Protein Science (1998), 7: 1458-1468. Cambridge University Press. Printed in the USA.
Copyright © 1998 The Protein Society

ARTICLE

Prediction of functional residues in water channels and related proteins

A. FROGER, ¹ B. TALLUR, ² D. THOMAS, ¹ and C. DELAMARCHE ¹

¹ UPRES-A CNRS 6026, Biologie Cellulaire et Reproduction, Équipe "Canaux et Récepteurs Membranaires," Université de Rennes1 bâtiment 13, Campus de Beaulieu, 35042 Rennes Cedex, France
² IRISA, Institut de Recherche en Informatique et Systèmes Aléatoires, Campus de Beaulieu, 35042 Rennes Cedex, France

(Received December 22, 1997; Accepted February 25, 1998)

Reprint requests to: Christian Delamarche, UPRES-A CNRS 6026, Équipe Canaux et Récepteurs Membranaires bâtiment 13, Campus de Beaulieu, 35042 Rennes Cedex, France;
e-mail:cdelam@univ-rennes1.fr.

Abbreviations: AQP, aquaporin protein; CA, correspondence analysis; GlpF, glycerol facilitator protein; MIP, major intrinsic protein; ORF, open reading frame; PC, personal computer.

Abstract

In this paper, we present an updated classification of the ubiquitous MIP (Major Intrinsic Protein) family proteins, including 153 fully or partially sequenced members available in public databases. Presently, about 30 of these proteins have been functionally characterized, exhibiting essentially two distinct types of channel properties: (1) specific water transport by the aquaporins, and (2) small neutral solutes transport, such as glycerol by the glycerol facilitators. Sequence alignments were used to predict amino acids and motifs discriminant in channel specificity. The protein sequences were also analyzed using statistical tools (comparisons of means and correspondence analysis). Five key positions were clearly identified where the residues are specific for each functional subgroup and exhibit high dissimilar physico-chemical properties. Moreover, we have found that the putative channels for small neutral solutes clearly differ from the aquaporins by the amino acid content and the length of predicted loop regions, suggesting a substrate filter function for these loops. From these results, we propose a signature pattern for water transport.

Keywords: aquaporin; correspondence analysis; glycerol facilitator; MIP family; multiple sequence alignment; protein function prediction

Article Contents

(You can also go directly to the beginning of the text.)

Introduction: Fig. 1. Predicted membrane topology of the MIP family proteins
Results: The MIP family; Table 1. Members of the MIP family analyzed in this study; Table 2. Sequences of the MIP family analyzed to check the predictions; Sequence alignment analysis; Fig. 2. Part of the average similarity plots for the MIP family proteins; Fig. 3. Portion of the 40 multiple sequence alignment; Conventional statistical analysis; Multivariate statistical analysis; Fig. 4. Correspondence analysis applied on the MIP family proteins
Discussion: Equation 1
Materials and methods: Selection of protein sequences; Multiple alignment software; Alignment analysis; Correspondence analysis
Acknowledgments
References

Introduction

Water is the most ubiquitous molecule of living systems and its movement across cell membranes accompanies essential physiological functions. All biological membranes exhibit some water permeability as a result of diffusion through the lipid bilayer, under the driving force of the osmotic gradient. However, some cells are able to transport water at greatly accelerated rates by way of water-selective channels called aquaporins. The discovery of these specialized proteins has led to new information on both the physiological and molecular mechanisms of membrane water permeability and on links between aquaporins and human diseases (King & Agre, 1996).

The first functionally characterized aquaporin, AQP1 (original name CHIP28), was discovered in the membrane of red blood cells (Preston et al., 1992). This protein is distributed in many water permeable tissues. AQP1 is constitutively expressed in the proximal tubules and descending thin limbs of the kidney where it mediates 90% of bulk water reabsorption (Sabolic & Brown, 1994). Presently, seven other mammalian aquaporins have further been identified (reviewed by Brown et al., 1995): AQP2, the vasopressin sensitive water channel expressed in the renal collecting tubules is implicated in a form of nephrogenic diabetes insipidus; AQP3 also expressed in kidney, exhibits water, glycerol and urea permeability; AQP4 is predominantly expressed in the brain where it probably plays a role in cerebrospinal fluid outflow regulation; AQP5 is distributed in a variety of exocrine glands; AQP6 (hKID) is exclusively expressed in the kidney and is not regulated by antidiuretic hormone. Finally AQP7 and AQP8, two novel aquaporins, are predominantly expressed in testis (Ishibashi et al., 1997a, 1997b), AQP7 being a mixed channel, like AQP3.

