The most obvious plant characteristic - photosynthesis - depends on terrestrial plants acquiring, transporting and transpiring water to maintain the functional integrity of their photosynthetic apparatus, sustain water-splitting activity and carry out carbon dioxide fixation. There is no growth without water movement through cells and into and out of the plant vascular system, and most plants are severely damaged by even a few hours of water deficit. Evaporation is hormonally regulated through stomata, which adjust in response to an imbalance between water flux in the vascular system and the rate of transpiration. Apart from the macroeconomy of water supply and allocation, we know little about how water movement is regulated within the cell and between cells in different tissues. Until recently, we were ignorant of the existence of proteins facilitating water movement [1,2], although the existence of proteinaceous water channels in plants had been suggested long ago, and received strong experimental support a decade ago [3,4]. The discovery of a family of genes encoding these protein channels, now known as aquaporins (AQPs), had been anticipated in principle by the earlier studies because water permeability, at least in some tissues and cells, has characteristics that cannot be explained if the lipid membrane is the sole barrier to water exchange (for reviews see [5,6,7,8]).

In macroscopic terms, water movement through plants had long been thought of as a biophysical problem, as a topic of capillary resistance in pressurized pipes terminating in 'sprinkler' systems - the stomata - whose apertures adjust when responding to changes in hydraulic pressure. A combined pathway for water through the apoplastic (cell wall) space and through cells, either transcellularly across the cell wall, or through the symplast via plasmodesmata, placed little emphasis on membrane barriers impeding such movement. Reports of an energy-dependent metabolic component to water flux and the poisoning of water flux by mercurials, however, pointed to additional complexity. Nevertheless, the existence of water channels does present a real puzzle, considering that water movement through simple lipid bilayers is significant and the AQPs increase conductance by a factor of maybe 10 (range 0 to 30) [6,8]. Model experiments with Xenopus oocytes injected with AQP cRNA have amply confirmed that many AQPs do facilitate water movement. Most important, AQP functions have been demonstrated in vivo in plants [9,10,11,12].

Aquaporins belong to a group of often abundant membrane proteins of molecular mass 26-35 kDa. They have six transmembrane α helices (TM) and two additional membrane-embedded domains. Their amino and carboxyl termini are located on the cytoplasmic surface of the membrane. The amino- and carboxy-terminal halves of the polypeptide show sequence similarity to each other and are arranged as a tandem repeat, which apparently originated from the duplication of a half-sized gene encoding three transmembrane domains. Each half bears one hydrophobic loop which includes a highly, though not absolutely, conserved Asn-Pro-Ala (NPA) motif [13,14]. The loops containing the NPA motif, located to the carboxy-terminal side of TM2 and TM5, respectively, overlap in the middle of the lipid bilayer ('hourglass' model) and form two hemipores that generate a narrow channel [15,16,17]. Erythrocyte CHIP28 (human AQP1) was the first AQP shown to form water-selective channels [18]. AQP1 forms tetramers [16,17,19] but monomers alone can facilitate water flow [20,21].

Two high-resolution X-ray structures are now available, one for human AQP1 (HsAQP1, CHIP28) and one for the Escherichia coli aqua-glyceroporin GLP-F [22,23,24], which transports glycerol. The model for HsAQP1 outlines the pore geometry and suggests a single-file passage of water molecules while the passage of protons is prevented. The two asparagine residues (76 and 192, both in NPA motifs) form hydrogen bonds with a single water molecule, interrupt the hydrogen-bonded file of water molecules in the pore, and thus control water passage. The GLP-F structure [23] describes a channel whose selectivity filter and pore are different, forming an amphipathic channel. The alkyl backbone of the transported glycerol is adjacent to the hydrophobic surface and the hydroxyl groups form hydrogen bonds with the hydrophilic surfaces [23,24]. This bacterial GLP-F is most similar to the NIP/SIP group of plant AQPs, whereas the PIP and TIP plant AQPs are more similar to the animal AQPs (see below). GLP-F transports ribitol and xylitol at a rate lower than that of glycerol, adding an interesting feature to GLP-F's possible physiological functions. The GLP-F structure model is consistent with a string of glycerol (or polyol) molecules following a path that involves successive hydrogen bonding along the channel and through the pore. Also, induced conformation changes may determine which metabolites (possibly including ions) pass through, providing additional flexibility and diversity. The GLP-F structure also suggests that the center of the tetrameric complex might constitute a different channel on its own [23]. As more structure models become available, AQP diversity is becoming obvious because, irrespective of overall sequence homology, the surface topographies of different AQPs as seen by atomic force microscopy seem to vary widely [25,26]. On the basis of deduced amino-acid sequences and one crystallographic model [27], plant AQPs seem to include the same general structural motifs that have been detected in mammalian and bacterial AQPs. Figure 1 presents a schematic view of notable domain structures and functional features [24,28].

