Processing heterogeneous biomass

Abstract

Plants have been developed as an alternative platform for the production of biopharmaceutical proteins, culminating recently with the FDA approval of the first plant-derived recombinant pharmaceutical enzyme for human use (ELELYSOÔ by Protalix Biotherapeutics). Among the many different plant-based technologies that have been proposed, transient expression mediated by Agrobacterium tumefaciens has proven to be particularly suitable for the rapid production of vaccines in response to emerging pandemics. However, one potential drawback of transient expression in whole plants is the large variation in recombinant protein expression levels among different leaves, which introduces a level of uncertainty in process design that can increase the regulatory burden and production costs. Transient expression is also used to test expression constructs prior to the longer and more expensive process of generating transgenic plants, and here the variation can produce misleading results leading to erroneous conclusions about the relative activity of different promoters and other regulatory elements. Such variation can be caused by loosely controlled environmental and process factors such incubation temperature, plant characteristics and the method and timing of harvesting. Here we discuss differences between transgenic plants and transient expression in intact plants, and their specific pitfalls for model building. We also highlight which aspects researchers should consider when using a DoE approach to investigate protein expression in plants, both for fundamental research and process development.

Plants have been developed as an alternative platform for the production of biopharmaceutical proteins, culminating recently with the FDA approval of the first plant-derived recombinant pharmaceutical enzyme for human use (ELELYSO^™ by Protalix Biotherapeutics). Among the many different plant-based technologies that have been proposed, transient expression mediated by Agrobacterium tumefaciens has proven to be particularly suitable for the rapid production of vaccines in response to emerging pandemics.¹ However, one potential drawback of transient expression in whole plants is the large variation in recombinant protein expression levels among different leaves, which introduces a level of uncertainty in process design that can increase the regulatory burden and production costs.^2-4 Transient expression is also used to test expression constructs prior to the longer and more expensive process of generating transgenic plants, and here the variation can produce misleading results leading to erroneous conclusions about the relative activity of different promoters and other regulatory elements.⁵ Such variation can be caused by loosely controlled environmental and process factors such incubation temperature, plant characteristics and the method and timing of harvesting.

We have recently quantified the effect of several of these factors by compiling the data in a design of experiments (DoE) approach, thus building a model that can predict the amount of recombinant protein produced under different process conditions.⁴ The model was also used to evaluate the economic impact of different process conditions—and the resulting protein concentrations—on the overall production costs for a plant-derived monoclonal antibody. Here we discuss differences between transgenic plants and transient expression in intact plants, and their specific pitfalls for model building. We also highlight which aspects researchers should consider when using a DoE approach to investigate protein expression in plants, both for fundamental research and process development.

A detailed description of the methods can be found in our earlier manuscript.⁴ Briefly, Agrobacterium tumefaciens strain GV3101:pMP90RK was transformed with plant expression vector pGFD, a derivative of pPAM (GenBank AY027531) allowing transient expression of the fluorescent protein DsRed (GenBank AF168419; R2G mutant) and monoclonal antibody 2G12 (kappa light chain F62 version and gamma heavy chain) under the control of the Cauliflower mosaic virus (CaMV) 35S promoter.⁶ Bacteria were diluted with water and 2× infiltration medium to achieve the necessary OD_600nm value and were injected into tobacco leaves using a 1-mL syringe. The tobacco plants (Nicotiana tabacum cv Petit Havana SR1) were cultivated in a greenhouse at 25/22°C day/night temperature with a 16 h photoperiod (180 µmol s⁻¹ m⁻²; λ = 400–700 nm) and at 70% relative humidity for either 35 or 42 d prior to infiltration. Infiltrated plants were incubated in phytotrons for 5 d at 70% relative humidity with a 16 h photoperiod (75 µmol s⁻¹ m⁻²; λ = 400–700 nm) at 15, 17, 20, 22, 25, 28 or 30°C. Leaf samples were taken 40 or 47 d post seeding (dps) corresponding to the growing and mature bud stages, respectively. Proteins were extracted by grinding samples in 3 µL extraction buffer (50 mM sodium phosphate, 500 mM NaCl; pH 8.0) per 1 mg of fresh biomass and solids were removed by centrifugation at 16,000 × g. DsRed fluorescence in the supernatant was determined using a plate reader fitted with 530/25 (excitation) and 590/35 (emission) filter set, and a standard curve was generated with dilutions in the range 0–225 µg/mL. The quantity of 2G12 antibody was determined by surface plasmon resonance spectroscopy using a chip coated with staphylococcal protein A, and by comparison to a 585 ng/mL reference solution. Design Expert 8.0 (Stat-Ease) was used to build and evaluate IV-optimal response surface designs with categorical balance. We modeled DsRed and 2G12 concentrations in 142 leaf extracts harvested at 40 dps and 465 leaf extracts harvested at 47 dps. Factors showing a significant influence on protein expression were pre-selected from a fourth-order model by automatic backward selection using a p-value threshold of 0.100, and factors with p-values greater than 0.050 were removed manually. Exceptions were made if a factor was needed to maintain the model hierarchy. The resulting models for plants harvested after 40 and 47 d were compared, and a consensus model was established including all factors present in both models.

