge, K_v is more consistent with the centre vial (location C). Therefore, K_v data calculated for location C is representative of approximately 87% of the 2322 vials on each shelf. The Lyostar II data shows K_v is equivalent to that of the centre location at rank 2 vials deep. These differences in K_v raise questions regarding the representative nature of edge vials at laboratory scale and how they represent the edge effect at commercial scale. It should be noted that the vial pack was surrounded by a metal ring that sits in contact with the shelf surface in the Lyostar II. Pisano et al. discussed how metal bands provide an additional contribution to heat transfer via conduction due to contact with the edge vial while also shielding radiative heat input from the chamber walls, thus reducing the heat input to edge vials (Pikal et al. 2016). There is still a radiative contribution by the metal band,however, it is limited as the surface temperature is low (Pisano et al. 2013). In comparison, during lyophilisation in the Lyomax system, the vials are not shielded from radiation which may explain why the edge effect is more prominent in the commercial scale units. For lyophiliser 01, the distance from the side edges of the shelf to the side walls is approximately 400 mm,the distance from the front and back edges to the adjacent wall/door is approximately 85 mm. For lyophilisers 02 and 03, the distance from the side edges of the shelf to the side walls is also approximately 400 mm; however, the distance from the front and back edges to the adjacent wall/door is approximately 170 mm.

Example of the K_v distribution outlined as heat map for commercial scale lyophiliser 02, shelf 07, lyophiliser 03 shelf 04 and the laboratory scale lyophiliser, middle shelf at SP2 (133 μBar)

Figure 7 provides a summary of the average edge vs centre K_v across SP1–SP3 (A) and a breakdown of the normalised edge vs centre factor per shelf (B). The figure was generated using values from the outer row of vials. The data provides further evidence showing consistent higher edge effect impact at commercial scale when compared with the Lyostar II, particularly regarding lyophiliser 01. Figure 7B provides some further insight; shelf 1 of lyophiliser 01 shows the highest edge effect normalisation factor, where shelf 7 and 13 are comparable with that of lyophilisers 02 and 03. The closure proximity of the front and back shelf edges of lyophiliser 01 with the adjacent surfaces does not appear to impact the weight loss in these local areas when compared with lyophilisers 02 and 03. Further assessment of the geometry specifically the ceiling area of lyophiliser 01 is required to further understand this identified hotspot, and the elevated radiative contribution needs to be accounted for in process robustness. In each case, for all 3 commercial units’, shelf 1 (top) presented the highest degree of edge effect, this is likely due to increased radiation as a result to proximity to not only the chamber walls but also the ceiling.

A K_v difference between centre and edge vials and B normalised edge effect factor for K_v for lyophilisers (Lyo) 01, 02 and 03 and the Lyostar II (LS) lyophiliser. Normalised weight loss data was used to calculate representative K_v at the edge locations

Other areas of consideration include the presence of sight glasses in the equipment walls and the contribution of radiative heat input from the Ellab shuttles. The Lyostar II has an integrated sight glass door. The commercial lyophilisers have an integrated sight glass adjacent to shelf 13 in the back engineering side door as well as both side walls. There was no evidence of additional contribution of radiative heat at these locations; in these cases, the design of the site glasses includes a peak extending over the top of the glass minimising direct light entry to the chamber. Furthermore, as discussed previously for the Lyostar II, vials in this vicinity are shielded by a metal band. There were limited preweighed vials placed adjacent to site glasses in the commercial units; thus, there is opportunity to investigate further. Vials in direct contact with Ellab shuttles showed consistent weight loss with that of edge vials as detected during Lyostar II studies specifically with weight loss data associated with Fig. 5A. Taking this into account, weight loss from vials in contact with Ellab shuttles was not included for over K_v calculations for the Lyostar II. During commercial scale studies, Ellab shuttles were placed strategically away from the seeded vials measured for weight loss to mitigate any impact to the resulting data.

It has been reported that radiative heat transfer is independent of pressure (Kuu et al. 2005). During these experiments however, a gradient was observed for the commercial scale equipment where the edge effect was more pronounced at lower P_c. This may be explained because at lower pressures, the contribution of radiation to the K_v is more prominent at the edge where overall K_v across the vial pack becomes less dependent on convection. This observation is consistent for Lyomax lyophiliser data shared by Tchessalov (2016) who showed a more prominent edge effect for a Lyomax 6 at P_c of 30 mTorr vs 500 mTorr. In contrast however, the Lyostar II is in more agreement with literature and exhibits a consistent edge effect factor as a function of pressure. This may be due to the inclusion of the metal band around the vial pack resulting in an edge effect less dependent on pressure due to the reduction in radiative heat input. Overall, the data shows the requirements for consideration in scale-up, if a reduction in the P_c set point is used to achieve a lower T_p, but this might increase variability from edge to centre at commercial scale. Alternatively, lowering the T_s set point may be more beneficial to minimise the risk of exceeding a critical product temperature.

