Plant aquaporin selectivity: where transport assays, computer simulations and physiology meet

Abstract

Plants contain a large number of aquaporins with different selectivity. These channels generally conduct water, but some additionally conduct NH₃, CO₂ and/or H₂O₂. The experimental evidence and molecular basis for the transport of a given solute, the validation with molecular dynamics simulations and the physiological impact of the selectivity are reviewed here. The aromatic/arginine (ar/R) constriction is most important for solute selection, but the exact pore requirements for efficient conduction of small solutes remain difficult to predict. Yeast growth assays are valuable for screening substrate selectivity and are explicitly shown for hydrogen peroxide and methylamine, a transport analog of ammonia. Independent assays need to address the relevance of different substrates for each channel in its physiological context. This is emphasized by the fact that several plant NIP channels, which conduct several solutes, are specifically involved in the transport of metalloids, such as silicic acid, arsenite, or boric acid in planta.

Introduction

The membrane that surrounds the cell and cellular compartments not only separates life from death, but also provides a low-dielectric barrier for charged and polar molecules. Classical studies have provided a solid basis for the lipid permeability of water and all physiologically relevant solutes [1, 2]. In addition to the intrinsic passive bilayer permeability, water and small solute flux across membranes is facilitated by aquaporins in all organisms. Their importance for human physiology is underscored by the fact that their loss can result in diseases [3]. Similarly, reduced growth or viability is associated with the loss of aquaporins in plants under laboratory conditions, and the relevance of these channels for survival in native environments is only beginning to be understood. Several excellent recent reviews have covered the function of plant aquaporins and their associated physiology [4, 5] and the in planta effects of the genetic modulation of aquaporin expression [6]. The transport of unconventional substrates, mostly by microbial and mammalian aquaporins, has also been summarized [7].

In this review, we will give an update on aquaporin selectivity with a focus on plant proteins. In order to understand their potential physiological function, we will discuss experimental findings and compare them with molecular dynamics (MD) computer simulations. Although we will discuss the permeation of small solutes, such as NH₃, CO₂ and O₂, we will not use the term “gas channel” in this context. This term may be misleading, as it implies that the surrounding waters have no impact on the diffusion of solutes. However, water–solute interactions and solute-induced water–protein interactions participate to determine channel selectivity. Furthermore, a “gas” defines a physical state of matter, but the volatile molecules referred to are dissolved in liquid water, and occur as gases only in their pure form under ambient conditions.

Plant aquaporin structure

Plants, such as rice, maize, and the model plant Arabidopsis, encode 31–39 aquaporin family members in the genome [8], while humans have 13. More than ten different high-resolution structures of aquaporin homologs have been determined, including an open and a closed structure of SoPIP2;1 from spinach [9]. All structures show an extraordinary conservation in their overall transmembrane architecture, but differ in the loops exposed to the aqueous solutions. Each monomer forms a hydrophobic pore and aggregates to tetramers. A monomer is composed of six tilted, membrane-spanning helices that are connected by five loops (A–E) (Fig. 1a–c). In two of these loops (B and E), short helices dip back into the membrane from opposite sides and contain two NPA (Asn-Pro-Ala) motifs that are conserved in most aquaporins and meet at the center of the membrane. The aromatic/arginine (ar/R) region is located ~7 Å from the center towards the external pore vestibule, and is formed by residues from TM2, TM5 and two residues from loop E. These residues define the most constricted site of the channel, which is ~2 Å in diameter in the most water-selective aquaporins, such as SoPIP2;1 (Fig. 1d). The pore diameter is ~1 Å larger in glycerol-conducting aquaporins (also termed aquaglyceroporins), such as the E. coli glycerol uptake facilitator GlpF. With only a single exception, mammalian AQP6 that also transports anions, all family members exclusively facilitate the passive flow of small, uncharged solutes and/or water across membranes. Although the term aquaporin was originally used exclusively for the water-selective channels of the family, we will follow the now common nomenclature that adopted this term for all family members.

Fig. 1.

Molecular structure of aquaporin monomers and the ar/R constriction. a Side view of the open SoPIP2;1 pore in ribbon representation. Helices 1 and 2 were removed to open the view into the pore. The aromatic/arginine (ar/R) residues, F, H, T, and R are explicitly shown in green, the two NPA-stretches in yellow. b Schematic two-dimensional representation of the structure. Ar/R residues are shown in diamonds, whereas residues involved in gating are shown in circles. c Top view: the four residues of the constriction site (ar/R) are shown with their van der Waals radii. d Pore radius of SoPIP2;1 (blue) and a mutant that has the ar/R filter as NIP5;1 (violet) calculated using the program HOLE. The structures were derived from the closed pore (PDB accession number 1Z98), in which Phe180–Leu200 was replaced by the open structure (2B5F). e Monomer of the NIP5;1-like mutant with pore calculated by hole as green mesh. f Residues in the constriction of SoPIP2;1 with TIP-like ar/R constriction. A histidine residue that replaces the phenylalanine bends into the pore (red), as long as no compensatory mutation in an adjacent threonine (to glycine) is made (green)

Plant aquaporins are often categorized according to their phylogenetic relationship, which by coincidence groups the family members according to their subcellular localization. The PIP1s (five members in Arabidopsis) and the PIP2s (eight members) are very closely related by sequence and are plasma membrane intrinsic proteins. The TIPs are tonoplast intrinsic proteins (ten members), small basic SIPs are localized in the endoplasmic reticulum (three members) [10] and most NOD26-related NIPs are localized to the plasma membrane (nine members). When heterologously expressed, all the plant PIPs, TIPs, SIPs, and NIPs (except for AtNIP6;1, [11]) that could be functionally expressed, were permeable to water.

Water conduction

Water has an exceptional importance for plants, and the transpiration from the leaves of a large tree can approach 200 l per day. Plants are exposed to rapidly changing hostile environments, where periods of drought can be eventually followed by rainfall, and water logging of roots may suddenly occur. Many plant species are particularly efficient in coping with changes in water supply, to keep their metabolism intact. This is reflected by the various and complex levels of aquaporin regulation, which have been reviewed in more detail elsewhere [4–6].

Initial studies observed that many PIP2 members strongly increased the water permeability in the plasma membrane of heterologous expression systems, such as oocytes, but many PIP1 members exhibited low or no transport activity. PIP1s from maize form heterotetramers with PIP2s in oocytes [12] and in planta [13] and require these interactions to traffic to the plasma membrane. In the absence of PIP2s, the PIP1 proteins remained trapped in intracellular compartments [13]. Genetic interactions between PIP1 and PIP2 members were also suggested by a study in Arabidopsis, where members of both PIP groups were separately silenced, but the effects were not additive [14]. In this work, a lower root hydraulic conductivity was observed with individually silenced PIP1 or PIP2 isoforms, and the same reduction was observed when both families were suppressed by antisense [14].