After the first aquaporin from plants (γ-TIP) was cloned and functionally expressed [29], plant AQPs received considerable attention. If functional, their bearing on our understanding of the biophysics of water flow across plant membranes could be significant [5,6,8,30,31,32]. New evidence indicates that the function of AQPs and aquaporin-related proteins may not solely, or not explicitly, be in water-flux facilitation. In addition, glycerol, urea, ammonia, other uncharged small metabolites, and possibly carbon dioxide and ions, seem to traverse membranes containing these 'aquaporins' [6,8,31,32,33,34]. During the past decade, several plant AQPs have been characterized using Xenopus oocyte, yeast, or plant protoplast swelling assays. Although such analyses are far from exhaustive, the evidence supports disparate water-flux characteristics among AQPs, possibly reflecting the fact that their real substrates might be quite different [35,36].

The AQP gene family in plants is unexpectedly large. In contrast to the 10 or 11 AQP genes in the human genome [24], more than 20 expressed AQPs were detected in Arabidopsis by cDNA library screens or RT-PCR [37]. A growing number of transcripts, detected as expressed sequence tags (ESTs), indicated that the actual number might be even higher in maize, where at least 31 unique AQPs have been detected [28]. With the sequence of the A. thaliana (Columbia) genome completed, classical approaches to assessing AQP gene complexity can be replaced by analysis of the genome [38]. We analyzed the Arabidopsis genome sequence for the presence of aquaporin-related sequences and summarize what is known about this gene and protein family.

The Arabidopsis AQP gene family

Thirty-eight AQPs and AQP-like sequences were found in the Arabidopsis genome, ranging in size from 2 to 3 kb. Their positions on chromosomes, accession numbers and names are given in Table 1. The nomenclature of these genes has been chosen such that established names are maintained to a degree, and we label unnamed putative PIPs (for plasma membrane intrinsic protein) as PIP, all putative TIPs (tonoplast intrinsic protein) as TIP, and the various NLMs (for nodulin-26-like membrane protein), whose functions are largely unknown, are put into two subfamilies termed NIPs and SIPs, respectively [39]. When aligned and compared by the Clustal-X/TreeView programs [40,41], the deduced protein sequences separated into the four major branches (PIP, TIP, NIP and SIP; Figure 2) that fit with a preliminary consensus on nomenclature that is developing among workers in this field [39,42]. Included in Figure 2 are HsAQP1 and the bacterial GLP-F protein sequences (GLP-F E. coli) for comparison. Location of the TIP subfamily of 11 sequences in vacuole membranes has often been documented, whereas PIP location is less certain and may be variable, depending on the conditions [32,43,44,45]. The group for which the least information is available includes 14 sequences in two subfamilies - NIP (11) and SIP (3). NIP/SIP family members show higher similarity to bacterial glycerol facilitators than the genes in the two other groups.

Homologies

Of the four groups of Arabidopsis AQPs, identity between the 13 homologs of the PIP subfamily at the deduced amino-acid sequence level was found to be highest, ranging from 71.8 to 97.2%. Thus, the PIP cluster is quite distinct and separate from all other AQPs (Figure 2). The TIP subfamily was found to have 11 members, with intra-group identities ranging from 44.1 to 93.1%. The NIP group includes 11 genes with even lower intra-group identity scores, between 38.9 and 84.7%. In the fourth subfamily, the three SIP genes are 28.1-71.2% identical to each other. The percent identity between sequences of different AQP subfamilies is approximately as low as the identity between all Arabidopsis AQPs and the human AQP1 and bacterial GLP-F. The range there is 22.1 to 33.1%. Homology analysis showed that PIPs are 32.5-37.6% identical with TIPs, 26.8-29.3% with NIPs and 21.8-25.5% with SIPs. These findings underscore the overall sequence diversity, and could indicate significant functional differences.

Distribution of AQPs on the chromosomes

Figure 3 shows the distribution of the 38 AQP-like sequences on the five chromosomes of Arabidopsis [38]. AQP genes are not obviously clustered, but two regions, which resulted from a duplication of segments in chromosomes 2 and 3, respectively, contain a higher than normal density. In total, nine AQP sequences were located in these two regions (PIP1;1, PIP1;2, PIP2;1, PIP2;3, PIP2;5, PIP2;6, SIP2;1, PIP2;8 and TIP1;1). The duplication of a segment within chromosome 1 includes an AQP in each copy, the closely related TIP3;1 and TIP3;2. Yet another duplication, on chromosomes 3 and 5, harbors SIP1;2 and SIP1;1, respectively. Three AQP genes are located within a few BAC clones from the ends of chromosomes. PIP1;3 and NIP6;1 occupy either end of chromosome 1. PIP1;4 is located close to end of the short arm of chromosome 4. For four genes, duplications within a short range are obvious; in-tandem arrangements are found, namely NIP1;1 and NIP1;2 (chromosome 4), and NIP4;1 and N1P4;2 (chromosome 5), respectively. Whereas the duplications and translocations outline mechanisms that have contributed to the evolution of the large AQP gene family, the large number of genes and the transcription patterns we found in the EST database (see below) seems to argue for specific functions in plants.