Aquaporins have also been identified in amphibian, plant, bacteria, and insect tissues. For example, FA-CHIP has been characterized in frog urinary bladder (Abrami et al., 1994), gamma-TIP and RD28 have been identified, respectively, in the tonoplast and the plasma membrane of Arabidopsis thaliana (Maurel et al., 1993; Daniels et al., 1994). AqpZ was discovered in Escherichia coli where it seems to play a role in osmoadaptation by contributing to cellular volume regulation in hypoosmolar media (Calamita et al., 1995, 1997). One aquaporin, called AQPcic, was recently characterized in the digestive tract of an homopteran sap-sucking insect, Cicadella viridis (Beuron et al., 1995; Le Cahérec et al., 1996).

The aquaporins belong to an ancient and ubiquitous family of channel proteins called the MIP family with reference to MIP26 (AQP0), the Major Intrinsic Protein expressed in lens fiber cells (Gorin et al., 1984). The function of the archetype MIP26 is unclear. When expressed in Xenopus oocytes, MIP26 weakly enhances permeability for ions, water, and glycerol, suggesting multiple physiological functions for this protein (Chandy et al., 1995; Kushmerick et al., 1995; Mulders et al., 1995).

Sequence comparisons revealed that the glycerol facilitators (GlpF) are also members of the MIP family. Bacteria use glycerol as a carbon source for glycolysis, and for lipid biogenesis. The transport of glycerol into the cytoplasm involves two types of mechanisms: a passive diffusion across the lipid bilayer, or a facilitated uptake when the external concentration of glycerol is low (Richey & Lin, 1972). The GlpF protein acts as a selective pore for uncharged molecules, depending on the molecular size of the substrates (Heller et al., 1980; Sanders et al., 1997). A glpF gene has been cloned from Bacillus subtilis, E. coli, Haemophilus influenzae, Pseudomonas aeruginosa, and Shigella flexneri. A glycerol facilitator was also characterized in Saccharomyces cerevisiae (Luyten et al., 1995).

Members of the MIP family are now included in the PROSITE database (Bairoch et al., 1996) with the signature sequence [HNQA]-x-N-P-[STA]-[LIVMF]-[ST]-[LIVMF]-[GSTAFY], with x as any residue and alternative residues in brackets. All the MIP proteins are about 260 residues long, with the exception for two MIP yeast proteins that have more than 600 residues, as a result of extended N- and C-terminal segments. Hydropathy plots and experimental investigations by epitope insertions and mutagenesis reveal a common topology for these molecules, with six transmembrane domains. After sequence alignment, the most divergent members of the family display less than 20% identity, but like the other MIP proteins, they share highly conserved residues distributed throughout the sequence. A characteristic homology common to the members of the MIP family is the Asn-Pro-Ala (NPA) motif repeated in two opposite loops (Fig. 1). Moreover, the NH2 and COOH-terminal halves are sequence-related, suggesting an ancestral intragenic gene duplication as known for other channel families (Pao et al., 1991; Wistow et al., 1991; Reizer et al., 1993; Saier, 1994).

Fig. 1 Predicted membrane topology of the MIP family proteins. Primary structure of a monomer based on the hourglass model of aquaporins proposed by Jung et al. (1994). The molecule consists of six membrane-spanning domains connected by loops A to E, with cytoplasmic NH2- and COOH- termini. Highly conserved residues including the two Asn-Pro-Ala (NPA) repeats are indicated. The stars indicate positions 1 to 5 predicted from the present study to play a functional role. In the hourglass model, loops B and E protrude into the lipid bilayer, and the NPA boxes joints themselves in the middle of the channel to form a single aqueous pathway. Indices are used to identify segments studied in the correspondence analysis.

The detection of a functional and/or structural site in a nucleic acid or protein is of major interest. Unfortunately, there is no simple and straightforward method, since the specific properties of a given molecule often result from a limited number of key residues. For example, single nucleotide changes in given positions of tRNA^Trp and tRNA^Tyr are sufficient to transform the identity of these tRNAs to glutamine (Cavarelli & Moras, 1993). Single amino acid substitutions also modify the ion-selection properties of the sodium channel protein into a calcium channel (Heinemann et al., 1992). The most common way to find the residues of functional importance is by site-directed mutagenesis. However, to avoid random targeting that can be both laborious and expensive, computational methods are of particular relevance. Since the enigmas of the structure/function of proteins are hidden within their sequences, a careful comparison of homologous sequences from a large number of organisms should enable us to make certain predictions.