In several studies, the PIP expression of different species was genetically reduced. The osmotic water permeability or root hydraulic conductivity was consistently decreased, but only to a limited extent, probably due to some redundancy. Early research with transgenic tobacco plants over-expressing aquaporins indicated that plant growth was enhanced by additional channels. These studies identified that the native, rather low symplastic water transport, was limiting optimal growth even under favorable conditions [15]. The strong AtPIP1;2 expression in tobacco was not beneficial under salt stress and was deleterious during drought stress [15].

In several species, the expression levels of PIPs have been documented, together with their transcriptional changes upon stress. In a few cases, these analyses additionally quantified the different protein levels and revealed a rather consistent down-regulation of PIPs in roots with salt [16], drought [17], or cold stress [18, 19]. As exceptions, AtPIP1;4, AtPIP2;5, and AtPIP2;6, which were more abundant in leaves and/or flowers, were unchanged or more strongly expressed under drought [17]. Upon salt treatment, the transcripts and proteins of all the significantly expressed PIPs and TIPs decreased in a coordinated manner. PIPs and TIPs were subcellularly re-localized to internal membranes or to vacuolar invaginations [16]. All these treatments reduced the water permeability across the plasma membranes, confirming the primary involvement of PIPs in water homeostasis in planta.

Since high-resolution structures are not available for all (or all classes of) aquaporins, it is common practice that important residues are deduced from sequence alignments. Homology models are frequently viewed with some skepticism, since biologically important structural details are often found in disparate loops that vary between individual homologs. However, surprisingly accurate predictions on membrane protein structures have been revealed by homology modeling, especially in the catalytically important membrane regions [20]. This modeling gave initial insights into the minimal pore diameters of aquaporins from Arabidopsis [21], and later from other plants [8]. Based on the available structures from many species, a model of SoPIP2;1 showed an excellent agreement with the X-ray structures in the transmembrane region ([8]; Dynowski, unpublished).

All PIPs from Arabidopsis, maize and rice have an identical narrow ar/R selectivity filter (Fig. 1; Table 1). Structural modeling identified that almost the same residues line the pore, which suggests that the overall pore selectivity of all PIPs is virtually identical, irrespective of whether they are PIP1s or PIP2s. Many studies confirmed the large water transport capacity of PIPs, but experimental evidence suggests that individual PIPs may additionally transport other solutes. For example, NtAQP1, a PIP1 member, transported as much glycerol as the bacterial glycerol channel GlpF in oocytes [22]. This is quite surprising, as mutational evidence from other aquaporins, molecular dynamics computer simulations, and the fact that glycerol transport was exclusively found with aquaporins that have a larger pore diameter, contradict the view that PIPs can efficiently transport glycerol [23–25].

Table 1.

Pore residues in plant PIP, TIP, and NIP channels at the ar/R filter

Position	H2	H5	LE1	LE2	NH3	H2O2	PIP mutant NH3	PIP mutant H2O2
At, Zm, Os: PIP1, PIP2	F	H	T	R	−	+	−	+
At, Zm, Os: TIP1	H	I	A	V	+	++	−	−
Zm, Os: TIP2, AtTIP2, 3, 4	H	I, M, V	A, G, S	R	+	+	−	+
ZmTIP4, 5, OsTIP4, 5	H, Q, T	S, T, V	A	R			−	+
AtTIP5;1	N	V	G	C			−	+
Zm, Os: NIP1, At NIP1, 2, 3, 4	W	I, V	A	R	+	+	+	+
OsNIP3, ZmNIP3, AtNIP5, 6, 7	A	I, V, A	G, A, P	R			±	+
ZmNIP2, OsNIP2	G	S	G	R
OsNIP4;1	C	G	G	R
PfAQP	W	G	F	R	+
hAQP1	F	H	C	R	±
hAQP1 mut (Ref. [42])	F, A	H, A	C	V	+
hAQP3	F	G	Y	R	+
hAQP7	E	G	Y	R	+
hAQP8	H	I	G	R	+
hAQP9	G	A	C	R	+

Aquaporins for which NH₃ and H₂O₂ permeability has been experimentally determined are indicated with plus. PIP mutant constructs: mutations were introduced into AtPIP2;1 [52]. The selectivity filter was identical to TIPs and NIPs, but failed to confirm the proposed permeability of native channels. The amino acids are given in single letter code

H2 Helix 2, H5 helix 5, LE1 loop E1, LE2 loop E2

Plant plasma membrane aquaporin gating is complex

The molecular mechanism how plant aquaporins switch between open and closed states has recently been identified in PIP2s. A conserved histidine in the second intracellular loop (D) plays a key role and is protonated upon a drop in the cytoplasmic pH [26]. This acidification leads to a conformational change of loop D, which contains several hydrophobic residues that fold into a cavity near the cytosolic entrance, to block water permeation in the closed state [9]. This closing involves only minor structural changes in the overall pore structure and is restricted to the cytoplasmic vestibule, where it also involves the binding of divalent cations such as Ca²⁺, Mn²⁺ or Cd²⁺ [27]. In the closed conformation, a divalent cation is coordinated by charged residues from loop D, the N-terminus and the first cytoplasmic loop [9]. The inter-dependence of the open state on the pH, divalent cations and phosphorylation of two serine residues was supported by MD computer simulations [9]. Recent mutational analysis confirmed the importance of the histidine in loop D for intracellular, pH-dependent opening and (partial) closing in other PIP aquaporins, but showed that the gating mechanism appears to be more complex [27, 28]. The importance of phosphorylation on the pH-dependent modification of water permeability could not be confirmed by transport assays with PIP channels that were mutated in the two key serine residues [28]. NIP and TIP channels are also reversibly phosphorylated, but at different positions, and whether they are gated is less clear.

To obtain a stable open structure for computer simulations, phosphorylation was necessary [9]. A modified high-resolution pore of SoPIP2;1 in which only loop D from the low-resolution open structure was incorporated into the high-resolution closed structure (at 2.1 Å) yielded a stable structure, but it was necessary to phosphorylate the protein at the known serines (Fig. 1b) to maintain the open status of the chimeric structure [29].

Molecular dynamics simulations on aquaporins

Molecular dynamics simulations have become standard methods to understand lipid and protein function at atomic resolution. As early as high-resolution structures of aquaporins became available, the mechanism of water conduction was studied by MD simulations [23, 30]. The setup for these simulations typically involved a box containing the whole tetramer, embedded in a simple membrane and surrounded with water (Fig. 2a). The enormous increase in computer power and a better knowledge of the principles of membrane protein structure recently allowed the study of very detailed aspects of aquaporin function and yielded an atomic picture of the molecular mechanisms involved.