Exon-intron structure

The 38 sequences were also analyzed for distribution of introns and exons; the results are shown in Figure 4 and Table 1. The division into four subfamilies inferred from comparison of the deduced protein sequences (see above) is mirrored in the intron-exon structures. PIPs invariably include three introns. For the TIPs, two introns is usual, but several of the genes (TIP2;3, TIP1;2, TIP2;2, TIP2;x-pseudo and TIP1;1) have lost intron one. TIP1;3 has no introns at all and no cDNAs or ESTs have been reported for this gene. TIP2;x-pseudo with one intron seems to be a pseudogene (see below). The NIP group is characterized by four introns. NIP2;1 and NIP5;1 in this group lack intron 2, and intron 3 is missing from NIP3;1. The genes in the SIP group are characterized by two introns. The intron insertion positions are different between the four sub-groups but are conserved within a sub-family, with the exception of PIP2;4 where intron 2 is located at approximately the same position as the introns 2 in TIPs. Intron length varies widely in the range of 80 to 300 nucleotides with two exceptions. Introns in NIP4;1 are between approximately 500 and 700 nucleotides in length and NIP5;1 is characterized by a first intron of 1,726 nucleotides. We take the number and largely conserved location of the introns as a strong support for classifying four sub-families. This grouping is also consistent with the relative similarity of deduced amino-acid sequences and our analysis of transcript abundance (below).

Genes and pseudogenes

Our sequence analyses indicate that certain modifications of AQP annotations should be made in the genome sequence and also that 3 of the 38 sequences are pseudogenes. For NIP3;1 the computer prediction omitted exon 1 and intron 1. A methionine in exon 2 was taken instead for the start codon. The proposed change adds 44 (exon 1) and 10 (exon 2) residues to this annotation (AAG31308). For NIP1;1, 17 residues at the amino terminus were not included in the published deduced protein sequence (T05028 [37]). They have been added on the basis of our examination of the homology of these residues with those of the closely related NIP1;2. The genome annotation has already accommodated this change (CAA16760). Also, we suggest a different splice site for PIP1;4 at the carboxy-terminal end of exon 2 (AAF02782). Removing eight amino-acid residues from the genome-deduced protein renders the sequence more like other proteins in this group. This suggestion is supported by cDNA sequences for PIP1;4 (TMPC; BAA05654).

Upon close inspection, we consider three sequences as genuine pseudogenes: NIP3;1-pseudo, NIP2;1-pseudo, and TIP2;x-pseudo (Table 2; Figure 5a,b,c). ESTs or cDNA sequences for these three genes have not been reported. TIP2;x-pseudo seems to encode a truncated TIP-like protein (Figure 5a). The computer-designed first exon encoding 36 residues does not show any significant homology with corresponding sequences in other TIPs. The last exon is truncated shortly before TM6 owing to an in-frame point mutation, which introduces a stop codon at that position. Homology with other family members continues beyond this stop codon in a different reading frame for another 32 amino-acid residues.

NIP2;1-pseudo seems to be a partial duplication of NIP2;1. The predicted sequence of NIP2;1-pseudo corresponds to a 740 bp segment of NIP2;1, which includes the second half of that gene starting at residue 4 of exon 3 and extending 60 bp into the 3' untranslated region (3'-UTR) (Figure 5b). The percentage identity between the duplicated segments was found to be 96%. A single nucleotide deletion at position 105 of the homologous region results in a frameshift and the computer-deduced amino-acid sequence uses a start codon located 12 residues upstream from that point mutation. This would result in an out-of-frame translation with the NIP2;1-like coding sequence for 35 residues upstream of the position of the frameshift. A 221 bp element showing 90% identity with fragments of the Athila retrotransposon, which is found in the centromeric region of chromosome 5, is located 89 nucleotides upstream of the NIP2;1-like region of NIP2;1-pseudo.

NIP3;1-pseudo is closely related to NIP3;1 at the nucleotide level and apparently resulted from duplication. The identity with AQPs extends beyond the putative coding sequence to approximately 1,300 bp upstream and 200 bp downstream of the two genes and ranges from 84 to 94%. However, an internal segment including exons 3 and 4 and parts of introns 2 and 4 was found to be missing from the NIP3;1-pseudo sequence (Figure 5c). The computer-designed NIP3;1-pseudo last exon (132 residues) uses a different reading frame downstream of the conserved exon 5' splice site and extends beyond the conserved region for approximately 90 amino-acid residues.

One interesting facet is that each of these three genes or pseudogenes contains three hydrophobic regions that could form transmembrane helices. Also, each of the three genes retains one of the NPA domains, and thus, if translated, they might constitute shorter proteins with AQP-like functions. Such 'three-TM-spanning ORFs with AQP homology' have been reported from Saccharomyces cerevisiae, although recent reports suggested that they might be pseudogenes [46,47]. Also, polymorphisms in AQP genes, including frameshifts leading to truncated proteins, have been reported in different laboratory strains of S. cerevisiae [48]. Only transcript analysis and a functional investigation of the putative encoded proteins will answer questions about a possible function of three-transmembrane AQPs or otherwise altered proteins.

