The optimization of protein structures to improve specific features like enzymatic activity, specificity, binding to ligands, affinity, or stability is an important task for the development of biotechnological products and processes. While rational design of proteins is a challenging task 15, screening of random libraries has been proven to be a valuable alternative in numerous cases (for reviews see 1617). A classical random screening approach is phage display of antibody libraries (first described in 18). Phage display can be regarded as the first example of surface display techniques, which share the common principle that a protein which is encoded in the genome of a host cell (or a virus) is displayed as a fusion protein on the outer surface of the same cell. Consequently, the genetic information is always linked to the respective protein variant in the same cell. Numerous applications of phage and cell surface display have been published. As a comprehensive overview, typical examples of pro- and eukaryotic surface display systems are illustrated in Fig. 2. As a general rule, the target (poly)peptide is fused to a native surface bound protein, be it on a virus capsid, the cell membrane or the cell wall.

Expanding the field beyond protein engineering, one can envisage an extremely wide array of cell engineering applications of single cell sorting. Despite the opportunities to screen for a multitude of different cellular features, cell sorting has been applied so far only in a rather small number of cases to cell engineering.

One of the basic challenges of strain improvement is the isolation of successful hybridization events, e.g. after cell (or protoplast) fusion or cell mating. As an example, FACS was applied to isolate rare yeast mating hybrids without selective markers. The parents strains were stained with two different fluorescent dyes, and a third, double stained population appeared appr. 16 h after mixing, which contained several mated cells 42.

A conceptual expansion would enable the screening for widely different genetic modifications, like the cloning of a random library to a specific position, or of genes with specific desired functions. The dual-fluorescence reporter system described in 43 refers to a method of library screening without plating, by the use of a plasmid encoding two different fluorescent proteins. Upon insertion of a DNA fragment, expression, and hence fluorescence, of one marker is impaired, so that all cells containing a plasmid with insert can be sorted based on the single fluorescence of the remaining dye. There is only a limited number of examples for FACS based cell engineering in the scientific literature, however, some patents describe high-throughput screening for novel bioactivities 4445. These methods are based on homologous or heterologous gene libraries employing either intracellular enzyme substrates or gel microdroplet encapsulation of cells. The gel microdroplet technique, first published by Weaver and coworkers 1988 46 is also described in section 3.3. for use as a method to sort for high production rates. However, it can also be used for physiological studies, measuring individual growth rates, acid production or other physiological parameters 474849.

Dunn and Handelsman described a method for promoter trapping by cloning a Bacillus cereus genomic library upstream of GFP 50. However, the authors only tested the efficacy of the sorting protocol with cell mixtures containing a 10⁴ fold excess of non-fluorescent cells (resulting in 95 % fluorescent cells after 3 sorts), but did not prove the applicability of the method to the initial goal of isolating promoters. In a similar approach strong promoter elements of Mycobaterium smegmatis were isolated by FACS sorting of a genomic library cloned upstream of a GFP gene 51.

Some other examples are the optimization of protocols for transfection of recombinant cells as well as the selection of new and strong promoters. A destabilized GFP, which does not accumulate in the cells, was used to characterize the kinetics of new promoters or the identification of sequences that enhance secretion 52. Frequently used tools for this type of research are reporter genes such as β-galactosidase 53 or GFP 54.

While the examples described above relate to broad method development, there are some specific applications of cell sorting to improve cellular characteristics for technological application. Viscardi and coworkers applied cell sorting for immunoselection of phage-resistant Streptococcus thermophilus (to be used in the dairy industry). By incubating the cells first with phage, followed by anti-phage antibodies and a fluorescent secondary antibody, the authors could isolate clones that lost interaction with the phage. While it is widely accepted that such a system of "negative staining", i.e. sorting of rare non-fluorescent cells, is more cumbersome and error-prone, these authors were able to successfully isolate desired clones after a single sort 55.

The improvement of cellular properties like viability or stress tolerance may be directly connected to improved overproduction of a desired biotechnological product. As an obvious example, it was demonstrated that the screening for yeast cells with enhanced resistance to weak organic acids in an acidic environment leads directly to strains with enhanced ability to produce lactic acid. Based on the observation that cells with a higher intracellular pH (pH_i) have a better tolerance to acidic conditions, a FACS sorting strategy was developed. By screening for cells within the highest range of pH_i, clones with a higher tolerance to acidic environment and a higher productivity of lactic acid were achieved (M. Valli, D. Mattanovich et al., manuscript in preparation).

Other cellular properties that have been selected by cell sorting concern the behaviour of cells during large scale production as well as the stability of recombinant gene expression in CHO cells. Böhm et al. selected for cells with high expression rates during stationary phase or at high cell densities, so that the resulting clones would be better suited for a high density fermentation system 56. As the production of therapeutic proteins needs to be performed without the usually toxic substances used as selective markers, they also screened for increased stability of recombinant gene expression under these conditions. With a similar objective, a bicistronic vector expressing GFP and interferon gamma was used to sort for cells expressing GFP after serum deprivation under growth arrested conditions 57. Schlatter and coworkers used a surface marker to sort for proliferation controlled cells, employing both FACS and MACS 58.

Recently, Borth and coworkers were able to select for recombinant CHO cells with altered glycolytic metabolism: cells selected for low mitochondrial membrane potential were found to have lower uptake rates for energy substrates such as glucose and glutamine. At the same time the production rate of lactate was decreased, while the growth rate of cells increased due to the more efficient energy metabolism. Incidentally, the specific production rate for monoclonal antibody was also increased (G. Brugger, N. Borth et al., manuscript in preparation).