Fig. 2.

Setup of MD simulations on tetrameric, open SoPIP2;1 and energetic pore profile. a Side view of the tetramer embedded in a lipid bilayer with the individual starting positions of the five waters and H₂O₂ explicitly shown (arrows). Two monomers, in blue and orange, are seen in the foreground. b The energy profile deduced from the residence time for water (blue) and hydrogen peroxide (red) along the SoPIP2;1 pore. The variance along the z-axis of the entrance, selectivity filter (SF, ar/R), NPA-region and exit on the cytosolic side is also shown. Modified after [29]

Molecular dynamics simulations describe the interactions of all atoms within a system using a relatively simple empirical potential function, a compromise between accuracy and computational efficiency. This potential function accounts for all the atomic interactions, such as stretching, bending, torsional interactions, van der Waals forces, and long-range electrostatic Coulomb interactions. From this potential function, the forces on all atoms in the system, often more than 100,000 atoms in total, are calculated and integrated in time, using time steps in the order of 10⁻¹⁵ s. This generates a trajectory that contains the diffusive movement of all lipids, proteins, waters, and other solutes, over millions of time steps. The further analysis can then provide insight into the fundamental mechanisms of the particular system and its thermodynamic properties.

Since ‘real-time’ MD simulations monitor rapid processes within a few ns at atomic resolution, spontaneous conduction attempts or full water permeation events in aquaporins were visualized and revealed molecular interaction energies and effects on the surrounding water [23, 30]. The simulations identified that the dynamic hydrogen bond interaction network in bulk water was efficiently replaced by equivalent hydrogen interactions donated by the pore. A favorable water-pore arrangement compensated for the energetic cost of water transfer into and through the hydrophobic pore. As a consequence, water moves into and inside the pore almost as fast as in bulk solution. Although individual permeation events were fast, net fluxes along an osmotic gradient were comparably small. Their calculation required further considerations, for example osmotic pressure in non-equilibrium simulations or even external steering forces [31]. The quantitative values of water permeability, Pf, for simulations from different groups yielded 7.1 × 10⁻¹⁴ [32] and 7.5 × 10⁻¹⁴ cm³ s⁻¹ [30] for the single channel osmotic permeability of mammalian AQP1, which is in excellent agreement with the experimentally determined 5.4 × 10⁻¹⁴ cm³ s⁻¹ [33].

Despite that all molecules move in MD simulations, a stable high-resolution protein structure, that will not collapse, is required for meaningful calculations. Simulations with membrane-embedded mammalian AQP1 at lower resolution (3.8 Å) had identified instabilities of critical residues lining the pore and failed to predict permeation speed correctly [31]. A study investigating the stability of various X-ray structures and homology models revealed that the low resolution structure of mammalian AQP1 did not provide sufficient accuracy for full atomic simulations [34]. Instead, a homology model produced an acceptable description of water inside the channel [34]. The lower resolution open SoPIP2;1 structure (3.9 Å) also did not have sufficient stability for reliable MD simulations [29]. Homology models of selected aquaporins, that were based on a large number of templates, were more stable during simulations than low(er) resolution structures, such as that of the open SoPIP2;1 (Dynowski, unpublished). This suggests that simulations should be possible even for those plant aquaporins, for which high-resolution X-ray structures are not yet available.

To reduce calculation times and for a quick routine test of their selectivity, simulations were also performed on monomers. These were compared with those on tetramers, both embedded in lipid bilayers. Both types of simulations gave a similar outcome, indicating that the central tetramer vestibule was of minor importance for the flux of selected solutes [29]. In the computer simulations, usually POPC and/or POPE (1-palmitoyl-2-oleoyl-phosphatidylethanolamine)-lipid membranes without sterols were used, and direct full conduction events across the lipid membrane were observed for water and other solutes [25, 29, 35, 36]. Such model membranes, however, probably do not adequately reflect the complex properties of a native plant membrane. The artificial simple membranes can be expected to confer lower diffusion barriers for solutes than the native membranes.

Despite their success, the atomic simulations should be treated with caution and their limitations should not be ignored. MD simulations are restricted to rather fast phenomena; this precludes the modeling of relatively slow biological processes and rare events. More recent MD simulations have expanded the simulation times to more than 150 ns, but unusually high solute concentrations or other simplifications were employed to simulate spontaneous permeation events [35]. Furthermore, the lipid/protein complex under study must still be limited to the most important functional units, and starting configurations can bias the system in an undesirable way. Different force field parameters and different computational approaches, such as non-equilibrium and equilibrium simulations, have been used and may have biased the results. Finally, the pH and the transmembrane potential are difficult to implement and are not considered in most studies. Classical MD simulations neglect quantum mechanical effects and therefore do not consider proton diffusion, chemical bond formation, and breakage. On the other hand, the success of MD simulations to quantitatively reproduce experimental parameters, even when different force field parameters were used, strongly validates these computational methods.

H⁺ exclusion by aquaporins

In MD simulations, waters were arranged in a bipolar manner inside the pore, meaning that the average dipole orientation reversed in the center of the membrane, close to the NPA region. This arrangement initially suggested that proton transfer might be blocked by a mechanism that disrupts a potential Grotthuss-like proton wire transfer inside the narrow pore [23]. However, this proposal was not confirmed by more sophisticated analysis. Simulations revealed that the large electrostatic barrier impeded H⁺ flux [37–40], together with the dehydration cost of moving a proton into the narrow hydrophobic channel [40, 41]. In fact, the large electrostatic barrier excluded protons from permeating even in the presence of an intact proton wire [40, 41].

Voltage clamp measurements on mammalian AQP1 mutants revealed that the arginine of the ar/R filter, which is conserved in most plant aquaporins, had an important impact on proton exclusion [42]. A proton leak was associated with a large pore diameter and the replacement of the arginine from the ar/R filter [42]. The fact that such unconventional mutant pores transported H⁺ was convincingly reproduced by MD simulations using sophisticated algorithms [41]. Interestingly, some TIP isoforms, namely TIP1s and AtTIP5;1, lack the arginine that is critical for proton blockade in AQP1 (Table 1). However, TIP1;1 has been thoroughly investigated with electrophysiology and no H⁺-leak was found [43]. There is currently no experimental evidence that any plant aquaporin efficiently transports H⁺. The missing arginine in TIP1s may be compensated by neighboring residues and a narrower pore compared to AQP1 mutants. Proton barriers are of exceptional importance in plants, as the acidic apoplast and vesicular/vacuolar lumen are maintained at approximately 2 pH units below that of the cytosol. PIPs, NIPs, and TIPs are thus exposed to a ~100-fold difference in H⁺ concentration, which is actively established by the energy consuming proton pumps. We expect that an evolution-driven process efficiently selected for pores that exclude proton transfer across TIPs, while leaving the transport of other substrates intact.