The impact of radiation at edge locations has been well explained for primary drying (Pikal et al. 2016; Rambhatla 2003). In our experience at locations that exhibit a hotspot such as shelf 1 of lyophiliser 01, T_p profiles measured during scale-up and technical transfer are impacted. Data collected from a scale-up technical batch for a sucrose-based formulation in the 20 mL SCHOTT vial used for these characterisation studies provided T_p data at the edge of shelf 1 location A and the centre location C for a Lyomax 29. In this technical batch, T_p measured data provided evidence of a hotspot at a similar location to that observed during the K_v assessment of lyophiliser 01. T_p profiles measured using Ellab RTDs on shelf 1 for location A (n = 2) and for location C (n = 3) demonstrated worst-case edge vs centre T_p profile from a Lyomax 29. During freezing, T_p at location A trended approximately 4 °C higher than T_p at location C. During primary drying, T_p trended approximately 2 °C higher at location A when compared with location C, and the duration of sublimation was approximately 15% shorter based on T_p equilibrating with T_s. During secondary drying, T_p trends about 3 °C lower at location A than location C (data available in the associated supplementary material).

The edge T_p characteristics observed during freezing and secondary drying during this technical batch may be attributed also to the chamber wall which in this case is not temperature controlled. Figure 8 shows data from lyophiliser 02 SP3 (200 µbar) where RTDs were fixed to the chamber wall adjacent to shelf 10 (empty shelf) and shelf 13 (bottom shelf containing vials of water). During freezing, with T_s at − 40 °C, the wall surface temperature decreases gradually from ambient to between 0 and − 5 °C. During primary drying with T_s at 0 °C and T_p <  − 30 °C, the wall surface temperature gradually increases to approximately 5 °C during 5 h of sublimation. This information provides rationale for the edge T_p behaviours described above, as the chamber wall surface temperature is higher than the T_s during freezing and lower than T_s transitioning from primary drying into secondary drying. This may directly impact T_p at the edge.

K_v lyocycle trace for lyophiliser 02 SP3 showing chamber wall surface temperature adjacent to shelf 10 and shelf 13

As outlined previously, lyophiliser 03 is a single-storey configuration with the condenser chute opening adjacent to shelf 4. This is an area of interest as edge vials adjacent to the chute opening are not proximal to the chamber wall surface. This location may also have a localised vacuum pressure lower than anywhere else in the product chamber during primary drying due to proximity with the condenser (Kshirsagar et al. 2019). It was found that vials in this location exhibited comparable edge effect characteristics to other edge locations assessed (results not shown).

In summary, the data demonstrates a more significant edge effect at commercial scale where edge vials are not shielded from radiation in comparison with the Lyostar II where radiation is shielded by the steel ring surrounding the vials. K_v and T_p profiles of edge locations versus centre demonstrate the necessity for a robust formulation development and consideration during scale-up.

Impact of shelf inter-distance on K_v

Lyophiliser 01 has a shelf inter-distance of 100 mm. Lyophilisers 02 and 03 have a shelf inter-distance of 110 mm. Data in this study suggests shelf inter-distance does not impact K_v. This is in agreement with work performed by Ganguly et al. who proposed using CFD modelling that at an inter-shelf distance of 90 mm, there is a nearly uniform distribution in pressure (Ganguly et al. 2017).

Impact of load on K_v

As outlined in “Materials and methods”, for commercial scale equipment, 3 of the 13 shelves were utilised for this study. The study design was considered to facilitate an assessment of the top, middle and bottom of the shelf stack under a limited commercial equipment capacity and primary packaging component availability. There are in this case some limitations to consider with respect to the data’s representation of fully loaded lyophiliser production cycles.

