The promoter and 5′‐untranslated region (5′UTR) play a key role in determining the efficiency of recombinant protein expression in plants. Comparative experiments are used to identify suitable elements but these are usually tested in transgenic plants or in transformed protoplasts/suspension cells, so their relevance in whole‐plant transient expression systems is unclear given the greater heterogeneity in expression levels among different leaves. Furthermore, little is known about the impact of promoter/5′UTR interactions on protein accumulation. We therefore established a predictive model using a design of experiments (DoE) approach to compare the strong double‐enhanced Cauliflower mosaic virus 35S promoter (CaMV 35SS) and the weaker Agrobacterium tumefaciens Ti‐plasmid nos promoter in whole tobacco plants transiently expressing the fluorescent marker protein DsRed. The promoters were combined with one of three 5′UTRs (one of which was tested with and without an additional protein targeting motif) and the accumulation of DsRed was measured following different post‐agroinfiltration incubation periods in all leaves and at different leaf positions. The model predictions were quantitative, allowing the rapid identification of promoter/5′UTR combinations stimulating the highest and quickest accumulation of the marker protein in all leaves. The model also suggested that increasing the incubation time from 5 to 8 days would reduce batch‐to‐batch variability in protein yields. We used the model to identify promoter/5′UTR pairs that resulted in the least spatiotemporal variation in expression levels. These ideal pairs are suitable for the simultaneous, balanced production of several proteins in whole plants by transient expression. Biotechnol. Bioeng. 2013; 110: 471–482. © 2012 Wiley Periodicals, Inc.

Eight different promoter/5′UTR combinations were cloned, introduced into Agrobacterium tumefaciens and used for transient expression of the fluorescent reporter protein DsRed in different leaves of Nicotiana tabacum. The resulting expression profiles were compiled to a predictive model using a design of experiments approach.