Membrane permeability

The channel-mediated water and solute passage by aquaporins can increase the permeability of a biological membrane by up to two orders of magnitude, depending on the expression level. Since individual aquaporin tetramers within a membrane patch usually occupy only a small fraction of the total membrane area, they will only increase the total membrane permeability when their transport is significantly larger than that of the small lipid membrane area that they occupy.

The intrinsic membrane permeability of a given solute is proportional to its oil–water partition coefficient, its molecular size, and the diffusion coefficient in lipids [1]. For molecules of equal size, the diffusion through the membrane is quicker for those with the greater solubility in lipids. The permeability also depends on the lipid composition and the permeability coefficient for water (P_f) of an artificial lipid bilayer membrane composed of 1-palmitoyl-2-oleoyl phosphatidylcholine (POPC) is P _f = 7.2 × 10⁻³ cm s⁻¹ [2]. Urea diffuses less well across such a membrane (P _urea = 1.3 × 10⁻⁶ cm s⁻¹), while ammonia permeates more readily,  = 3.6 × 10⁻² cm s⁻¹ [2]. The reduction of the membrane fluidity usually reduces the crossing rate of the solute, so that the permeability is often lower in membranes containing 35% POPC, 25% sphingomyelin, and 40% cholesterol: P _f = 8.5 × 10⁻³ cm s⁻¹, P _urea = 4.4 × 10⁻⁸ cm s⁻¹, and  = 6.8 × 10⁻³ cm s⁻¹ [2].

In this context, it is essential to note that biological membranes are more complex. They are composed of different types of lipids that may form lipid heterogenities, known as lipid rafts, and are packed with integral and associated proteins. Furthermore, plant plasma membranes exhibit a rather high sterol content [44]. The osmotic permeability of native biological membranes is therefore lower than that of artificial membranes. P _f of the yeast plasma membrane was recently measured as P _f = 2 × 10⁻⁴ cm s⁻¹ [45].

Furthermore, unstirred layers can contribute significantly to, or even dominate, the resistance of a membrane to a given solute. Unstirred layer effects result from the fact that solute concentrations adjacent to the membrane surface differ from the bulk solution, even when the bulk solutions are well mixed [46]. The thickness of unstirred layers is considered to be of particular importance for the volatile CO₂ and artificial lipid bilayers, but does not exceed ~1 μm in small cells [46]. The estimated permeability to CO₂ in 50% POPC and 50% cholesterol was  = 3.5 × 10^-1 cm s⁻¹ [47] or tenfold larger [48], but measurements on native membranes have reported values that were more similar to those of NH₃.

Selected plant aquaporins conduct urea

While urea is an important excretion product of nitrogen waste in animals, it is a ubiquitous, rather low-abundant metabolite in plant nitrogen metabolism. Its specific interest for plant physiology stems from its vast agricultural use. Urea (NH₂CONH₂) has become the most popular nitrogen fertilizer by mass in recent years, in part due to its ~15% lower costs compared to other fertilizers. Urea is degraded by soil microorganisms to ammonia and nitrate within days or weeks, but is also directly acquired and utilized by plants [49].

NtAQP1, a PIP1 from tobacco [50], and ZmPIP1;5 from maize [51] were reported to conduct at least some urea. However, unlike several mammalian and microbial AQPs that also conducted urea, these PIPs have a narrow, rather water-selective ar/R filter. Other PIPs with the identical ar/R region did not conduct measurable urea, e.g., AtPIP2;1 [52], and one should be cautious about their relative urea permeability compared to that for water. A low urea permeability has been reported for the plant plasma membrane [53]. Transport measurements with native AQP9 and AQP7, which have a large pore diameter, and mutational analysis in mammalian AQPs, including AQP1 [42], suggested that size selection is an important determinant for urea transport.

In a functional complementation screen using a yeast mutant that lacked the endogenous high affinity urea transport, four TIPs, TIP1;1, TIP1;2, TIP2;1, and TIP4;1, but no NIPs or PIPs, were isolated from a cDNA library from Arabidopsis [54]. The TIP transcripts were regulated by the plant nitrogen level and may function in plants to equilibrate urea between the cytoplasm and the vacuole. The fact that the vacuolar membrane is particularly permeable to urea had been identified earlier, and Nt-TIPα was suggested to account for this in tobacco [53]. However, the question whether urea transport by TIPs is of physiological relevance has not yet been answered.

A mutational analysis, in which AtPIP2;1 was exchanged at the ar/R filter into the residues that are found in TIPs and NIPs, indicated that all TIP- and NIP-like pores conducted urea at a sufficiently high rate to promote yeast growth [52]. Urea conduction correlates with the larger pore diameter in TIPs and NIPs [11] and suggested that all native TIPs allow urea passage. Urea transport proved that these mutant aquaporins were functional [52]. The pore requirement for efficient urea transport is explicitly shown for NIP5;1-like mutations in Fig. 1d, e. The urea conductance of the pollen specific AtTIP5;1 has now been confirmed experimentally [55]. The low urea permeability of PIPs was confirmed by MD simulations on SoPIP2;1 [52] and other aquaporins [25].

In addition to the ar/R residues, the exchange of residues without direct contact to the pore can indirectly influence the pore diameter [52]. The structural modeling of mutants with altered ar/R filter revealed a threonine in SoPIP2;1 is crucial to position the phenylalanine side chain of the ar/R. Replacement of the phenylalanine by histidine, as in TIPs (Fig. 1f), required a further concomitant threonine to glycine exchange. Pores were only functional when the histidine was exchanged in combination with the glycine, the residue in TIPs and NIPs (Fig. 1; Table 1).

Although initial studies had proposed that TIP isoforms were localized to different vacuole types, recent studies with GFP-tagged TIP isoforms in Arabidopsis have challenged that proposal [56]. All isoforms localized to the same large central vacuole. Vacuolar subdomains may, however, exist as AtTIP1;1, but not AtTIP2;1, localized to invaginated bulbs after salt treatment [16]. Although minor differences between the selectivity of TIP isoforms were observed, there is currently little evidence that this property is reflected in physiologically relevant compartment-specific differences.