Functionally, for the MIP family, we know that the aquaporins present a highly selective, but quantitatively heterogenous, water permeability (Yang & Verkman, 1997), and that the GlpF channels transport glycerol but exclude water. Therefore, in order to understand the molecular mechanisms of substrate selectivity and permeation, one must find the residues that are specifically linked to each subclass. In this report, we present the results obtained from a sequence analysis of the MIP proteins and point out some residues that could modulate the selectivity for water or glycerol. Two different approaches were used: (1) a systematic comparison of the physico-chemical properties of the amino acids at each position in multiple sequence alignments and (2) a statistical analysis (conventional and multivariate) to compare the amino acid composition in sequence segments. The results are assessed by comparison with published experimental data.

Results

The MIP family

In Tables 1 and 2, we present an updated compilation of 122 complete and 27 partial sequences of the MIP family available from public databases.

Table 1. Members of the MIP family analyzed in this study^a

Table 2. Sequences of the MIP family analyzed to check the predictions^a

These sequences were retrieved by combining searches with specific keywords and/or with the PROSITE signature sequence for the MIP family. The number of MIP sequences has rapidly increased since the recent publication of 84 members by Park and Saier (1996). However, the last update of the PROSITE database (PS00221, November 1997) lists only 63 MIP proteins. All of the proteins presented in Tables 1 and 2 have a nonredundant accession number in the databases. It is obvious that only a small number of closed sequences results from genetic polymorphism. For example, AqpZ (U38664) differs from Bnip (D49469) by only three amino acids, but both are homologous genes of different E. coli strains.

Proteins of the MIP family are abundant in plant cells, where they are expressed both in plasma and vascular membranes. For A. thaliana, the model plant for genome sequencing, we retrieved 20 different amino acid sequences from standard databases, but found 84 sequences (not shown) in the specialized database of the Institute for Genomic Research (http://www.tigr.org). Most of A. thaliana MIPs are believed to be aquaporins. The plant aquaporins play important physiological roles, including different levels of water permeability, probably for adaptation to a variety of water stresses and in relation to the plant cell compartmentalization (Maurel et al., 1997; Weig et al., 1997).

Complete genome sequences provide the opportunity to examine the presence of MIP sequences in these genomes. Twelve complete genomes were examined: S. cerevisiae possess four MIP encoding genes, one of them, Q12302, cited in the conclusion paragraph, contains a frameshift mutation. E. coli possess two distinct and functionally identified MIP proteins, one aquaporin and one glycerol facilitator. Two distinct MIP coding sequences have also been described in the genome of H. influenzae but only one in that of Archaeoglobus fulgidus, Bacillus subtilis, Borrelia burgdorferi, Mycoplasma genitalium, Mycoplasma pneumoniae, and Synechocystis sp. Interestingly, no MIP-protein-encoding sequences were retrieved from the genomes of Helicobacter pylori (eubacteria), Methanobacterium thermoautotrophicum, and Methanococcus jannaschii (archeobacteria).

Sequence alignment analysis

The quality of the alignments obtained using various software was similar. The major difference was in the number of gaps introduced into the alignments. For this reason, further experiments were done with two independent methods: a precise analysis of the alignment content in regions of high conservation between the sequences, and a statistical analysis on sequence segments less dependent on alignment methods.

The PILEUP program was used to create a multiple alignment with the test set of 40 sequences. In the resulting dendrogram, the data are clearly separated into two major clusters and each cluster fits into a main functional subgroup: cluster I corresponds to glycerol transport and cluster II corresponds to water transport (Table 1). In all the sequence alignment studies, the proteins AQP3 (Nos. 5 and 6), which are able to transport either glycerol or water when expressed in the Xenopus oocyte, were allocated to cluster I. MIP26, the archetype of the MIP family, was allocated to cluster II. The function of some of the 40 proteins is yet unknown, but the partition into two major clusters was not affected by adding or removing new sequences to create the multiple alignment. This is in agreement with a recent phylogenetic study (Park & Saier, 1996), concluding that most and probably all the MIP family members will exhibit specificity for water and/or small neutral solutes. At this point in our analysis, we postulated that it should be possible to identify residues that are directly linked to function by a careful inspection of the multiple alignments. In the following experiments, we will consider the alignment of 40 sequences and the two subgroups extracted from this alignment: cluster I, 14 sequences, which will be named the GlpF cluster, and cluster II, 26 sequences, which will be named the AQP cluster.

One of the most popular prediction methods for getting information concerning the protein membrane topology consists in averaging the residue hydropathy for consecutive 19-residue segments (Kyte & Doolittle, 1982). We compared the average hydrophobicity profiles of the two clusters, using different hydrophobicity scales (data not shown). The resulting profile supports the prediction of six membrane spanning segments for the MIP proteins (Reizer et al., 1993). The transmembrane segments are clearly superimposable for the two functional subgroups, while for predicted loop segments, the hydrophobicity scores differ significantly, suggesting the presence of residues linked to function.