Transcript abundance

From an analysis of more than 120,000 Arabidopsis ESTs included in the public databases, a distribution of AQP transcripts emerged that provides a rough measure of their abundance, with the caveat that the tissues from which libraries have been generated are not sufficiently representative. Also, tissue specificity as it is deduced from cDNA library history does not provide information about the expression in specific cells of the tissue (for example [49]). The findings are summarized in Table 2, from which it is clear that abundant transcripts predominantly belong to the PIPs and TIPs. Most PIPs are represented in more than one tissue, and many are found in high numbers in roots (PIP2;2 and PIP2;4 seem to be expressed exclusively in roots) and in green siliques (seed pods). Similarly, most TIPs show a broad tissue distribution. The abundant transcripts for the genes TIP1;2 (230 ESTs), TIP2;2 (70), TIP2;1 (79) and TIP1;1 (180) are highest in roots. Comparatively few transcripts, not exceeding 10 for any of them, have been found for members of the NIP/SIP subfamilies. For six genes in the NIP group and three in the TIP group no cDNAs have been reported. No ESTs have been detected for the three genes that we considered pseudogenes (or 'short' AQPs). The pattern of transcription inferred from numbers of ESTs does not conform to any simple functional assignment. It seems that some AQPs carry out functions that are common to many tissues. Others seem to be reserved for more specialized tasks, and the function of yet others, represented by rare transcripts, could be required only under specific conditions or in very few cells, which might not have been sampled by the available cDNA libraries.

Promoter regions

In the absence of systematic studies targeting AQP regulatory elements, our knowledge of the expression characteristics of the AQP promoters comes almost exclusively from the representation of transcripts in cDNA libraries (Table 2). The regulation of transcripts in response to light or temperature, osmotic stress, high salinity, or water deficit, or in expanding cells in and adjacent to meristems, in developing siliques, during embryo and seed formation or in flower structures, requires further investigation. We carried out an in silico analysis of the Arabidopsis regions (1.5 kb) upstream of the coding regions, and found a few putative cis-regulatory elements. If any of these elements are functional, they seem to be for the control of circadian and/or diurnal rhythmicity, light-regulated expression and the binding of Myb-type transcription factors. In fact, diurnal expression changes in root aquaporin expression have been reported [50]. Similar elements are also present in Mesembryanthemum AQP gene promoters (J. Bennett, F.Q. and H.J.B., unpublished results). The literature on AQP expression - on over 90 AQPs in more than 20 species - supports tissue-specific expression, the presence of AQPs in cells and tissues strongly involved in water conduction, and control by hormonal signals [31,36,51]. ABA, auxin, light and circadian rhythms have been shown to affect expression of individual or groups of AQPs [6,36,52,53].

Our sequence analysis also revealed a mutator-like DNA element [54] 178 bp upstream of the protein initiation codon of the Arabidopsis TIP1;2 (Figure 6). TIP1;2 is the most highly expressed AQP among the Arabidopsis ESTs. As the MudrA transposase sequence of this element, which is adjacent to the TIP1;2 sequence, is transcribed from the opposite strand, it seems that the TIP1;2 basal promoter is either extremely compact or that intra-transposon sequences serve a regulatory function.

We note that there has been insufficient experimental work to allow a thorough comparison of AQP promoter sequences in different plant families. The protein encoded by AtPip2;2 is most closely related to McMipB from M. crystallinum (the ice plant). In this plant, McMIPB is highly expressed in the root vasculature although not in the primary meristem of the root [55]. When the promoter of McMipB is transferred into either tobacco (to drive β-glucuronidase (GUS) expression) or Arabidopsis (to drive green fluorescent protein (GFP) expression), this expression pattern is largely conserved ([56], and S. Luan, H.J.B. and D.W. Galbraith, unpublished results). There is, however, no recognizable sequence conservation between the ice plant McMipB and the Arabidopsis AtPIP1;5 promoters. Similar high protein homology exists between AtTIP1;3 and ice plant McMIPI without recognizable homology in their respective 5' regulatory regions.

Deduced proteins

From our examination of the genomic sequence, Arabidopsis AQPs are predicted to be between 240 (SIP1;1) and 323 (NIP3;1) amino acids long (Table 3). This range compares to 269 amino acids for HsAQP1 and 281 amino acids for E. coli GLP-F. Of exceptional length are the yeast Fps1p (NP013057) and the Drosophila BiB (AAF52844), containing 669 and 696 residues, respectively, in which the amino-terminal portions show homology with AQPs. In the deduced amino-acid sequences we found little variation in the length of loops between the six TMs, variations are largely restricted to the amino terminus of the proteins (Table 3; see also [57]). The PIPs were observed to have two different amino-terminal extensions, one approximately 50, the second approximately 30 amino-acid residues longer than the AQPs with the shortest amino terminus (SIP1;2, SIP2;1). Amino termini of the TIPs are predicted to be 10-15 residues long. Sequence analysis showed a higher variability in the NIP/SIP classes. Among the SIPs there is no extension, the putative start of TM1 is about five residues downstream of the amino terminus. In the NIP class, NIP5;1 has approximately 70 additional amino acids at the amino terminus and shorter extensions of around 40 residues characterize others in this group. Among those, the amino terminus of NIP6;1 has some characteristics of a mitochondrial signal peptide sequence - a series of positive charges separated by two to four amino acids. This segment could form an amphipathic helix, but no studies have been carried out on the function of this sequence.

An analysis of the amino- and carboxy-terminal regions and of loops that connect the putative TM helices did not reveal any obvious domains. One characteristic specific for the PIP group seems to be that the amino-terminal region of 20-30 residues is invariably terminated by a string of three proline residues. It is highly probable that this would affect the orientation of the amino terminus with respect to the rest of the protein but a function, if any, for such a domain is not known.