A PIP-mutant containing the ar/R filter of NIP6;1 transported urea [52], which is in agreement with urea transport by native AtNIP6;1 in oocytes [11]. However, this urea conductance of AtNIP6;1 may be of minor physiological importance, as AtNIP6;1 distributes boric acid in plants, particularly in young and developing shoots [57]. In summary, experimental and computational evidence suggests that size selection is a major determinant for urea conductance in plant aquaporins (Fig. 1e, f), although other factors also contribute to establish efficient urea transport [25, 52].

Permeability of membranes and channels to signaling molecules

Atomic structures provide an estimate on the maximal size of a solute that can enter and pass the pore. However, whether small solutes that pass the size limit are efficiently conducted is difficult to predict from static structures. H₂O₂ has somewhat similar molecular properties as water, can form hydrogen bonds, and has the same permeability as water,  = 2 × 10⁻⁴ cm s⁻¹, in yeast [58]. H₂O₂ is a relatively long-lived reactive oxygen species and functions in plants, which are more tolerant to H₂O₂ than animals, as a signaling molecule in abiotic and biotic stresses [59]. The amount of H₂O₂ is dependent on its production, which is compartmentalized to the peroxisomes, mitochondria, chloroplasts, and the apoplast. This is counteracted by complex scavenging redox chemistry in all cellular compartments [60]. The H₂O₂ compartmentation and accumulation had early been visualized with H₂O₂-specific precipitates [61] and other dyes and reactants. H₂O₂ is further produced during the hypersensitive response as a defense reaction to restrict pathogen infection and to confine the pathogen spread outside the plant. Salicylic acid and salt treatment [62] or chilling [18, 63] lead to the extracellular accumulation of reactive oxygen species, including H₂O₂. The same treatments reduced the root water permeability via PIP internalization in Arabidopsis roots [62], suggesting that H₂O₂ functions as a common regulator of water homeostasis.

The possibility that aquaporins might conduct hydrogen peroxide had initially been suggested by pressure probe measurements in Chara, at huge H₂O₂ concentrations [64]. In a later screen using yeast, AQP8, TaTIP1;1, and TaTIP1;2 were found to conduct H₂O₂, out of 24 candidates [65]. Using a similar experimental setup, further plant aquaporins, including AtPIP2;1, AtPIP2;4, and TIP and NIP isoforms, were identified to conduct H₂O₂ [29]. Although the hydrogen peroxide permeability was confirmed with other assays, these studies critically relied on a rather indirect yeast-based growth assay, where H₂O₂ conducting aquaporins caused higher susceptibility to the growth on plates that contained H₂O₂.

Because of the inherent difficulty to experimentally quantify H₂O₂ conduction by PIPs, steered MD simulations were set up with membrane-embedded SoPIP2;1 (Fig. 2a). A vectorial constant force was applied to the molecules, to increase the probability that the solute entered and crossed the pore within 16 ns. Similar simulations were performed with water. H₂O and H₂O₂ encountered a distinct barrier profile (Fig. 2b) that was estimated from solute trajectories and dwell times in the pore [29]. However, a similar maximal barrier height, close to the ar/R filter, was calculated (Fig. 2). Mutations at the ar/R region, that widened the pore, allowed larger H₂O₂ fluxes [29]. Yeast growth in the presence of H₂O₂ is explicitly shown for mutants that have ar/R filters representative of all PIP, TIP and NIP isoforms of Arabidopsis (Fig. 3). Control experiments under non-selective conditions confirmed that these aquaporin mutants did not affect normal growth [29]. Mutants containing ar/R filters of SIPs were not included, since the SIPs are quite divergent in the entire pore and have only very low identity, making homology models less reliable. In accordance with the fact that the ar/R filter was most important for H₂O₂ conduction, these experiments suggest that a larger pore diameter increased the H₂O₂ influx, and triggered cell death (Fig. 3).

Fig. 3.

Mutations in the ar/R region affect the resistance to methylamine and to hydrogen peroxide. AtPIP2;1 and its pore mutants were expressed in wild type yeast and did not affect growth under non-selective control conditions (upper panel). Yeast sensitivity to methylamine (MeA, middle panel) and H₂O₂ (lower panel) was changed by aquaporin mutants. The lower spot in each lane corresponds to a 25-fold dilution. All mutants were introduced into AtPIP2;1 and were expressed from the same plasmid. Mutant nomenclature: NIP1;2-like (NIP1;2L) had ar/R residues as in AtNIP1;2, etc. The TIP1234L and NIP1234L mutants contained an additional Thr55 to Gly mutation. Yeast survival on MeA correlates with MeA transport capacity (efflux from the cytosol), while survival on H₂O₂ indicates inability to transport H₂O₂

In plants, the H₂O₂-induced PIP internalization might also participate in auto-regulating and controlling H₂O₂-signaling. In a chilling-sensitive maize cultivar, chilling increased H₂O₂ levels, which coincided with low water permeability [18]. In contrast, no H₂O₂ accumulation was observed in a tolerant cultivar with high water permeability [18]. These cultivar differences may suggest that the dissipation of high H₂O₂ involved water channels.

Nitric oxide (NO), another signaling molecule, is sufficiently stable to diffuse and communicate between cells. In mammalian cells and liposomes, the rate of NO influx directly correlated with the abundance of AQP1, measured using the NO-specific fluorescent dye diaminofluorescein-2 [66]. Although, intriguingly, earlier studies had suggested that the membrane itself was not a barrier for NO in erythrocytes [67]. NO often works in tandem with H₂O₂ and is involved in programmed cell death, pathogen defense, flowering, stomatal closure, and gravitropism in plants [68]. NO evolution and signaling involve different cellular compartments, but whether plant membranes present a barrier for NO diffusion remains unclear. A possible role of PIPs in NO signaling must therefore await a thorough test on aquaporin-deficient mutant plants. Cross-talk with other signaling molecules and the inevitable concomitant loss of water permeability in PIP mutants will complicate the interpretation of possible effects. However, loss-of-function mutants are essential to determine whether the H₂O₂ and/or NO transport through plant aquaporins is of any physiological importance, or just a peculiar side effect without impact.

Molecular requirements for NH₃ conduction and physiological relevance

The first plant aquaporins that conducted ammonia were molecularly identified by yeast complementation. Yeast was transfected with cDNA libraries and only cells expressing TIP2s from wheat and Arabidopsis grew under selective conditions [69, 70]. TIP2s have a unique ar/R filter that is distinct from other TIPs, PIPs and NIPs (Table 1). Such library screens usually select for the most abundant cDNAs, while low-abundance channels might have been overlooked with this approach. Indeed, AtTIP1;2 also complemented the growth defect of a yeast mutant that lacked endogenous ammonium transporters on low ammonia and thus conducted this solute [52]. Correlative evidence suggested that NOD26 channels contribute to the ammonia transfer across peribacteroid membranes [71].

