Several recent papers suggest that manipulation of a few genes can greatly increase the plant ability to survive under water deficit. Briefly, these genes have two types of functions. (i) A first category of genes is involved in the resistance to partial cell dehydration. The effects of these genes is either to alleviate biochemical effects of stress by an accumulation of molecules that stabilise proteins, membranes and the biological structures of partially dehydrated cells [13], or that avoid the accumulation of toxic species due for instance to oxidative stress [30]. A source for such genes lies in desert ‘resurrection plants’ such as Craterostigma plantagineum, which can survive complete desiccation and resume growth after rehydration. (ii) A second category of genes is involved in stress avoidance by reducing transpiration via rapid stomatal closure or limitation of leaf growth. This can be achieved by several means, for instance manipulating the hormonal balance of the plant [18] or transcriptional activators [17]. Note that most of the transformed plants with increased tolerance or resistance have reduced growth even in well-watered conditions. Because they deplete soil water more slowly than wild types, transformed plants stay green longer and die after the wild plants during a sequence of soil drying.

These studies have led to a considerable progress in the understanding of the mechanisms that allow plants to survive severe water stress. This progress will probably continue because the accumulation of cell protections does not face obvious physical or physiological limits. Besides an obvious biological interest, it may have applications for developing plants able to colonise slopes and thus avoid erosion in Mediterranean or semi-arid climates, or to grow in very dry environments (especially for cattle use). However, they may well be of limited interest in most agricultural contexts, because they address the problem of survival rather than that of production. Plant survival is usually not relevant to agriculture, because farmers avoid sub-lethal conditions (accompanied by extremely low yields) by using an ‘escape’ strategy when choosing their cropping system (e.g., crops with a shorter cycle or pasture instead of crops). Until now, no clear demonstration has been made that the transfer into crop species of exotic genes that confer survival capacity can help to maintain yields under water deficit. Rather, there are theoretical reasons to consider that rapid biomass accumulation is linked to high water use (see below), so genetic engineering will never produce plants able to rapidly accumulate biomass in a desert.

Transpiration per unit biomass is several orders of magnitude greater in plants than in animals. For instance, a typical transpiration rate of 1 m² of plant canopy in summer in southern France is 6 l d⁻¹ m⁻². This volume of water taken up by roots every day, which flows in the xylem and is transpired by leaves via the stomata, is similar to the water volume in a plant. When the water availability to roots decreases, plants tend to decrease transpiration by two means. A short-term effect consists in closing the stomata, thereby reducing the water flux through the plant. A longer-term effect consists in reducing leaf growth, which results in a smaller transpiring leaf area. These two mechanisms of transpiration reduction are adaptive processes that conserve water for later stages of plant development, but they also contribute to a partial homeostasis of the water status of shoot tissues.

When stomata partially close, thereby decreasing transpiration, the leaf water potential (the physical variable that represents the degree of tissue hydration) increases, i.e. leaves become more hydrated. This mechanism allows leaves to maintain their water status in a narrow range. Roots have a signalling role in this process via the circulation of hydraulic and chemical messages to leaves when they begin to sense water depletion [34]. This is exemplified in Fig. 1, which represents a sequence of soil drying and its consequences on a field-grown maize canopy [36]. As the non-irrigated soil dried, stomatal conductance (a physical variable whose variations represent the degree of stomatal aperture) decreased and the concentration of a stress hormone (abscisic acid, which controls stomatal conductance) increased, but the water status of leaves (day-time water potential) remained essentially the same in well-watered and drought-affected plants. These results changed our views concerning the effect of water deficit on growth and transpiration: the cells of drought-affected plants experienced the same level of ‘stress’ as those of well-watered plants in terms of their water potential.

The above observations, among others, resulted in a multi-species model involving water and hormone fluxes in the xylem, the control of stomatal conductance and the changes in water potential of different organs of the plant [34,35]. The model was recently tested in transgenic plants [4] and extended to predict CO₂ fluxes [11]. Briefly, any change in transpiration rate is accompanied by a change in the gradient of water potential between roots and leaves with a mechanism similar to the Ohm's law. In this way, fine-tuning of stomatal conductance via root messages, in particular abscisic acid, allows plants to keep a partial homeostasy of leaf water status. In most circumstances of water deficits compatible with agriculture, the leaf cells therefore avoid severe dehydration via a reduction in transpiration. Conversely, there is a constitutive link between high transpiration rate (a ‘favourable’ trait linked to photosynthesis) and low leaf water potential (an ‘unfavourable’ trait associated with tissue dehydration).