Inferences about functionally important AQP residues

The characteristics of a water channel and residues that distinguish a true water facilitator from a glycerol facilitator have previously been analyzed by comparing sequences of AQPs with known functions [58]. That comparison resulted in the identification of a few invariant or nearly invariant residues in AQPs and glycerol facilitators on the basis of 153 sequences ranging from bacteria to humans but including only a few plant AQPs. We applied this analysis to the entire Arabidopsis AQP gene family, and the results are shown in Table 4. The PIP and TIP classes were found to fall clearly in the 'AQP-type' family group (Table 4). Recently, another indicator residue has been pointed out in TM6 - the conservation of a (Y/F)(L/I/W) pair in this helix [31]. All TIPs and PIPs were found to show the pair YW and FW. Like the TIPs, the SIP group was found to contain YW residues in this position, but only further experiments can show whether these residues are sufficient to locate the protein to the tonoplast or whether they might be determinants of substrate specificity. In NIP7;1 the pair is YM, whereas the remainder of sequences in the NIP group include (Y)(L/I) in this position, which might assign them to the plasma membrane. In the case of the NIP subfamily characters emerge that are a composite of the AQP and GLP groupings, and the subfamily SIP shows even stronger similarities to both GLP and AQP types. We found that the pore-defining NPA motifs [23] are absolutely conserved among the predicted amino-acid sequences of PIPs and TIPs but vary in the NIPs and SIPs. This may also be significant in determining substrate selectivity.

Phosphorylation sites have been identified in several AQPs by Johansson et al. [31]. They identified conserved residues (indicated in bold type below) in a loop following the first NPA motif (R-K-X-S-X-X-R/K) and close to the carboxyl terminus (K-K/X-X-X-S-X-R/K-S). We analyzed the predicted amino-acid sequences of the Arabidopsis AQPs for such motifs. A site with the sequence R-K-X-S-X-X-R is present at the internal loop in most PIPs. The exceptions are AtPIP2;7 (GKXSXXR) and AtPIP2;6 (SKXSXXK). Seven out of 11 PIPs were found to contain the predicted sequence KXXXSXRS at the carboxyl terminus. Four TIPs (TIP3;1, 3;2, 4;1, 5;1) include a sequence reminiscent of the consensus at the putative internal phosphorylation site (G/R-K/R/H-X-S/T-X-X-R/T) but all TIPs lack such a site at their carboxyl terminus. No such internal site is found among the NIP/SIP groups, but five NIPS (NIP1;1, 1;2, 2;1, 4;1 and 4;2) contain the carboxy-terminal sequence K-X-X-S-X-X-K/R-S/T/R/A.

Sensitivity to mercurials is characteristic of some AQPs. In HsAQP1, the cysteine residue that confers sensitivity is C189, adjacent to the second NPA motif, but no plant AQP shows a cysteine in this position although effects of mercury have been documented in plants in many studies. The cysteines responsible for mercury poisoning of water flux in AtTIP2;1 (C116) and AtTIP1;1 (C118) [59] were found in the predicted sequences of all Arabidopsis TIPs. This residue is located in TM4, downstream of the first NPA motif. A cysteine was found in all PIPs at a similar position along the deduced protein sequences. It does, however, not occupy a similar three-dimensional position (assuming that the TIP and PIP TM are similar); this means that the cysteine in PIPs would point in a different direction from that in TIPs. In fact, other studies have shown that some PIPs are mercury-insensitive [59,60].

A combined NMR and modeling approach to analyzing water movement through plant tissues

Typically, AQP water-channel activity is tested by the oocyte swelling assay [18], in which Xenopus oocytes expressing AQPs are exposed to hypo-osmotic shock and the subsequent water influx is measured. Apart from the caveats mentioned above, the level of expression of AQPs, that is, the number of active AQPs in the oocyte plasma membrane, is often not known and some AQP might even not be targeted to this membrane in oocytes. These limitations can, however, be addressed by measurements of membrane localization and oligomerization state [61,62]. TIPs have been reported to be more 'active' in these experiments, which suggests that they might have a higher conductance in vivo and/or that their structure is more compatible with the oocyte expression or targeting machinery. Other techniques use artificial membranes into which AQPs are embedded, yeast cells in which AQP is overexpressed [63] or isolated plant protoplasts [12].

Important advances in studying water movement at the cell and tissue level include the use of pressure probes for roots and for individual cells [64]. The effects of a single AQP have been addressed by antisense experiments that measured changes in xylem pressure and compared water permeability of protoplasts from control and transgenic antisense AtPIP1;2 (PIP1b) plants [12]. Because the movement of water through plants and plant tissues involves several barriers, including apoplastic, symplastic and transcellular ones, a single set of measures of overall hydraulic conductivity, although useful, is not sufficient for determining how particular AQPs function in planta.

Although informative, the results of pressure-probe experiments require careful interpretation, and independent nondestructive ways of measuring water fluxes are desirable. Such methods are available but have been little used so far for analyzing AQP functions. Nuclear magnetic resonance (NMR) imaging (MRI) allows the distribution of tissue water to be quantified, and also allows the mapping of flow and of diffusion [65,66,67,68,69]. The potential for directly discriminating intracellular and extracellular water signals in plant cells and root tissues has been shown [70] and this opens the door to spectroscopic and imaging-based measurements of water flow in vivo ([70,71] and J. Rosenberg, Y.S-H., H. Wang and H.J.B., unpublished results).