The hydrophobicity profile method is based on an average value calculated along a sliding window. Thus, the presence of gaps in the alignment will alter the comparison between the two clusters. Therefore, we used another method to extract the information at each position of the alignment: the similarity profile method, which was applied with a one residue window (Fig. 2).

Fig. 2 Part of the average similarity plots for the MIP family proteins. At each position of the 40 sequence alignment, the average similarity score is calculated for each cluster. The similarity score was calculated using the modified Dayhoff table (Gribskov & Burgess, 1986) scale, and plotted with a sliding window of one. This figure illustrates the similarity plot from position 620 to 640 of the global alignment. The similarity for the 40 sequences is shown as a broken curve, the GlpF cluster is shown as a dotted curve and the AQP cluster as a continuous curve. A maximum score of 1.5 is obtained when all amino acids at a given position of the alignment are identical. Highly conserved residues presented in Figure 1 are indicated above the alignment. The positions P2 and P3 predicted to have a functional role are indicated.

Potential residues of high interest are directly visualized on the curves, when the similarity score for each cluster is higher than the score for all sequences. Five discriminating positions were identified, where the physicochemical properties are conserved within each subgroup, but differ between the subgroups. These positions are located in highly conserved regions, and can be easily retrieved from any sequence (Figs. 1, 3).

Fig. 3 Portion of the 40 multiple sequence alignment. Sequences are numbered according to Table 1 and the position of the first residue in each sequence segment is indicated in parenthesis. The positions P1 to P5 predicted to have a functional role in the MIP proteins are boxed. The highly conserved residues presented in Figure 1 are indicated below the alignment (*).

From our observations, we deduced the following rule:

Position 1, located in the terminal part of the third transmembrane segment, is an aromatic residue in the GlpF cluster. This residue is not aromatic in the AQP cluster.

Positions 2 and 3 are located in loop E, just behind the second "NPA" box. They correspond respectively to an acidic then a basic residue (D, then R or K) in the GlpF cluster and to two small uncharged residues in the AQP cluster.

Positions 4 and 5 correspond to two consecutive amino acids located in the sixth transmembrane segment. These positions can be defined as two aromatic residues in the AQP cluster compared with a proline followed by a nonaromatic residue in the GlpF cluster (except for the yeast protein YFF4, which possess an alanine in P4 and a tryptophan in P5).

A close inspection of the second set of sequences (Table 2) confirms that the five positions described above are composed of remarkably conserved residues. Of 112 proteins, 21 are functionally characterized, and 20 of them present a perfect correspondence with the rule, confirming possible functional role for key residues. The exception concerns a newly characterized aquaporin from A. thaliana, NLM1 (Weig et al., 1997), which possess mixed key residues of GlpF cluster for P1 and P5, and AQP cluster for P2-P4. Of 450 key residues available from Table 2, only 22 (4.8%) differ from those observed in Figure 3. In the absence of a functional characterization of the corresponding proteins, it is difficult to relate these differences to extensions or divergences of the rule, or to errors in the sequences. Residues particularly well conserved concern each of the couples P2-P3 and P4-P5. These couples are closely tied to each other: charges of opposite sign in P2-P3 associate with nonaromatic residues in P4-P5 or uncharged residues in P2-P3 associate with aromatic residues in P4-P5. Presently, there are only two exceptions to this observation and they concern two MIP proteins of Caenorhabditis elegans (U40415, Z35595). Another observation concerns the position P1, which is generally the counterpart of P4-P5 (aromatic/nonaromatic) between the two clusters.

Conventional statistical analysis

The differences between the glycerol facilitators and the aquaporins, revealed by the average similarity profiles, result in changes of amino acid content between the subgroups. These compositional differences can be quantified by conventional statistical analysis. The mean amino acid content and the mean length in amino acids of the segments, predicted to be inside or outside the membrane, were calculated for each subgroup and compared with the "Student t" test. Such a trivial analysis revealed some of the residues that participate in the key positions: position 1 in TM3 (Y), position 2 and 3 in loop E (D,R) and position 4 in TM6 (P), as defined previously. We also observed that some amino acids are over-represented in glycerol facilitators (W in TM2; P in loop B; I, F, P, T in loop C; N in TM4 and loop D; G, L in loop E), and that others are significantly more frequent in aquaporins (A, S, W in TM1; V in loop B; C in TM3, R and K in loop D; H in TM5 and TM6). Moreover, the statistical analysis revealed significant differences (Student t test, P < 0.001) in the predicted length of two external loops. Loops C and E are longer for the GlpF cluster than for the AQP cluster. The respective mean lengths, calculated from the 40 sequences of Table 1 are 27.92 compared with 18.24 for loop C and 28.35 compared with 18.52 for loop E.