Transgenic plants tend to accumulate similar levels of recombinant proteins in all leaves, e.g., we recently observed a coefficient of variation (C.V.) of 36% across all leaves in transgenic tobacco plants producing DsRed and 2G12 (Fig. 1). Therefore, all leaves are approximately equal in value in terms of the product content, so a simple sampling procedure, such as the isolation and testing of one leaf from each plant, is sufficient to estimate the total amount of product formed in a batch and downstream equipment can be scaled accordingly. In contrast, our recent model predicted a sharp increase in protein levels between leaves 4 and 5, counting from the bottom of the plant, when transient expression is used to produce the same two recombinant proteins in plants of similar age. The prediction was confirmed empirically, resulting in a C.V. of 69% (Fig. 1). There was no sharp increase when younger plants were used, probably reflecting (1) the absence of a distinct sink-source transition between the leaves of young plants and (2) the more homogeneous capacity for protein biosynthesis in these leaves. A more complex sampling procedure is therefore required for transient expression systems in order to predict batch yields accurately.

Figure 1. Relative levels of DsRed expression in transgenic plant leaves (left) and in leaves infiltrated with Agrobacterium tumefaciens (right) as percentage of total expression. Expression is more evenly distributed among the transgenic plant leaves as indicated by the averages over all leaves and the corresponding standard deviations (lowest bars). Only the youngest leaves of transgenic plants show substantially lower expression levels than the mean, because recombinant proteins accumulate to reach a steady-state over time. Conversely, transient expression is highest in young, rapidly-dividing cells, hence the tendency for expression to decline with leaf age.

One way to mitigate the additional sampling requirements in transient expression experiments is to normalize our recently-described model. Normalization can be achieved by dividing the amount of target protein produced in the different leaves by the value of a “standard” leaf, e.g., the leaf with the highest expression level. The values for 2G12 and DsRed can be used to accommodate proteins targeted to different subcellular compartments, because 2G12 is secreted to the apoplast whereas DsRed is imported into the plastids (Table 1). The normalized values can be multiplied with the amount of a target protein detected in the “standard” leaf. We are currently investigating whether the normalized expression values obtained for 2G12 and DsRed can be extrapolated to other proteins and are transferable between laboratories that use different cultivation practices.

Table 1. Expression levels of recombinant proteins achieved 5 d post-infiltration, with values normalized to “standard” leaf six for plants harvested 47 d after seeding or leaf 4 for plants harvested 40 d after seeding.

Protein	2G12				DsRed
Plant age at harvest [d]	47		401		47		401
Leaf	Amount per leaf [µg]	Normalized [-]	Amount per leaf [µg]	Normalized [-]	Amount per leaf [µg]	Normalized [-]	Amount per leaf [µg]	Normalized [-]
1	0.05	0.09	0.15	0.32	0.02	0.03	0.12	0.10
2	0.10	0.18	0.32	0.67	0.06	0.09	0.40	0.34
3	0.28	0.54	0.47	0.97	0.16	0.25	0.79	0.68
4	0.28	0.53	0.48	1.00	0.20	0.31	1.17	1.00
5	0.55	1.03	0.20	0.42	0.58	0.90	0.57	0.49
6	0.53	1.00	—	—	0.64	1.00	—	—
7	0.43	0.81	—	—	0.78	1.22	—	—
8	0.27	0.50	—	—	0.48	0.74	—	—

¹ Plants harvested 40 d after seeding only possess five leaves, therefore no values are available for leaves 6–8.