The separation of intracellular from extracellular water NMR signals is restricted to roots or other tissues, such as some stems, which have elongated cells oriented in parallel [70] and is illustrated in Figure 6 for maize roots. The assignment of the two signals to represent the total intracellular water and the total extracellular water, including the apoplast, has been confirmed by the selective imaging of intracellular and extracellular water using this approach [71]. NMR spectroscopic applications enable the rates of osmotic shrinkage and swelling to be measured in situ and also enable measurement of transport of small molecules and ions in vivo [70].

We found that the total intracellular water signal (see Figure 6) shrinks in response to osmotically driven efflux when the tissue is exposed to hypertonic solutions (with salt or sucrose in the perfusate spectra; data not shown). The advantage here is twofold: the isolation of the intracellular signal removes apoplastic tracer from the data; and, in contrast to other NMR methods using shift reagents, the movement of any small molecule that gives an NMR signal can be followed (this includes any molecule containing carbon, nitrogen or nonlabile hydrogen atoms).

It has also been reported that the method illustrated in Figure 6 allows diffusional water fluxes to be measured by isotopic or magnetic labeling [70]. Diffusional gradients are not the ones that drive water movement in plants, but diffusional measurements, especially in combination with hydraulic ones can be very informative about the barriers responsible for regulating water fluxes. The in vivo NMR isotope flux data available from this approach were obtained by following the time course of the intracellular ²H₂O signal after switching between unlabeled buffer and buffer containing tracer levels of added D₂O (see Materials and methods). The spectra giving rise to Figure 7a for hydroponically grown maize have a high time resolution (10 seconds) and high signal-to-noise ratio. This means that rapid processes can be followed and a complete influx and efflux dataset is collected from one root sample, with each sample acting as its own control. This means that simple mathematical models, embodying different hypotheses about the location and magnitudes of barriers to water movement through the root, can be tested by using them to generate fits to the isotopic flux data. An example of a 'first generation' approach to modeling water movement is shown in Figure 7b. The fluxes generated by this model were shown to fit kinetic data and such a fit is shown in Figure 7a (see Materials and methods), where it is apparent that the data were well accounted for by a model that has separate barriers to water movement at the exodermis, endodermis and between one cell layer and the next. In the model illustrated in Figure 7b, the relative volumes correspond to the anatomically determined locations of exodermis, endodermis and number of cell layers in the roots of maize seedlings. This simple model was found to be sufficient to account for widely differing time courses of water movement from plants grown under a range of conditions that affect water transport (data not shown). As cell-to-cell water movement is likely to be controlled by the permeabilities of the plasmalemmal and tonoplast membranes, the parameter corresponding to this permeability potentially provides a noninvasive measure of membrane water conductivities in intact tissue. In combination with transcript and protein expression data, which also monitor cell specificity, NMR therefore has the potential to test ideas about aquaporin function in planta by combining localization information with modeling of flux measurements from control and knockout plants and with plants grown under conditions that alter both water conductivity and AQP expression.

Intracellular locations of AQPs

The original definition of the PIP and TIP subfamilies was their assumed location in the plasma membrane and tonoplast, respectively. When antibodies were used to locate proteins of the TIP group, the signal was always confined to tonoplast membrane fractions [72]. PIP localization seems less well defined. Daniels et al. [73] located a PIP-family protein to the plasma membrane in Arabidopsis. For M. crystallinum, Barkla et al. [43] detected PIPs to a small extent in the plasma membrane but mostly in a vacuolar fraction in continuous sucrose gradients or, more likely, in a membrane fraction with a density similar to tonoplasts. This is not surprising considering that mammalian AQPs have been documented as cycling between the plasma membrane and internal vesicles, a process that seems to be under hormonal control [74,75]. Such mobility may also be at the basis of the distribution of different TIPs in distinct plant vacuoles [61,76,77]

Functions of AQP

We have outlined how the genome of Arabidopsis can provide new information, taking the aquaporin gene and protein family as an example. There are, however, limits on what can at present be deduced from gene sequences alone. The fundamental questions of whether the primary role of all or most AQPs is in increasing water conductivities, and what physiological functions such water permeability increases may serve in plant function, are not settled. They can only be addressed by functional measurements. The straightforward view that most aquaporins in plants are there to regulate water flow, and a subset may facilitate the movement of glycerol or other small molecules, is the most widespread general interpretation of the physiological and functional assays outlined at the start of this article. This dogma is expounded more thoroughly elsewhere [6,30,31,32,36]. This view, in fact, conceals a number of quite different roles for water-channel activity, involving the regulation of water flux and homeostasis at the subcellular, transcellular, tissue and whole-plant levels. Therefore, there are two levels at which functional questions must be addressed for each AQP. One is whether each AQP functions as a water channel, and it seems that a significant number of PIPs, for example, do not in fact greatly increase the water conductivity of the cell membranes in which they reside (at least not in oocytes; see for example [55,78]). In reality we are ignorant about the natural substrates for most AQPs.