The results of the above comparisons strongly suggest that several amino acids could contribute to structural and functional differences between the glycerol facilitators and the aquaporins. However, these results were obtained by averaging the amino acid content over already predetermined functional subgroups. Thus, to check our observations without a priori classification, we have carried out a multivariate analysis on the 40 MIP proteins.

Multivariate statistical analysis

Correspondence analysis was used to compare the amino acid frequencies in the MIP proteins, from segments of different lengths. CA does not take into account the amino acid order along the polypeptide chain or the presence of gaps in the alignment. The results of the correspondence analysis are shown on a factorial map in which each amino acid is represented by its letter, and each MIP protein by the number corresponding to the sequence in Table 1. Proteins having a close resemblance are positioned closely on the map. Thus, if the two functional subgroups are clearly separated on the map, this suggests that the segment participates in the channel specificity. Preliminary experiments using entire sequences or short segments gave unsatisfactory results. The best separations were obtained for medium-sized segments located between but excluding two strictly conserved amino acids. These selected segments are marked on Figure 1 from NH2- to COOH termini: E₁-G₁, G₁-Q, Q-E₂, E₂-G₂, G₂-P. In our study, the projection map obtained with the segment Q-E₂, including the major part of the third transmembrane domain and loop C, gave a clear and complete separation of the two functional subgroups (Fig. 4).

Fig. 4 Correspondence analysis applied on the MIP family proteins. Correspondence analysis was applied to the segment located between the residues Q and E₂, principally including the third transmembrane domain and loop C (Fig. 1). Each segment in the MIP proteins is represented as a vector point in a 20-dimensional space, where each dimension corresponds to the relative frequency of one of the 20 amino acids. The cloud of proteins as well as that of amino acids are then projected on the plane of first and second factors and these projections are overlaid. Proteins of similar composition appear as neighbors and amino acids contributing to the separation are easily identified.

The first factor accounts for 20.9% of the total variance between MIP proteins, while factors 2, 3, and 4 account for 13.7%, 10.4%, and 9.7%, respectively. Thus, the first factor corresponds to the most predominant differences present in the population sequence. Along this factor (F1), all glycerol facilitators have positive values and all aquaporins have negative ones. The simultaneous representation of proteins and amino acids leads to sensitive means of identifying those amino acids responsible for the separation on the factor map. Tryptophane, aspartic acid, phenylalanine, and tyrosine have a high relative contribution on the first factor to separate glycerol facilitators and aquaporins. These results support our key residues rule given for position P1 (located inside the Q-E₂ segment): aromatic residues (phenylalanine and tyrosine) are characteristic of glycerol facilitators. Along factor 2, aquaporins are separated into two major subgroups, one of which is a subgroup of plant aquaporins (numbers 15 to 20). Lysine, cysteine, and glycine seem responsible for this separation. Intriguingly, two AQP2 aquaporins, from rat and human (numbers 34 and 35), are positioned in the extreme negative values of factor 2. The presence of glutamic acid in their segment appears to be responsible for this separation.

Discussion

Recently, Park and Saier (1996) reported a phylogenic study of the MIP family based on the comparison of 84 protein sequences. The authors concluded that most, if not all, MIP proteins fall into one of two physiological groups: (1) the water transport by the aquaporins and (2) the small neutral solutes transport such as glycerol by the glycerol facilitators. Considering these two main functions, the present work focuses on the characterization of functional residues in the MIP proteins. The study was carried out by the analysis of multiple protein sequence alignments, a standard strategy to highlight residues of structural and functional importance in a protein family (Livingstone & Barton, 1993), and by a new strategy based on amino acid frequencies in sequence segments.