Stomatal aperture and leaf area, which are the main determinants of plant transpiration, are also the main determinants of carbon accumulation by plants because CO₂ flows to photosynthetic sites via the stomata. This leads to well-documented relationships between light interception, transpiration, photosynthesis, and biomass production [21]. There is therefore a ‘built-in’ contradiction between stress avoidance via a reduction of transpiration and biomass accumulation. The latter can be temporarily buffered by carbon storage in the stem [2] or by changes in carbon metabolism [20], but fast biomass accumulation depends in the longer term on leaf area and stomatal conductance.

A second contradiction exists between avoidance of water stress and avoidance of heat stress. During a sunny summer day, the major part of the incident solar energy is dissipated via transpiration, so non-transpiring leaves are warmer than transpiring leaves by up to 10 °C [19]. A reduction in transpiration rate via stomatal closure can therefore be accompanied by heat stress, and its consequent effects on development time and growth.

Overall, the above suggests that manipulating genes involved in cell protection, which increases the survival of severely stressed plants, may have less influence than frequently believed. This is because of a partial homeostasy of leaf water status under water deficit linked to a fine-tuned reduction in transpiration via the controls of stomatal aperture and leaf growth. These reductions in transpiration, which have a positive effect on plant water status, have two negative effects, namely a reduction in photosynthesis and a higher risk of heat stress. This suggests that tolerance to water deficit might be regarded as an optimising process limited by several physical constraints and trade-offs, rather than as a classical resistance process driven by a series of genes.

An efficient way to obtain plants that can cope with water deficit is to adapt their cycle to expected climate conditions. In particular, reducing the duration of the crop cycle reduces the total soil water depletion, so that crucial stages such as flowering or grain filling occur in a soil with higher soil moisture, resulting in an acceptable yield. The drawback of this strategy is that the total amount of light intercepted by plants is reduced, and so is the potential accumulation of biomass (Eq. (2)). By using a genotype with a shorter cycle, the farmer therefore obtains an acceptable biomass and yield in exchange for a limitation of the maximum yield that can be expected.

A way to maintain transpiration and photosynthesis without negative effects on the plant water status is to increase the ability of the root system to take up water. This can be achieved either by improving the soil tillage [32], or by selecting genotypes with increased root growth or branching [39]. Spectacular results have been achieved in this way [38], but it should be mentioned that they apply essentially to cases in which there is a soil water reserve that is not exploited by roots (e.g., water table or deep soil). In cases where plants grow on a limited amount of water (e.g., shallow soil), improving the root system ability to take up water is of little interest or even counter-productive [26]. This may explain a counter-intuitive result in which the selection of maize lines with the highest yields under drought conditions resulted in a reduction of the carbon investment in the roots [3].

Although biomass accumulation and transpiration are intrinsically linked, one might expect that their ratio can be manipulated to improve the amount of CO₂ captured by the plant per transpired unit of H₂O (intrinsic water-use efficiency). Farquhar et al. [12] showed that there is a striking similarity between, on the one hand, the equations of water and carbon transfer in the leaf and, on the other hand, the laws of discrimination against the heavy isotope of carbon, ¹³C, in relation to ¹²C during the process of photosynthesis. This results in a strong relationship between carbon isotope discrimination (termed Δ13C) and the intrinsic water use efficiency. This (negative) relationship has been observed experimentally in a large number of species, with the notable exception of plants such as maize, sorghum, or sugarcane, which have a special photosynthesis metabolism. Because Δ13C can easily be measured with a mass spectrometer, agronomists and breeders have an efficient way to indirectly measure water use efficiency and carry out breeding programmes to improve it.