When sorting is used for the selection of over-producing cells, there are three main objectives: (1) to reduce the work load necessary to find over-producers, (2) to reduce the time required and (3) to find the cells with the highest possible production rates, ideally combined with other advantageous properties (see section 3.2.). Brezinsky et al. have nicely illustrated in their work how cell sorting can effectively resolve all three of these objectives 59. When sorting for high producing recombinant CHO cells, the obtained clones fell into the following categories: 12% very low, 50% low, 32% average and 8% high producers. Clones obtained by limited dilution cloning were 90% very low, 10% low producers, no average or high producers. The timeline for sorting was 6–12 weeks, compared to at least the double with traditional methods, as it was necessary to amplify the gene copy number to reach comparable production rates. Amplification takes more time, it requires another round of subcloning by limited dilution and an additional load of immunoassays for clone testing. In general, simply by eliminating the non- or low-producing cells, sorting will reduce the number of assays to be performed, because only the interesting cells will get to this stage. To obtain stable high producers, it is necessary to isolate single cells, which is not guaranteed by classical limited dilution, because otherwise the faster growing low producers will outgrow any high producers present 6061.

As many biotechnological products are secreted and thus dissociated from the cells that produced them, three strategies have been described to sort cells directly for high production rates: the first is to catch the product on the surface of the cell that secreted it by applying an artificial matrix on the surface of cells which will bind the product. This affinity matrix approach was establish by Manz and coworkers to study the secretion of immunoglobulin G (IgG) by plasma cells and later on of cytokines by T-cells 62. It has been used both for the isolation of hybridoma cells 6364 and for the selection of recombinant CHO cells 65. An outline of the affinity matrix approach is presented in Fig. 3A. The use of the same approach for yeast cells was less successful for so far unknown reasons 66. The second strategy was established in 1988 by Weaver et al. and relies on the capture of the secreted product within a microcapsule or gel that surrounds the producer cell 466768. Finally, Brezinsky et al. developed a strategy that simply "freezes" the product secreted at that moment in the cell membrane by putting the cells on ice. Subsequent staining with a fluorescent antibody against the product allowed to discriminate between cells according to their secretion rate 59.

Before the establishment of these methods, several authors used the correlation found in some hybridoma cell lines between surface expressed IgG and IgG production rate to sort for high producers. This correlation is a specific feature of some hybridoma cell lines, deriving from B-cell development 6970717273, and does not apply for recombinant proteins. Native surface expression must be differentiated from the surface capture or entrapment methods described above. Sorting for surface expression directly after hybridoma fusion will loose many specific monoclonal antibody producing cells, which happen to have no surface expression, while both the microcapsule technique and the affinity matrix method have been used successfully for this purpose 64. Another strategy is the use of GFP-product fusion genes to sort for cells with high GFP production, which will also have high production rates of the gene of interest. However, this strategy has an inherent problem: the cells need to produce two proteins, of which only one is required. This puts a considerable, unnecessary stress on the cell. Although it was shown that the clones with a 3× higher specific product formation rate could be found and that the expression of both product and GFP is reasonably stable, no data on growth rates of these clones or on performance in bioreactors was given 74. However, these properties are of equal importance for an industrial production cell line as high productivity.

Even though cell fixation to access intracellular product seems to be contradictory, as the cells are necessarily killed prior to sorting, immunofluorescence staining of fixed cells was successfully applied for sorting of E. coli cells expressing recombinant superoxide dismutase 75. As strong overexpression of this as well as many other proteins proved to retard cell growth, the rationale was to screen for promoter mutants out of a mutation library which guide less overexpression, thus enabling continued growth. As the mutation library was localized on a plasmid, it was possible to isolate the plasmids out of fixed sorted cells, retransform them into E. coli, thus rescuing the desired promoter variant. Interestingly, most of the sorted promoter variants had multiple point mutations, which essentially cannot be screened by low throughput techniques. Figure 3B outlines the principle of plasmid screening with fixed E. coli cells.

While it seems obvious to apply cell sorting for the selection of overproducing clones, the number of published examples dealing with non-protein products is interestingly quite limited. One example is the screening of high producers of polyhydroxyalkanoates (PHAs). Vidal-Mas et al. have described a flow cytometry protocol to measure PHA content of Pseudomonas aeruginosa after Nile red and SYTO-13 staining, indicating the utility of this method for cell sorting 76. Alternatively, Vijayasankaran et al. used heterologous co-expression of GFP in Pichia pastoris for the flow cytometry screening of clones with increased overexpression of three PHA biosynthesis genes, thus isolating strains that accumulated appr. 7 % PHA of the cell dry weight instead of 3 % for unsorted strains 77. As GFP was expressed with an inducible and the PHA biosynthesis genes with a constitutive promoter, the above described problem of an extra load of GFP synthesis during production was avoided. In this setup, GFP was induced only for sorting, while for PHA production the gene was silent. While the same group has developed sensitive staining of PHA by BODIPY 493/503 78, this probe was apparently not used for sorting.

Several studies have been published on the sorting of plant protoplasts, e.g. for alkaloid production [e.g. 79]. For more details the reader is referred to references in 80. In some cases quite unspecific staining has been successfully used for sorting of overproducers. One example is gramicidin S production by Bacillus brevis, based on the observation that gramicidin S overproducing cells gave higher fluorescence signals after fluorescein-isothiocyanate (FITC) staining as compared to low or non producers. By sorting for high FITC fluorescence these authors were able to isolate a strain with appr. twofold more gramicidin S production 81. It should be noted that the potentials of sorting based on autofluorescence or unspecific staining are extensive, hence it is not the intention of this review to list every single application available in the literature.