The present work highlights five key positions that could play an important role in the structure and function of each MIP family subgroup, and consequently, raises questions on the mechanisms of substrate selectivity by conserved residues. An hourglass topological model was previously proposed for AQP1 (Jung et al., 1994). In this model, the loops B and E dip and join into the channel. According to this scheme, the second "NPA" region, which includes P2 and P3, should be very close to P4 and P5 and possibly to P1 (Fig. 1), thus forming an atomic network whose bond properties are differentiated by these key residues, resulting in channel specificity. However, as we have also shown, the amino acid length and composition of the external loops C and E differ significantly from glycerol facilitators to aquaporins, and these parameters may probably play a crucial role in pore selectivity. In summary, we predict that the external aperture of the channel, surrounded by the key residues and protected by the external loops act as a substrate filter. Further sequence analysis will be required to combine our observations in a single rule, and to identify the amino acids able to interact at distance. Mutagenesis experiments on AQPcic (an insect aquaporin) and on GlpF (the glycerol facilitator of E. coli) are currently in progress in our laboratory to assess these predictions. Problems with this experimental approach is that missense mutations can induce multiple effects, particularly in protein folding and trafficking (Mulders et al., 1996). In these conditions, it is difficult to determine if the mutation also affects the functional properties of the channel.

Meanwhile, by examining the current literature in the field, we can evaluate the accuracy of our predictions. The following results support our predictions:

The insertion of 25 amino acids in loop E of hAQP1 (V201-E1) resulted in a markedly reduced water channel activity (Preston et al., 1994). We have shown by statistical analysis that the length of this loop is significantly longer in glycerol facilitators than in aquaporins.
In the aquaporin hAQP2, the substitution of a segment of loop C (eight amino acids) by an other segment of loop C (eight amino acids) from GlpF abolished water transport function (Bai et al., 1996). This substitution introduced into the loop two proline, one isoleucine, and one phenylalanine, residues that we demonstrated to be more abundant in loop C in glycerol facilitator than in aquaporins.
The substitution of a serine by a cysteine at position 196 of hAQP1 greatly reduced the water transport activity of this aquaporin (Jung et al., 1994). This serine corresponds to position P2 in our rule.
The aquaporin AQP7 was recently cloned from rat testis. Expression experiments in Xenopus oocytes have shown that AQP7 also transports glycerol (Ishibashi et al., 1997a, 1997b). The five key residues of AQP7 are Y, D, R, P, and V and correspond, according to our rule, to the solute transport signature. Interestingly, like AQP7, AQP3, which is also a mixed-channel, bears the signature of glycerol facilitators, as if the key residues are more important for solute transport than for water transport.

The results presented below do not follow our predictions but allows refinements of our rule:

The insertion of 25 amino acids in loop C of hAQP1 (T120-E1) had no significant effect on water permeability (Preston et al., 1994), even though this loop is significantly longer in glycerol facilitators. This result suggests that the length of loop C does not affect the water permeability but probably the solute transport properties.
The permeability to water and glycerol for MIP26 and NOD26 have been recently demonstrated (Kushmerick et al., 1995; Mulders et al., 1995; Rivers et al., 1997). These functional properties should assign the two proteins to cluster I, by analogy with AQP3. However, MIP26 and NOD26 also exhibit ion channel activity (Weaver et al., 1994; Kushmerick et al., 1995; Lee et al., 1995; Modesto et al., 1996). The SmpX protein (D43774) of the Gram-negative bacterium Synechococcus might also be involved in ion transport, particularly in copper transport (Kashiwagi et al., 1995). Therefore, cluster II proteins could include functional subgroups other than water transport.

A great number of published MIP sequences were obtained by the RT-PCR technique using degenerate primers directed against the two conserved "NPA" boxes. Moreover, a large number of sequences were obtained from poly(A)⁺ RNA prepared from the kidneys of vertebrates, and in plants, most MIP genes were cloned after induction by water-deficit stress. As a result of such strategies, a bias could be expected with a great number of published MIP sequences being aquaporins bearing the two "NPA" boxes. A question arises immediately: Is it possible to define aquaporin sequences without reference to the "NPA" boxes?

We examined this question by retrieving all sequences in the databases matching with a new signature sequence, which excludes the "NPA" motif. We have focused this preliminary search on key amino acids P2 to P5 deduced from the 26 sequences of cluster II (Fig. 3), and located between highly conserved residues (G2-P, Fig. 1). From these, we designed a signature pattern specific for water but not for solute transport:

(the signature is reported in the PROSITE format, x corresponds to any residue and brackets to alternative residues).

This signature sequence based on only six residues was used to extract related proteins from the Swiss-Prot and the TREMBL databases. These databases include more than 170,000 entries corresponding to the translation of all CDS in the EMBL Nucleotide Sequence Database (Apweiler et al., 1997). We retrieved 91 sequences, from which 81 sequences have, as expected, the signature we proposed for aquaporin and possess the NPA boxes. From these 81 sequences, we have retrieved four new MIP sequence proteins: U58207 from Allium cepa, P28238 from Gallus gallus, Q12302 from S. cerevisiae and P93683 from Sorghum bicolor. These sequences were not retrieved in our previous analysis because they correspond to partial sequences without the first "NPA" box. Consequently, the total number of MIP sequences recorded in this study is 153. The other ten sequences retained with the signature sequence do not appear to be, at a first glance, water channels (P80517, Q09652, P44843, O00213, P30208, P47542, P78018, Q10782, P46933, P74003). In conclusion, we have shown that, if the presence of the "NPA" boxes is a characteristic of the MIP family, water transport function can be described by using a specific signature devoid of any "NPA" box, such as AQP6 (AB006190).