It has been demonstrated that a full capacity load produces a lower K_v and thus lower T_p during sublimation with a longer primary drying duration (Patel et al. 2010). The calculated K_v in this study could be higher than what is expected under full load conditions as a function of the thermal impact of a larger number of vials subject to sublimation cooling in the product chamber. On the contrary, the calculated K_v in this study could also be lower than what is expected under full load conditions as a function of the gas composition in the product chamber. Under maximum load conditions during sublimation, there is a larger fraction of water vapour making up the gas composition in the product chamber. This reduces the nitrogen supply required to maintain vacuum pressure set-point control. During these experiments however, there was likely a larger fraction of nitrogen in the overall gas composition due to the lower load in the product chamber. Water vapour has a molecular conductivity about 60% higher than that of nitrogen (Nail et al. 2017). Considering an identical experimental approach was taken for each commercial scale lyophiliser, there is minimal impact to the study’s comparability element. A confirmation K_v study under maximum load conditions would further verify the data (Barresi and Marchisio 2018).

Sublimation studies

Figure 9 shows the MCP trend from lyophiliser 02 as an example; trend data from lyophiliser 03 is not shown. As described in the method “Part 1: Minimal controllable pressure”, this data was generated as per part 1 of the study where the outputs were used to assess the equipment choke flow regime and generate the equipment limitation boundary of maximum sublimation rate vs P_c.

MCP Lyophilisation trace for sublimation study. Part 1 on lyophiliser 02

The graph illustrates how the incremental increases in T_s influence not only the vacuum pressure but also the ice temperature, condenser coil temperature and T_s surface. The delta between the T_s shelf fluid inlet and the T_s surface measured by RTD increases as the degree of sublimation and the endothermic cooling increases showing the impact of ice under sublimation on the T_s surface.

Figure 10 shows an overlay of MCP vs sublimation rate for lyophilisers 02 (horizontal configuration) 03 (vertical configuration) and the Lyostar II.

Equipment limitation data of the LyoStar II (LS) and commercial scale lyophilisers (Lyo) 02 and 03 in comparison with literature TDLAS data for Lyostar II by Mockus et al. Part 1 — refers to minimum controllable pressure cycles. Part 2 — refers to the gravimetric verification cycles executed considering part 1 results

The data shows the equipment capability boundaries of lyophilisers 02 and 03 and the LyosSar II, inside which the operational space should be defined. Outside this boundary, P_c control would be lost resulting in choked flow (Patel et al. 2010). An inflection point in the Lyostar II data represents the point at which the condenser is overloaded causing the unit to go into a “safe mode”. At this point, the condenser was no longer able to trap vapour which risks moisture entering the vacuum pump. During the commercial scale studies, this was avoided by monitoring the condenser coil temperature to ensure there was no risk to damaging commercial equipment.

Lyophiliser 02 offers the highest water vapour mass transfer capacity, whereas the Lyostar II unit is the most limited piece of equipment. Literature articles often describe scale-up scenarios where commercial equipment is not capable of facilitating lyophilisation recipes developed at laboratory scale due to choke limitations (Pisano et al. 2013). In this example however, the most aggressive recipe developed on the Lyostar II will not pose a risk of choke at commercial scale. Also included in Fig. 10 for comparative purposes is normalised Lyostar II data generated using TDLAS by Mockus et al. which provides further evidence of the higher capability of the Lyomax units when compared with a Lyostar model (Mockus et al. 2011).

The data also poses the question on condenser location impact to performance. The K_v assessment provided evidence of comparability between lyophiliser 02 (Lyomax 29 vertical configuration) and 03 (Lyomax 29 horizontal configuration). However, the equipment limitation assessment identified differences in performance. In contrast to Kshirsagar et al., data generated in this study showed a broader equipment boundary with respect to the vertical condenser configuration. The CFD model presented by Kshirsagar assumes the absence of the mushroom valve at the condenser chute, whereas the valve was intact during the studies described in this experiment. Further consideration for lyophilisers 02 and 03 with respect to the controlling capacitance manometer vacuum gauge is required. The controlling capacitance manometer is located in the ceiling of the product chamber; however, the distribution of P_c across the product chamber is likely to exhibit a gradient with proximity to the condenser which is more prominent in a vertical two-story configuration (Kshirsagar et al. 2019).

Another key variable to consider is the length and diameter of the condenser chute. Patel et al. demonstrated that gas velocity reaches the Mach I limit at the condenser chute exit under the choked flow conditions (Patel et al. 2010). The correlation between gas flow conductance and chamber pressure depends on the geometry of the chute where the dimensions are characterised by the ratio of chute length/diameter (L/D). For lyophiliser 02, the chute length is approximately 2.4 m with a diameter of about 0.7 m resulting in a L/D of 3.4. Lyophiliser 03 has chute length of approximately 1.7 m with a diameter of about 0.7 m resulting in a L/D of 2.4. Mach 1 and subsequent choked flow is observed in lyophiliser 03 at a higher P_c at a given T_s.