Preculture development
Assuming that the main culture shown in Fig. 2a represents a typical batch screening process that is terminated after 65 h, individual clones are harvested in completely different growth phases. One reason for unequal growth behavior in the main culture relates to the colony picking process. Colony picking causes a transfer of different colony volume and the picked colonies can be in unequal physiological stages [28]. These variations cannot be avoided by changing the picking method (by hand or by picking robot) or picking instrument (inoculation loop, toothpick, pipet tip, etc.) [29, 30]. However, although the colony picking method described in Fig. 2b was similar to the procedure in Fig. 2a, scattered light intensities (1–4) of the first preculture exhibit negligible lag-phases. The prevention of the lag-phases can be attributed to the applied culture medium. In Fig. 2a, colonies from complex LB agar plates were directly transferred into mineral salt medium of the main culture. The changing substrate availability from complex to mineral basis results in extended lag-phases and unpredictable growth kinetics. Besides the substrate source, also physical changes within the culture environment, as for example the medium osmolality, influence growth of B. licheniformis [31, 32]. By applying the sequential preculture procedure (Fig. 2b), colonies from complex LB agar plates were transferred into complex TB medium of the first preculture. Hence, the exemplarily shown cultivations using TB medium reveal efficient growth within the first preculture.
The positive effect on growth is not the only reason for using complex TB medium within the first preculture. Apart from the complex compounds, TB medium is supplemented with glycerol. In TB medium, glycerol represents the primary carbon source for B. licheniformis. The depletion of glycerol is indicated by a plateau of the scattered light intensity at approximately 12 h of cultivation (Fig. 2b, first preculture). At this point, the complex compounds act as slowly accessible substrate sources, which prevent B. licheniformis from undergoing carbon starvation. This results in equal growth conditions with equal biomass concentrations for each individually picked colony, which is indicated by the scattered light plateau (colony 1–4) reached after 16 h of cultivation (Fig. 2b, first preculture, Additional file 1). If B. licheniformis cultures face carbon starvation, the scattered light intensity exhibits a declining trend instead of showing a plateau (Fig. 2b, main culture; Additional files 1 and 2). Under carbon starvation, Bacillus cells exhibit cell lysis [33, 34] and morphological changes [35], resulting in decreasing scattered light intensities [36].
The reason for using a relatively small inoculum volume of 1% (v/v) to inoculate the second preculture was to minimize the transfer of unmetabolized complex compounds (Fig. 2b, second preculture). The short lag-phase is most probably caused by transferring the culture from complex to mineral medium. Diluting the complex compounds is not the only reason for introducing a second preculture. The second preculture also serves to adapt the B. licheniformis culture to the mineral salt medium, which is also applied in the main culture. However, when working with mineral salt media in the preculture, the time of inoculum transfer determines the inoculum quality. Since the mineral salt medium contains 20 g/L glucose as only carbon source, glucose depletion results in carbon starvation. Besides cell lysis [33, 34] and morphological changes [35], carbon starvation causes a variety of metabolic changes in Bacillus cultures [15, 37, 38]. To prevent the culture from undergoing these changes, the second preculture was harvested in the late exponential growth phase at 24 h (Fig. 2b, second preculture), prior to glucose depletion.
The batch main culture was inoculated with 10% (v/v) of the second preculture to ensure a high initial biomass concentration (Fig. 2b, main culture). This is of particular importance when the screening process is conducted with the fed-batch MTP. Due to the release principle of this microtiter plate (Fig. 1), glucose is released from the beginning of the cultivation (Additional file 3) [26, 27]. At this point, the biomass concentration is too low to consume all of the released glucose. This results in an initial batch phase. Its length is dependent on the initial biomass concentration [27]. Low initial biomass concentrations might cause an extended batch phase with oxygen limitation and overflow metabolism.
A commonly applied preculture procedure consists of a single preculture instead of a two-step procedure (Additional file 1). This single preculture can be performed either in complex medium or mineral salt medium. As described above, directly transferring picked colonies into mineral salt medium revealed substantial lag-phases with unequal growth (Fig. 2a), which does not result in synchronized cultures. Precultures performed in complex medium showed negligible lag-phases and equal growth, thereby enabling growth synchronization (Fig. 2b, Additional file 1). However, when performing a single preculture in complex medium, complex compounds are transferred into the main culture. At the same time, a short lag-phase was observed when transferring the culture from the complex medium to the mineral salt medium (Fig. 2b). Both, the transfer of complex compounds and lag-phases become crucial when working with fed-batch MTP’s. As mentioned above, the fed-batch MTP’s require high initial cell concentrations. Thus, with a single preculture it is difficult to find a trade-off between reaching high initial cell densities and reducing the amount of transferred complex compounds into the main culture. By introducing a second preculture with mineral salt medium, the complex compounds transferred to the main culture can be reduced, on the one hand, and the B. licheniformis culture can adapt to the mineral salt medium, on the other, which enables equal growth with negligible lag-phases in the main culture (Fig. 2b).