On the basis of the constantly increasing number of sequences available for the MIP family, together with more functional characterizations, we believe that a similar approach is feasible to design a specific signature for other transport functions.

Materials and methods

Selection of protein sequences

The sequences were extracted from GenBank, EMBL, and PROSITE databases (indexing date Dec 1997). We based our initial analysis on a test set of 40 MIP protein sequences representative of different groups of organisms (Table 1). In order to avoid any bias in the analysis, we took care to include divergent members rather than a too large number of very similar ones. This initial analysis was limited by the number of available sequences. Therefore, another set, including 109 other fully or partially sequenced MIP proteins, was used to amplify the predictive reliability of the results (Table 2).

Multiple alignment software

We have used three multiple alignment programs, using various values for the critical parameters, K-tuple and gap penalties: CLUSTALW (Thompson et al., 1994), MAP (Huang, 1994), and PILEUP from the GCG package (Devereux et al., 1984). We have tried three amino acid substitution tables: the Dayhoff table PAM250, (Dayhoff et al., 1978), the standard GCG table (Gribskov & Burgess, 1986), and the BLOSUM62 table (Henikoff & Henikoff, 1992). The computation from the GCG programs was performed using INFOBIOGEN resources (http://www.infobiogen.fr).

Alignment analysis

Assignments for transmembrane domains were made with the TMAP program (Persson & Argos, 1994), available on the Worldwide Web (http://www.embl-heidelberg.de/tmap/tmap_mul.html). The program CGD (Delamarche, unpubl. obs.) was used to analyze the amino acid composition along the sequence alignments: amino acid frequencies, similarities, charge distribution, hydrophobicity, etc. CGD runs on PC using the powerful computing and graphical tools of EXCEL worksheets. For the hydrophobicity profiles, each of the three hydropathy scales was used: Hopp and Woods (1981), Kyte and Doolittle (1982), Rao and Argos (1986). For the similarity profiles, CGD uses the same substitution matrices than for the multiple alignments. The similarity score at a given position of the alignment is the arithmetic mean of all the pairwise amino acid similarity scores at that position. The distance between the couples formed by two gaps or one gap and one amino acid is zero. In this way, a position in the alignment that contains many gaps will have a lower similarity score. This algorithm is similar to that used by the program PLOTSIMILARITY from the GCG package (Devereux et al., 1984).

Correspondence analysis

Correspondence analysis (CA) (Benzecri, 1973) may be applied to contingency tables whose rows and columns correspond, respectively, to "individuals" (or objects) and attributes (or categories). In our study, the data consist of 40 * 20 contingency tables whose rows correspond to 40 MIP protein sequences (entire sequences or segments situated between two strictly conserved amino acids) and whose columns represent 20 amino acids. The (i,j) cell of the contingency table contains the frequency of amino acid j in sequence i. CA was carried out on entire sequences and on five different segments using the CORRESP procedure of SAS/STAT software. In this paper, we present the results obtained on the segment Q-E2, which gave the highest inertia.

Acknowledgments

The authors thank Dr. Rebecca Hartley and Dr Isabelle Pellerin for helpful discussions. This work was supported by the Langlois Foundation (Rennes, France).

References (The NLM-formatted Reference List is also available.)

Abrami L, Simon M, Rousselet G, Berthonaud V, Buhler JM, Ripoche P. 1994. Sequence and functional expression of an amphibian water channel: A new member of the MIP family. Biochem Biophys Acta 1192:147-151.

Apweiler R, Gateau A, Contrino S, Martin MJ, Junker V, O'Donovan C, Lang F, Mitaritonna N, Kappus S, Bairoch A. 1997. Protein sequence annotation in the genome era: The annotation concept of SWISS-PROT+TREMBL. ISMB 5:33-43.

Bai L, Fushimi K, Sasaki S, Marumo F. 1996. Structure of aquaporin-2 vasopressin water channel. J Biol Chem 271:5171-5176.

Bairoch A, Bucher P, Hofmann K. 1996. The PROSITE database, its status in 1995. Nucleic Acids Res 24:189-196.