Although yeast growth assays on selective media are easy to perform and straightforward, the growth promotion by aquaporins was much weaker than that by unrelated NH₄ ⁺ transporters [52, 69, 70]. The ammonia transport by human AQP3, AQP7, AQP8, AQP9, and wheat TIP2;1 was measured in oocytes [72]. In that study, NH₃ transport was indirectly deduced via the associated pH change and the secondary induction of ionic currents that result from a large NH₃ influx in oocytes. The large NH₃ permeability of AQP8 was confirmed in lipid bilayer experiments [73], but a loss-of-function mouse did not reveal any physiological function for the ammonia permeability of AQP8 [74]. Whether mammalian AQP1 transports NH₃ is controversial; some studies have suggested that mammalian AQP1 did not conduct NH₃ in yeast and oocytes [42, 69, 72]. In contrast, AQP1 mutants, that had an altered ar/R selectivity filter, transported NH₃ [42]. Other studies using oocytes [75, 76] or red blood cell ghosts [77] identified NH₃ transport by AQP1, but not by AQP5. Since AQP1 and AQP5 have the identical ar/R filter and differ only minimally in their pore [78], their proposed different NH₃ transport activity is quite remarkable [76]. MD simulations on AQP1 revealed that NH₃ encountered a large permeation barrier in the pore and suggested that AQP1 is unlikely to increase NH₃ transport above the intrinsic membrane leak [25].

In AQP8 and TaTIP2;1, ionic currents were associated with ammonia transport when expressed in oocytes. Therefore, these channels were suggested to transport not only NH₃ but in addition H⁺, at physiological pH [72]. Whether this H⁺ was conducted through the aquaporin pore, or was an “artifact” of the oocyte expression system, remains to be resolved in other expression systems. Oocytes have complex endogenous background currents that are activated by the influx of NH₃. AQP8 excluded H⁺ and NH₄ ⁺ in lipid bilayer experiments, at least at low pH [73]. MD simulations revealed that the large electrostatic barrier and the hydrophobic nature of the pore impair the transport of any charged molecule [41]. Since this barrier remains when NH₃ enters the pore, an NH₃-induced H⁺ transport appears unlikely.

NH₃ is of similar size as water. The pore diameter thus cannot explain why only some aquaporins conduct NH₃. The mutational exchange of the ar/R filter by two residues in TaTIP2;1 created a channel that excluded NH₃, but had wild type water transport [72]. MD simulations confirmed that the ar/R selectivity filter is the major barrier for NH₃ in plant aquaporins [52]. PIP2;1 pores imposed a large barrier to NH₃, making NH₃ fluxes by PIPs unlikely, in accordance with experimental findings. The barrier for NH₃ did not only result from direct pore-NH₃ interactions themselves, but from perturbations that NH₃ exerted to water–pore interactions at the ar/R filter and to the water arrangement in the pore [52]. This is in agreement with another, more detailed study, where it was similarly concluded that, if the pore diameter is sufficiently large to be passed by the solute, the interactions of water close to the solute in the ar/R region determine whether it is transported at high rate [25]. Since the surrounding water is important for small solute transport, such molecules do not behave as free “gas”.

The experimental evidence from native aquaporins and the MD simulations indicated that NH₃ was most efficiently conducted through pores with wide, hydrophobic ar/R filters [25]. This view was challenged by the observation that inhibitor studies had revealed different sensitivities of the water and NH₃ fluxes of TaTIP2;2 expressed in yeast [45]. It was suggested that NH₃ was not transported in the same pore as water, but through a separate pore, possibly the central cavity formed by the tetramer. However, inhibitor studies must be treated with caution, especially when the relatively unspecific Hg²⁺ is used. Initial inhibitor studies on human AQP3 had suggested different pathways for water and larger solutes, such as urea or glycerol, but this proposal was not substantiated by subsequent studies [79]. Furthermore, the central pore was occluded in the crystal structure of human AQP5 by a lipid, which questions its potential role as solute pathway in some homologs [78].

Although several TIPs allowed NH₃ fluxes, AtPIP2;1 could not be converted into a NH₃ channel by TIP-like pore mutations. The mutational exchange of the ar/R sites into those resembling TIP1 or TIP2 did not lead to channels that allowed mutant yeast to grow on limiting ammonium [52]. Conversion of a PIP into a NIP-like channel promoted yeast growth on limiting ammonium, suggesting that NIPs conduct NH₃ [52]. The same growth tests were performed on methylamine (MeA, H₃C–NH₂, pK_a = 10.66) in wild type yeast and are explicitly shown (Fig. 3). MeA is often used as a transport analog of ammonia, but it is possible that methylamine and NH₃ do not share the identical pore requirements for conduction. Yeast-expressing methylamine-conducting channels survived on high MeA, probably due to the passive efflux of the uncharged form of this toxic compound. NIP1, NIP2, NIP3, NIP4, and NIP6-like pores most efficiently rescued growth on NH₃ [52] and MeA (Fig. 3), while TIP3, TIP4, NIP5 and NIP7-like pores were intermediate. No growth complementation was observed for TIP1 and TIP2-like pores, although TIP1s and TIP2s had been experimentally identified as NH₃ channels. It is thus likely that not only the ar/R region, but other parts of the pore are critical for conduction. These residues may be involved in positioning the TIP-like ar/R residues or may recruit NH₃/NH₄ ⁺ to the external entrance. Since the structures of the external pore vestibules of TIPs are not available, a rudimentary NH₄ ⁺ recruitment and de-protonation mechanism, similar to that in ammonium transporters, cannot be excluded in some aquaporins [80]. Finally, the tetrameric pore center may be involved in NH₃ passage.

In accordance with a redundant role for TIPs in the vacuole, over-expression of AtTIP2;1 [70] or knock-out of the two major TIP1 forms, AtTIP1;1 and AtTIP1;2, had little effect on plant growth and metabolism [81]. While there is a general tendency that aquaporins with a wide, hydrophobic ar/R filter conduct at least some NH₃, the exact pore requirements, and whether the central tetramer pore contributes, are still less than clear. Although likely, direct evidence for the physiological relevance of aquaporin-mediated NH₃ fluxes in plants still awaits discovery.

Are plant membranes highly permeable to CO₂ and O₂?