Selecting for increased water-use efficiency alone often results in decreased biomass accumulation and yield, with a detrimental effect that reaches a maximum under well-watered conditions [8]. This is part of the ‘water for carbon trade-off’ theory presented above: water-use efficiency increases when the stomata tend to close, reducing both transpiration and photosynthesis. Selecting for high water-use efficiency alone therefore results in selecting for slow growth and low yield. However, interesting results were obtained when special care was taken to overcome this contradiction [23]. Fig. 2 shows that, when tested in a series of experiments with varying rainfall, wheat lines selected for high water-use efficiency had an increased yield under very dry conditions, while the effect was negligible under milder water deficits. The most likely explanation was, again, that these lines had reduced transpiration and conserved soil water for the end of the crop cycle. It is noteworthy that both yields and rainfall were very low in these experiments (1 to 3 t ha⁻¹, i.e. one third of the yields classically observed in Europe), although they are typical of Australian production conditions.

A reduction in leaf expansion rate occurs before any reduction in photosynthesis or in other biochemical processes [5,27]. Growth maintenance can therefore be viewed as a physiological mechanism per se with several components. Cell turgor is a positive pressure in cells (usually ranging from 0.3 to 1 MPa), which is the driving force in tissue expansion. It is at least partially maintained when leaf water potential decreases. Because water cannot be forced into cells in the absence of biological pumps, plants use an indirect mechanism that consists in a build-up of solutes in the cell, which causes water to enter into the cell (osmotic adjustment). In situ measurements of turgor suggest that osmotic adjustment is already very efficient, with near-perfect turgor maintenance under water deficit [31], so there may be little room for genetic improvement of this characteristic. On the other hand, many breeding groups have improved osmotic adjustment with positive effects on yield [40], although this has been questioned recently [28]. Other mechanisms involved in the reduction of growth under water deficit are essentially reduced cell division rate [15] and cell wall stiffening, which blocks cell expansion, even with maintained turgor [9]. These mechanisms are themselves controlled by the plant hormonal balance and by the expression of several families of genes.

It can be argued that increasing growth under water deficit makes plants more susceptible to water deficit because of increased transpiration and soil water depletion. This is certainly true in very dry climates such as those presented in Fig. 2. However, several lines of reasoning suggest that this is not the case in moderately dry climates.

• • If growth response to water deficit is a physiological process per se, it is expected that the growth of several organs be affected by the same genetic variability, including that of the reproductive organs that are harvested. A ‘success story’ of plant breeding for tolerance to water deficit has consisted in selecting maize plants with a rapid initial growth of the reproductive organs under drought, indirectly observed by the time elapsed between male and female flowerings [3]. Selecting for improved silk growth under stress resulted in increased maize yields in moderately dry conditions [6], via an increase of the term HI in Eqs. (1) and (2).
• • The water-saving strategy consisting in a reduction of the leaf area is not as efficient as was expected in moderately dry climates, because of the direct evaporation from the soil when it remains partly wet [8].

Maintaining growth, either of the leaves to maintain photosynthesis, or of the reproductive organs to increase the assimilate storage in harvested organs, can therefore be an efficient strategy. Because a genetic change in growth rate cannot have lethal consequences (unlike manipulations of hormonal balances or of signal transduction), there is a considerable natural genetic variability of the degree of growth maintenance under water stress for both reproductive organs [7,25] and for leaves [24]. An example is presented in Fig. 3, in which the elongation rate of maize leaves is plotted against two components of water deficit, either in the air or in the soil, in several experiments. Two genotypes are presented, which have consistent differences in response to water deficit over several experiments. The existing genetic variability of growth maintenance under water deficit can therefore be exploited for developing genotypes with reduced sensitivity. However, it should be stressed again that such genotypes are expected to have a reduced tolerance to very severe water deficits because of faster soil water depletion. A farmer using such genotypes would therefore trade a higher potentiality of biomass accumulation under mild water deficits for an increased sensitivity to very severe water deficits.