The second level of question that requires nongenomic information is: if water-channel activity is significant for a particular AQP, what is the role of this activity in a whole-plant context. A simple example concerns the location of gene expression. The simple statement that a particular gene is expressed in roots, for example, conveys insufficient functional information, as different AQPs have different distributions within one tissue [44,72]. An AQP expressed in the epidermis is unlikely to function in vascular flux. Although our understanding of promoter and protein domains that determine expression and localization is growing, our ability to deduce from genomic sequences which cell types are likely to express a particular gene and where the product will reside under what conditions is still rudimentary. Thus direct determination of the location of each AQP within tissues is still required to understand its function in the plant.

The water-channel view of AQP function requires us to believe that there may be something about some PIPs that prevents them displaying their normal water-channel function in oocytes. This and other considerations arising from the imperfections of oocyte expression systems [79] call for functional measures of AQPs to be made primarily in plant cells or tissues. Assays on plant protoplasts, roots and whole plants are obviously important, and here the question is how the contribution of an AQP is to be determined. Such measurements generally rely on mercurials to perturb AQP function, but mercurials can act in a nonspecific way on membranes and proteins, and not all AQPs are mercury-sensitive, so this is not sufficient.

A path forward: reverse genetics and biophysical investigations

The use of gene knockout and/or antisense technologies is particularly appealing for plant AQPs. In animals, this approach has been used to reveal which AQPs and combinations thereof are, and are not, crucial for water fluxes in vivo in mice lacking the genes for AQP1 and AQP3 [80]. In the one case where it has been applied, lowering AQP expression by AtPIP1;2 (AtPIP1b) antisense expression resulted in plants with root systems that were larger but pressure probe measurements indicated the same overall xylem pressure as the smaller root systems of wild-type plants [52]. Protoplasts from these antisense plants showed lower water permeabilities [12].

We believe that applying a combination of biophysical measures of conductivity to transgenics involving each of the AQPs, and combinations of several AQPs, represents the most promising approach to the fundamental unanswered questions about general and individual AQP functions. By including whole-plant water relations and also metabolite measurements, the use of transgenic plants, preferably of the 'gene knockout' type, will yield a greater understanding and prevent errors of interpretation that could arise from simply correlating single-gene knockouts with changes in hydraulic conductivity parameters.

Only by addressing in a systematic way the functions of each AQP in planta will it be possible to understand such simple but puzzling issues as the reason behind the large number of AQP genes in plants. The number of AQPs in the Arabidopsis genome - 35 - compares with the presence of, for example, 15 potassium channels and 15 proton/metal antiport systems in the Arabidopsis genome. More detailed comparison with animal systems only increases the puzzle. In humans only AQP1 appears to be widely distributed in different tissues (although several other AQPs are expressed as rare transcripts in various tissues), whereas the data in Table 2 show that, in plants, most AQPs are expressed at significant levels in multiple tissue types. We suggest that the reason for the large number of plant AQPs will turn out to be that multiple genes represent multiple functions, of which water permeability is only one. Thus we postulate that plant genomes evolved more AQPs than animals because in plants they serve functions that are either not required or are met by other systems in animals.

Such functions may include metabolite permeation. As with the water permeability of most AQPs, permeability to a particular metabolite does not mean that this is a meaningful function in planta. Thus the permeability of an AQP to ribitol and xylitol, as found for a bacterial glycerol facilitator [23], cannot be significant in cells where these metabolites are not present. Indeed permeability in one species may have a very different role in another. Thus there is good evidence that glycerol permeability is important in yeast cells [81,82] in the secretion of this polyol under stress conditions, but in higher plants that do not accumulate glycerol as an osmoprotectant nor excrete it in large quantities, there is no apparent function for increasing glycerol permeability. One might speculate that by increasing membrane permeability to polyols and/or other metabolites, some plant AQPs could be involved in long-distance apoplastic transport or signaling rather than in the bulk secretion of, for example, glycerol. The discovery of such possible functions will require in planta examination of transgenic plants and an extension of analyses beyond highly selective water and small-molecule permeability assays in heterologous systems and protoplasts.

Biophysical measures of water movement in planta - whether based on pressure probes, NMR or other methods - as well as methods for determining localization and expression and for generating transgenics are available. A relatively neglected area is the development of quantitative anatomical models of water movement for interpreting flux measurements [83]. Such models are needed for testing the potential contribution of AQPs to regulating water content and transport, and indeed for determining which of the range of plausible but untested ideas about AQP function may or may not apply to different membranes, cells, tissues, plants and environmental conditions. The combination of localization of AQP proteins and using anatomically-based models, has the potential to allow function to be assigned to individual AQPs using mutants and transgenics in which expression of AQPs (singly or in combination) is reduced or eliminated.