A large number of studies suggested that lipid bilayers and cellular membranes have an exceptionally high permeability to CO₂. This high CO₂ permeability of artificial POPE and POPC membranes was confirmed by MD computer simulations and suggested that these membranes did not provide barriers for CO₂ fluxes [35, 36]. As CO₂ conduction is most frequently indirectly measured, via the associated pH change in the presence of carbonic anhydrase, unstirred layer effects and the diffusion of the associated HCO₃ ⁻ are of exceptional importance at physiological pH [47]. In fact, HCO₃ ⁻ diffusion, as well as carbonic anhydrase availability in unstirred layers, often limited net CO₂ diffusion across bilayers [47, 48].

Over-expression of mammalian AQP1 in epithelial layers [48], or the double knock-out of AQP1/AQP5 in erythrocytes, did not affect the CO₂ permeability, which argues against their significant CO₂ permeability [82, 83]. Carbonic anhydrase, the enzyme that is used to accelerate the pH changes associated with CO₂ fluxes in some studies, is inhibited by the rather non-specific blocker HgCl₂ [82]. This questions the significance of some related experiments that used HgCl₂ to block aquaporins. Furthermore, MD simulations showed that the pores of AQP1 and GlpF constitute rather impermeable barriers to CO₂. The central cavity in AQP1 tetramers also imposed a barrier, but this was lower than that of the water pores. Therefore, the central pore was considered as a potential pathway for dissolved CO₂ in unusual membranes with very low intrinsic CO₂ permeability, e.g., when much of the membrane was occupied by integral proteins [25, 35, 36].

Other studies, however, suggest that the intrinsic permeability of membranes can be increased by channels. Specific membranes in the mammalian body were suggested to be rather impermeable to CO₂ [84]. Mass spectrometric determinations of complex CO₂–HCO₃ ⁻ equilibrations that involved carbonic anhydrase, suggested that the intrinsic was at least one order lower in erythrocytes than the predicted by other methods [85]. Erythrocytes lacking AQP1 had less than 50% CO₂ permeability [85]. Reconstituted AQP1 in liposomes also indicated significant CO₂ transport. This study furthermore revealed that CO₂ fluxes were unusual in that they were not affected by changes in membrane fluidity [86]. Research on oocytes expressing carbonic anhydrase had suggested that AQP1 and a pore mutant facilitated CO₂ fluxes, based on cytosolic pH changes measured with impaled pH-electrodes [87]. Using the same method, slightly accelerated pH changes were recorded in oocytes expressing carbonic anhydrase and tobacco NtAQP1. These reflected a 45% increase in CO₂ uptake compared to controls [88]. Considering that several PIPs increased the water permeability of oocytes by more than 20-fold, the 1.45-fold increase in CO₂ transport by NtAQP1 in oocytes was rather small. The H₂O transport, for comparison, was roughly doubled by the expression of NtAQP1 [50]. Additional mammalian aquaporins were suggested to conduct CO₂ after expression in oocytes [76]. In the latter study, transport was deduced from transient external surface pH changes that were recorded in the absence of carbonic anhydrase. These were not measured from whole cells, but with pH-sensitive electrodes on small membrane patches of ~15 μm diameter.

Direct evidence for a large CO₂ permeability of plant aquaporins is still scarce, but plants with increased PIP expression levels had an improved carbon bio-availability. This was attributed to increased CO₂ transport rates across plant membranes (by NtAQP1) and within leaf tissue [88]. Several other studies confirmed this improved growth by PIP expression and identified that an altered internal CO₂ conductance was affected in PIP mutants. The internal conductance accounts for the movement of CO₂ from the stomatal cavities to the site of carboxylation, which is the enzyme RUBISCO (ribulose 1,5 biphosphate carboxylase) in chloroplasts [89]. RUBISCO has a relatively low affinity for CO₂, and the increased CO₂ bio-availability improves photosynthesis. As a consequence, the CO₂ assimilation rate increased in tobacco leaves that over-expressed AtPIP1;2 [15] and NtAQP1 [88], and in rice that strongly expressed HvPIP2;1 from barley [90]. PIPs were thus beneficial for plant photosynthesis under favorable conditions, but transgenic tobacco and rice with increased PIP levels were more heavily affected by water stress, due to their higher transpirational water loss [15, 90]. The internal conductance and stomatal opening are considered to be limiting factors for efficient photosynthesis and CO₂ assimilation, and ultimately affect the efficiency of nitrogen and water use in the leaf. The internal conductance increased by 40% in leaves of transgenic rice plants that over-expressed HvPIP2;1, and stomatal aperture and leaf morphology were also changed [90]. In contrast, the internal CO₂ conductance correlated similarly in tobacco with NtAQP1 levels, but other leaf parameters were unchanged [89]. The involvement of aquaporins in determining the internal conductance had already been predicted from experiments with the non-specific inhibitor Hg²⁺ [91].

How the aquaporins increase the internal conductance, a parameter that is experimentally difficult to access, is less clear, but recent work on tobacco sheds light on this issue. NtAQP1 was originally localized to the plasma membrane [22] and vesicular structures [92], but was recently additionally found in the chloroplast envelope [93]. Direct transport measurements on plasma membrane vesicles clarified that the CO₂ permeability of plasma membranes was unchanged by NtAQP1 down-regulation, although the water permeability was about half the magnitude in NtAQP1 RNAi plants [93]. The pH changes associated with the CO₂ transport identified that native chloroplast envelope vesicles had an unusually low permeability to CO₂, approximately fivefold lower than that of the plasma membrane [93]. Genetic reduction of NtAQP1 reduced the CO₂ permeability further, but did not affect the water permeability of these membranes [93]. At the same time, chloroplast envelopes had an about threefold higher water permeability than the plasma membrane, and their water permeability was not affected by NtAQP1. Whether PIP isoforms from other species are also localized and abundant in plastid envelopes should be analyzed in the future, as PIPs have not yet been identified in the plastid proteomes of diverse other plants [94].

Many classical studies suggested that O₂, like CO₂, crosses membranes with an intrinsically high rate, but direct oxygen transport is difficult to measure. MD simulations confirmed the large lipid permeability to oxygen and suggested that mammalian AQP1 and bacterial GlpF impose significant barriers for O₂ fluxes [25, 35]. However, it was considered that in unusual and exceptionally tight membranes, for example with very high protein content, some transport might occur, most likely through the tetramer center.

Although classical experimental evidence and computer simulations collectively exclude that oxygen transport across aquaporins occurs, this has been suggested in animal cells with a high rate of oxygen consumption. AQP1 was induced by low oxygen levels, and hypoxia-inducible gene expression correlated with mammalian AQP1 abundance in a manner that was consistent with O₂ transport by AQP1 [95]. However, direct oxygen transport rates were not measured, and hypoxia-related metabolic by-products might be involved in the AQP1-related regulation.