Unequal growth conditions can also be prevented with the use of a robotic platform (RoboLector®) consisting of a BioLector® and a liquid-handling robot [9, 39]. Applying fed-batch mode to synchronize growth of precultures is another functional and well-described procedure [28, 29]. However, with the used B. licheniformis strain, two sequential preculture cultivations in batch mode represent an easy to implement and efficient solution to synchronize the precultures.
Determination of the number of replicates
In every high-throughput screening there is a trade-off between the number of replicates used for each strain and the number of different strains investigated. While the reliability increases with more replicates, the throughput decreases at the same time. Determination of the minimal number of replicates needed for a statistically sound evaluation of strain performance can be used to find an optimum between statistical reliability and throughput. However, for calculating the minimal number of replicates the standard deviation, the minimal detectable difference and the statistical power is required.
The standard deviation of repetitive cultivations with fed-batch MTP’s is in the range of ± 10% (Fig. 4a). This standard deviation represents the sum of individual deviations occurring throughout the entire screening procedure. This includes the analytical procedure of the protease activity measurement. However, the largest deviation derives from the glucose-limited fed-batch cultivation with the fed-batch MTP. It has been found that glucose release from fed-batch MTP’s exhibits a mean coefficient of variation of 4.5% [26]. This deviation in glucose release directly influences the final protease activity.
The results in Fig. 4b show that besides the standard deviation, the detectable difference influences the number of replicates. The minimal detectable difference describes to which extend the protease activity must at least increase or decrease in order to be detected with the demanded statistical power. In contrast to the standard deviation, the minimal detectable difference is a parameter that cannot be determined experimentally. The value of the minimal detectable difference relies for example on previous experiences with strain improvement programs. Since the number of replicates decreases with increasing minimal detectable difference (Fig. 4b), a clone that has a change in protease activity > ± 15% is still securely detected with 6 replicates (statistical power ≥ 0.8). If only 3 replicates would be used, as often found in literature, only strains with a productivity increase > ± 30% compared to the control strain could be identified with a standard deviation of ± 10% and a statistical power of 0.8 (Additional file 4a). On the other hand, using only 3 replicates but selecting for strains with ± 15% improvement at ± 10% standard deviation would result in a statistical power of only 0.3 (Additional file 4b).
The statistical power describes the probability of correctly rejecting the null hypothesis, i.e. correctly identifying clones with different protease activities. Therefore, the statistical power is an important measure regarding the reliability of screening results. The higher the statistical power, the higher the number of replicates needed to correctly identify clones with a different protease activity (Additional file 4b). In order to find a trade-off between statistical reliability and number of replicates (Additional file 4b), a statistical power of 0.8 was chosen for this study.
Comparison of batch and fed-batch screening
The comparison of the predictability of batch (38%) and fed-batch screening mode (92%) highlights that implementing fed-batch conditions is essential to predict the clone performance similarly to the fed-batch fermenter that mimics production conditions. The necessity of implementing fed-batch conditions during screening is exemplarily discussed based on the results achieved with clone 4 and 5 from lineage #1 and clone 1 and 5 from lineage #2 (Fig. 5c). Clone 4 from lineage #1 exhibits a significantly higher protease activity under batch screening conditions (Fig. 5c, batch MTP). With the assumption that the screening process is only conducted in batch mode, clone 4 would be selected as a promising strain candidate and would be further investigated under lab-scale fed-batch conditions. Under these conditions, however, clone 4 shows no significantly higher protease activity and would be rejected (Fig. 5c, fermenter). Under fed-batch screening conditions, in contrast, clone 4 is correctly identified as a strain with no significantly higher protease activity (Fig. 5c, fed-batch MTP). Consequently, with a screening process conducted in fed-batch mode, the strain would not have been considered further, thereby saving time and costs. The results achieved with clone 5 from lineage #1 and clone 1 and 5 from lineage #2 represent an opposite scenario (Fig. 5c). Due to the similar (non-significant) or significantly lower performance achieved under batch screening conditions (Fig. 5c, batch MTP), these clones would directly be rejected. However, under fed-batch screening conditions, these clones were correctly identified as clones with a significantly higher protease activity than the control strain (Fig. 5c, fed-batch MTP and fermenter). This scenario shows that optimal protease producing clones can remain undetected under batch screening conditions, due to unequal physiological conditions between screening and production processes.