Studying genome sequences yields unique and useful information about unstudied and understudied genes. The aquaporin gene family in Arabidopsis is an interesting example of the current and potential limitations of this approach. Full sequence comparisons allow the assignment of the AQPs into subfamilies to be placed on a firm footing, and have functional implications for previously unknown genes. While we know about AQP sequences, structure, expression, post-translational modification, permeability properties and subcellular location from some case studies, our ability to extrapolate from gene sequence to functional properties for unstudied members of the same family is still limited. One limitation is that we cannot yet extrapolate reliably from sequence via three-dimensional structural features to conductivity, selectivity and other protein properties. This inability extends to regulatory regions that determine place, time and extent of transcript expression, and to protein characteristics that determine half-life, intracellular localization and membrane trafficking. This type of limitation will decrease as our understanding of sequence-structure and structure-function relationships improves and as our knowledge of expression and cellular targeting in plants expands. This means that genomic information will become more useful in the coming years and analyses such as the one presented here are likely to become common and more useful.

Irrespective of our steadily improving predictive tools for in silico analyses, fundamental questions of the physiological roles of individual genes will continue to require detailed functional assays. Here too, genomic information is important because, as we have argued above, assigning physiological function via transgenic reduction or removal of gene expression requires sequence information for precise targeting. Here, the AQP family provides an ideal opportunity. This is a set of genes whose functions are intuitively perceived as important, much isolated information has been accumulated, yet their function is far from being understood. Also, AQPs pose challenges because the family is likely to contain genes with divergent and unique biological roles in the plant. Indeed it is likely that in many cases individual members of the family have different roles, depending on when and where they are present in the plant. Clearly, this level of understanding will require a combination of biophysical measurements - which we think must be nondestructive, real-time and experimentally flexible - transgenic technology, knockout or precision antisense experiments, and modeling of plant function.

Genome and transcript analysis

Arabidopsis thaliana (ecotype Columbia) aquaporin coding sequences were identified by searching the Arabipdosis genomic data at NCBI [84] with the BLAST algorithms [85]. Intron-exon splicing sites were verified visually using available cDNA sequences and/or corresponding ESTs. The Arabidopsis Information Resource (TAIR) website [86] was used to localize Arabidopsis aquaporin genes on the chromosomes. Sequences were aligned with the pile-up program of the GCG software (Wisconsin Genetics Computer Group package). Tree building was performed with ClustalX [40] and drawn with the TreeView software [41]. Transmembrane regions were determined with the transmembrane prediction program TMpred [87,88].

Growth of seedlings

Maize seeds (FRB-73) were germinated for 3 days on trays on absorbent paper wetted with 0.1 mM CaSO₄ at 28°C in a humidified incubator. Seedlings were then transferred to hydroponic growth on 0.5 cm thick Styrofoam sheets (0.5 cm diameter holes) floating on aerated growth solution. The primary root of each seedling was threaded through a hole in the Styrofoam sheet into the growth solution. Hydroponic growth was under continuous illumination at 28°C in humidified air and growth solutions contained 0.1 mM CaSO₄.

Preparation for NMR

Tips 5 mm long were excised from each primary root and discarded. The next 25 mm of primary root was then cut and briefly vacuum infiltrated under reduced pressure to improve magnetic field homogeneity of in vivo spectra. Root segments were placed in 10 mm NMR tubes with their axes aligned parallel to the tube axis and perfused with aerated buffer solutions containing 50 mM glucose, 1 mM Ca²⁺, 10 mM potassium-gadolinium EDTA, 0.5 mM EDTA, 5 mM methylethylsulfonate (MES) at pH 6 with or without 3% D₂O. The gadolinium (Gd) was used to separate intracellular from extracellular signals [70] and is complexed to prevent toxicity. The composition of the perfusion medium was designed to keep the free [Gd³⁺] below 1 μM, while having [Ca²⁺] above 0.2 mM and below 1 mM. No toxicity to maize root tissues was observed by in vivo ³¹P NMR spectroscopy of root tips or segments exposed to the perfusion medium for several hours, and the growth of excised root tips and roots of intact seedlings was the same over 24 h in this medium as in medium without Gd or EDTA. The perfusion system used has been previously described [70]. Perfusion rates of 30-40 ml per min were used.

One-dimensional-pulse ¹H and ²H NMR spectra were acquired as previously described [70] using a Unity 400 Varian spectrometer with a 10 mm broadband probe. For tracer experiments, time courses of 150-300 ²H spectra were acquired at 1, 10 or 15 sec intervals. The intensities of intracellular ²H signals were measured from these spectra and normalized to final equilibrated intensities (for influxes) or initial intensities (for effluxes).

Model for the analysis of water flux data

The model illustrated in Figure 7 assumes that water diffusing radially through the root encounters several barriers. The three parameters of this model correspond to rate constants (or permeabilities) of the exodermis and the endodermis, as in model 2, with additional barriers for movement between cell layers, similar to model 3. The rate of change of concentration for the nth compartment is given by a set of n equations

in which C_n is the concentration in layer n, k_n is the rate constant between the nth and the (n - 1)th cell layer. The permeability for water movement between successive cell layers is assumed to be the same for all layers except for those for movement from the external medium into the root (across the exodermis) and movement from the eighth to the ninth layer, which are separated by the endodermis. Therefore the all k values except for k₁ and k₉ have the same value scaled by dividing by the relative radius of each cell layer. This scaling is because the rate constant for crossing a membrane or other barrier is equal to the permeability of that barrier multiplied by the area-to-volume ratio, and the area-to-volume ratio for an elongated cylinder is 2/(radius). Solution of this set of equations was implemented using a spreadsheet computer program.