Interestingly, in Arabidopsis, AtNIP2;1 was transcriptionally up-regulated upon water logging and oxygen deprivation (anoxia) [96]. AtNIP2;1 only minimally transported water and glycerol, but efficiently transported the uncharged form of lactic acid (CH₃CH(OH)COOH), which is synthesized after the metabolic shift from aerobic respiration to lactic acid fermentation in anoxia [96]. The export of lactic acid may contribute to long-term metabolic adaptation to anoxia.

Short-term effects of anoxia include the decrease of water conductance in the root by the rapid closure of PIPs [26]. This mechanism involves cytosolic acidification and PIP gating. Furthermore, a concerted transcriptional down-regulation of PIPs and TIPs upon anoxia leads to a further long-term reduction of water transport in the root [97]. Although there is currently no evidence for O₂ conduction by plant aquaporins, the rapid channel closure upon anoxia would potentially reduce external O₂ supply. It would also restrict further dissipation of the remaining O₂ in the tissue, assuming that these channels hypothetically conduct oxygen.

Specific functions of plant aquaporins in metalloid nutrition and toxicity

It was shown more than a decade ago that some aquaglyceroporins from microorganisms conducted metalloids, potentially as an unspecific side activity [98]. In plants, several NIPs are specifically involved in the regulated uptake and distribution of selected metalloids. Boron is a trace element that is important for proper plant cell wall structure, and in boron-deficient roots, AtNIP5;1 was up-regulated [99]. Its loss reduced the boric acid uptake in roots, lowered biomass production, and increased the sensitivity of roots, shoots, and seeds to boron deficiency [99]. AtNIP6;1 was also specifically involved in boron homeostasis, despite its capability to transport other solutes [57]. AtNIP5;1 and AtNIP6;1 transported boric acid, B(OH)₃, when expressed in oocytes [57, 99].

From a mechanistic viewpoint, it is also interesting that these channels transported little (AtNIP5;1) or no water (AtNIP6;1), despite their large pore diameter. Similarly, bacterial GlpF had only 17–30% of the water transport capacity of AQP1 [100]. The differential water transport rates in water-selective and less-selective pores with larger diameter has been evaluated using MD simulations, but unexpectedly, the water conductance in GlpF was predicted to exceed that of the narrower water-selective counterpart [101, 102]. The experimentally determined low water conductivity of the wider pores is one of the few details that have not yet been correctly predicted by MD simulations. Residues in the external pore vestibule may be relevant for water exclusion in pores with a wider ar/R filter by electrostatic repulsion. In the aquaglyceroporin of Plasmodium, a glutamate in the outer pore vestibule and in proximity to the arginine of the ar/R filter was responsible for high water permeability. A mutation in this glutamate abolished water, but not glycerol transport [103].

Silicic acid, Si(OH)₄, which has a slightly larger diameter than boric acid, is a beneficial nutrient for some plants, such as rice, and increases their disease resistance. Silicon accumulates and ultimately irreversibly precipitates on the leaves as amorphous silica or silica gel (SiO₂–nH₂O). The rice mutant lsi1 (low silicon rice 1) was defective in silicon uptake, accumulated less silicon in the shoot, yielded fewer grains, and was more susceptible to disease [104]. This mutant had a defect in the silicon channel OsNIP2;1 (Lsi1), which has a unique wide ar/R region (Table 1). This protein showed an in- and efflux capacity for Si(OH)₄ when expressed in oocytes and additionally transported urea and boric acid. OsNIP2;2 has the same ar/R filter and transported silicic acid in oocytes, while OsNIP1;1 and OsNIP3;1, which differ in their ar/R region, did not transport silicic acid [105]. A further, very interesting, feature of several NIPs is their polar localization pattern to the outer side of the root cells, allowing vectorial transport of metalloids across epidermal, exodermal, and endodermal lavers [104].

Arsenic is a critical soil contaminant and poses large problems to agriculture in south-east Asia. It exists in different oxidation forms in nature, predominantly as As(V) in aerated soils, but as the reduced form As(III) in paddy soils. While As(V) is a structural analog of phosphate and is transported by phosphate transporters, As(III) mimics Si(OH)₄ and is transported by OsNIP2;1 (Lsi1) [106]. In paddy rice, arsenite and silicon share the same transport pathway via the roots, which explains why rice is specifically susceptible to arsenic accumulation in contaminated soils. Loss of the silicon influx channel OsNIP2;1 (Lsi1) significantly decreased arsenite uptake in rice [106]. Similarly, loss of AtNIP1;1, which has a different selectivity filter (Table 1), led to Arabidopsis that were more resistant to As(OH)₃ [107]. Whether the physiological role of AtNIP1;1 is to facilitate arsenic uptake remains unclear, because most plants may try to exclude arsenic from their cytosol. AtNIP1;1 is the closest homolog of NOD26 in legumes and has the identical selectivity filter, and correlative evidence suggested that NOD26 conducts water and NH₃. NOD26 is localized in the peribacteroid membrane, which encloses nitrogen-fixing bacteroides in a vesicular compartment [108].

Antimonite, Sb(OH)₃, was transported by specific aquaporins in microorganisms and may be transported by NIPs, as there is evidence for channel-like uptake mechanisms in plant roots. Selenium is required in trace amounts for the mammalian diet, but is not essential for plants and may be taken up by NIPs in the form of selenic acid (HO)₂SeO₂.

The specificity of NIPs in metalloid transport, together with their already molecularly identified export facilitators, may lead to multiple biotechnological applications. Over-expression of these channels is likely the strategy of choice to overcome low uptake rates in soils, which are deficient in the respective metalloids. Likewise, metalloid excretion, or efflux to the vacuole through modified TIPs, may improve plant growth in areas where contaminant metalloids exceed tolerable levels. Such approaches have already begun for boron [109] and are likely to become of agronomical impact in the near future.

Summary and conclusion

Experimental, structural and computational evidence suggests that plant aquaporins select solutes by two major strategies. Firstly, larger solutes are sterically excluded to permeate narrower pores by size exclusion. Secondly, smaller solutes that can potentially pass the pore diameter are selected by a hydrophobic mechanism that involves water–protein interactions in the ar/R region. It is likely that the combination of experimental techniques and MD simulations will finally resolve the open questions associated with aquaporin physiology, but despite their value to visualize solute conduction and to predict the selectivity patterns, computer simulations cannot replace experimental work. The detailed analysis of mutant plants will finally resolve the physiological impact of the distinct selectivity of PIPs, TIPs and NIPs. A molecular understanding of aquaporin selectivity and regulation may ultimately help to minimize water losses by cultivated plants, as water shortage gets relevant in increasing areas of the world.