Although I have cast doubt on the value of transferring into crops a series of single genes that would enable them to survive the stress, more elaborate strategies of genetic engineering are now feasible. In particular, it is conceivable to transfer favourable alleles of several genes already present in the studied species, or to manipulate the expression of genes under water deficit by manipulating transcription factors. The application of this strategy to the case of water deficit remains to be tested. The main problem will probably be the prediction and the test of the combined effects of the changes in the expression of several genes, within the complex framework described above, characterised by contradictory effects and physical limitations. A possible way of tackling it could be to place the effects of all genes known to be involved in the genetically engineered function into a regulatory network that simulates the effects of fluctuating environmental conditions on their transcript levels, on the amount and the activities of corresponding proteins, and finally on the resulting phenotype. Plant transformation would therefore follow in silico research selecting the most interesting combinations of genes and alleles. This strategy, already used for simulating the behaviour of prokaryotes as a function of their environment [14], has been considered suitable for simulating the effects of water deficit on plants [10].

I have argued elsewhere that this is not feasible yet and may well face theoretical limits that make it impossible in the near future [33]. (i) Our understanding of controls and of involved genes is still poor. Only a small proportion of the genes involved in the response to water deficit is known, and conversely genes with almost any function may have a role in this control. (ii) It is virtually impossible to simulate the interaction between genes in complex systems. A study of only 15 genes with three alleles each would generate 14 million genotypes, an impossibly high number to analyse in enough detail to avoid misinterpretations [33]. (iii) In contrast to existing examples of regulation networks, the phenotype is usually quantitative (i.e. with an infinite number of possibilities) and cannot be interpreted without at least an implicit model. It is therefore proposed that a ‘bottom-up’ strategy consisting in a transfer of genes of known functions is not an appropriate strategy in the search for water-stress tolerance [16,29,36], although it is essential for analysing functions involved in the plant responses to water deficit (e.g., [37]). Strategies based on available genetic diversity would therefore be the most likely to succeed.

In many programmes of crop breeding, germplasm that is better adapted to dry conditions have been developed and released by using conventional high-throughput, long-term selection strategies that emphasise yield alone. There are fewer cases of selecting for drought performance by selecting for specific physiological traits. Two interesting cases are described here above: water-use efficiency [23] and growth maintenance [6]. In both cases, the genetic strategy was based on the search for favourable alleles associated with the traits, in the existing genetic material. Briefly, the most common strategy consists in crossing parents with different phenotypes and genetic backgrounds. In this way, one obtains a large offspring population whose genome is recombined, i.e. in which each chromosome consists of portions of the chromosomes of each parent. An increasingly common approach is to use statistical analysis to associate one allele at a given position of the genome with a phenotypic trait (quantitative trait loci, QTLs). Then, a molecular marker for the trait can be selected directly from seedlings, without having to grow the plants to yield. This labour-intensive technique has the advantage of associating a plant trait of interest with a locus in the genome, even though the underlying cellular-level mechanisms are not known. Its main disadvantage is that this association is statistical, and frequently varies with environmental conditions, i.e. a trait of interest is associated with different parts of the genome in different experiments.

The above strategy can be combined with ecophysiological models that integrate numerous functions and their rapid changes with environmental conditions. Ecophysiological modelling allows prediction of plant behaviour in fluctuating conditions, potentially making it possible to predict the behaviour of a plant in any pedoclimatic conditions. Such models are based on both physical equations, representing fluxes of mass and energy, and control equations that represent the plant response to environmental conditions. The value and weakness of these equations are described elsewhere [33]. Briefly, the fact that the response curves are common to several experiments (Fig. 3) allows one to consider them as ‘meta mechanisms’ at the plant level, although the underlying molecular mechanisms are not known. We have recently proposed such a combination of genetic and ecophysiological models applied to the genetic variability in leaf growth under water deficit [24]. Briefly, a genetic analysis was carried out on the slopes of the response curves shown in Fig. 3, thereby associating alleles with either increased or decreased responses to water deficit. Theoretically, this makes it possible to predict the behaviour of virtual genotypes with any combination of alleles. This possibility was tested by analysing genotypes whose behaviour was either simulated or measured in an experiment with varying environmental conditions (Fig. 4). The model predicted the general trends of the changes in leaf elongation rate with environmental conditions, but also the differences in elongation rates observed between genotypes that were known only by their genetic characteristics.

This study, carried out on a simple function, leaf growth, and in a simple genetic population, might be extended to more complex situations and open the way to an in silico prediction of virtual genotypes, thereby allowing a re-engineering of plants. Only the best genetic combinations would then be obtained genetically and tested in the field.