An alternative way to predict accurate values for the concentrations of target proteins synthesized in different leaves by transient expression is to set up a customized DoE and model, but this requires careful consideration of several characteristics of the transient expression system. We have already noted the crucial impact of leaf and plant age, but even though both parameters can adopt defined values (such as 47 d post seeding) these values do not necessarily describe the same physiological state of a plant at all times. For example, we found that seasonal variation has an impact on the growth rate and thus the maturation of plants even in a controlled greenhouse setting where temperature, light and humidity are maintained within narrowly defined ranges, resulting in batch-to-batch variability and reducing the precision of the collected data (Fig. 2). We therefore suggest the inclusion of additional parameters such as plant height, leaf diameter or budding status in order to describe the physiological age of a plant more accurately. The use of strictly-regulated growth conditions in phytotrons can also help to reduce batch-to-batch variation but this type of environment is difficult to replicate for process-scale production.

Figure 2. Changing seasons induce variations in protein expression levels and thus reduce data quality and precision during model building. The diameter of the plant (A) is less responsive to changing seasons than the plant height (B). Therefore, the latter parameter is a more sensitive indicator for changing growth conditions. Error bars represent standard deviations (n = 20).

We also found that post-infiltration incubation temperature had a major impact on transient expression levels (Fig. 3A). Even a 2°C change away from the optimal temperature reduced DsRed expression by 25% and 2G12 expression by 15%, yet temperature changes of this magnitude can occur even in carefully-regulated environments. For example, there are often temperature gradients between the air conditioning inlet, outlet and thermometer, and in summer months the air conditioning system can reach its capacity. Furthermore, local temperature gradients may form around individual plants because upper leaves are closer to the light source. In order to detect and compensate for this kind of variation, the temperature should be monitored throughout the post-infiltration incubation area and should be strictly controlled.

Figure 3. Variation in transient protein expression levels can be caused by lax incubation temperature control and the imprecise scheduling of sampling times. (A) Temperature has a major impact on protein expression with a steep and narrow optimum range of 2–3°C for DsRed, independent of plant age. (B) Concentrations of DsRed change substantially during the interval 3–7 d post-infiltration, as indicated by the steep slope of the response-surface model. Sampling times must be scheduled precisely to avoid introducing unnecessary additional variation in protein expression levels.

Finally, sampling times should be scheduled precisely as indicated by our latest results (manuscript in preparation) because recombinant protein levels vary substantially between 4 and 8 d post-infiltration, the typical range for harvesting.^4,7,8 For example, we found that DsRed concentrations changed by 20% over 8 h on the fifth day post-infiltration (Fig. 3B).

Regardless of any pitfalls of transient expression in plants, reliable detection methods for physiological parameters and protein accumulation must be implemented to ensure that the DoE approach yields high-quality data. Among the various design types discussed in the literature for different scientific applications, the D-optimal and IV-optimal designs combined with response surface methods are thought to be the most suitable for the development of accurate predictive models.⁹ These can be implemented using software suites such as DesignExpert, JMP, Minitab, MODDE or STATISTICA. Important aspects of the design include the number of factors considered in the model, the mean standard error of the model required throughout the design space, and thus the number of single experiments included in the design. So-called block factors can also be relevant for the modeling of biological systems if the individual experiments within the design cannot be performed at the same time, e.g., using a single batch of plants. This is because block factors compensate in part for the natural variation associated with different batches of biological samples, thus improving the quality of the final model. The base model is another important aspect when modeling biological systems. This defines the type, maximum number and polynomial order of factor interactions that can be resolved by the final model, i.e., a linear base model (first order) will not allow the quadratic effect of a factor (second order) to be incorporated into the final model. For example, the temperature optimum we identified for the transient expression of DsRed and 2G12 in tobacco leaves required a quartic (forth order) model for the temperature factor because of the asymmetric (non-parabolic) shape of the optimum. Therefore, additional data from the literature or from initial screening experiments can help to improve and adjust the base model of a design, thus reducing or even eliminating the need for additional experiments to correctly describe the effect of all factors contributing to a specific response.

DoE approaches including model building can therefore improve our understanding of biological systems as exemplified by the transient expression of recombinant proteins in intact tobacco plants. Such models can help to evaluate the significance of observed effects in fundamental research projects, but when applied to actual processes they are useful to improve process design and scalability by providing accurate predictions of the impact of process variations on expression levels and protein recovery.

Glossary

Abbreviations:

C.V.

coefficient of variation

DoE

design of experiments

dpi

days post injection

dps

days post seeding

FDA

Food and drug administration

Pos

position on a leaf

Buyel JF, Fischer R. Predictive models for transient protein expression in tobacco (Nicotiana tabacum L.) can optimize process time, yield, and downstream costs. Biotechnol Bioeng. 2012:n/a–n/a. doi: 10.1002/bit.24523.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.