Benzecri JP. 1973. L'Analyse des Correspondances, tome2. Paris: Dunod.

Beuron F, Le Cahérec F, Guillam G, Cavalier A, Garret A, Tassan JP, Delamarche C, Schultz P, Mallouh V, Rolland JP, Hubert JFH, Gouranton J, Thomas D. 1995. Structural analysis of a MIP family protein from the digestive tract of Cicadella viridis. J Biol Chem 270:17414-17422.

Brown D, Katsura T, Kawashima M, Verkman AS, Sabolic I. 1995. Cellular distribution of the aquaporins: A family of water channel proteins. Histochem Cell Biol 104:1-9.

Calamita G, Bishai WR, Preston GM, Guggino WB, Agre P. 1995. Molecular cloning and characterization of AQP2, a water channel from Escherichia coli. J Biol Chem 270:29063-29066.

Calamita G, Kempf B, Rudd KE, Bonhivers M, Kneip S, Bishai W, Bremer E, Agre P. 1997. The aquaporin-Z water channel gene of Escherichia coli: Structure, organization and phylogeny. Biol Cell 89:321-329.

Cavarelli J, Moras D. 1993. Recognition of tRNAs by aminoacyl-tRNA synthetases. FASEB J 7:79-86.

Chandy G, Kreman M, Laidlaw DL, Zampighi GA, Hall JE. 1995. The water permeability per molecule of MIP is less than that of CHIP. Biophys J 68A:353.

Daniels MJ, Mirkov TE, Chrispeels MJ. 1994. The plasma membrane of Arabidopsis thaliana contains a mercury-insensitive aquaporin that is a homolog of the tonoplast water channel protein TIP. Plant Physiol 106:1325-1333.

Dayhoff MO, Schwartz RM, Orcutt BC. 1978. A model of evolutionary changes in proteins. In Dayhoff MO, ed. Atlas of protein sequence and structure, vol 5, suppl 3. Washington, DC: National Biochemical Research Foundation. pp 345-358.

Devereux J, Haeberli P, Smithies O. 1984. A comprehensive set of sequence analysis programs for the VAX. Nucleic Acids Res 12:387-395.

Gorin MB, Yancey SB, Cline J, Revel JP, Horwitz J. 1984. The major intrinsic protein (MIP) of the bovine lens fiber membrane: Characterization and structure based on cDNA cloning. Cell 39:49-59.

Gribskov M, Burgess RR. 1986. Sigma factors from E. coli, B. subtilis, phage SPO1, and phage T4 are homologous proteins. Nucleic Acids Res 14:6745-6763.

Heinemann SH, Terlau H, Stuhmer W, Imoto K, Numa S. 1992. Calcium channel characteristics conferred on the sodium channel by single mutations. Nature 356:441-443.

Heller KB, Lin ECC, Wilson TH. 1980. Substrate specificity and transport properties of the glycerol facilitator of Escherichia coli. J Bacteriol 144:274-278.

Henikoff S, Henikoff GJ. 1992. Amino acid substitution matrices from protein blocks. Proc Natl Acad Sci USA 89:10915-10919.

Hopp TP, Woods KR. 1981. Prediction of protein antigenic determinants from amino acid sequences. Proc Natl Acad Sci USA 78:3824-3828.

Huang X. 1994. On global sequence alignment. Comput Applic Biosci 10:227-235.

Ishibashi K, Kuwahara M, Gu Y, Kageyama Y, Tohsaka A, Suzuki F, Marumo F, Sasaki S. 1997a. Cloning and functional expression of a new water channel abundantly expressed in the testis permeable to water, glycerol, and urea. J Biol Chem 272:20782-20786.

Ishibashi K, Kuwahara M, Kageyama Y, Tohsaka A, Marumo F, Sasaki S. 1997b. Cloning and functional expression of a second new aquaporin abundantly expressed testis. Biochem Biophys Res Comm 237:714-718.

Jung JS, Preston GM, Smith BL, Guggino WB, Agre P. 1994. Molecular structure of the water channel through Aquaporin CHIP. J Biol Chem 269:14648-14654.

Kashiwagi S, Kanamaru K, Mizuno T. 1995. A synechococcus gene encoding a putative pore-forming intrinsic membrane protein. Biochim Biophys Acta 1237:189-192.

King LS, Agre P. 1996. Pathophysiology of the aquaporin water channels. Annu Rev Physiol 58:619-648.

Kushmerick C, Rice SJ, Baldo GJ, Haspel HC, Mathias RT. 1995. Ion, water and neutral solute transport in Xenopus oocytes expressing frog lens MIP. Exp Eye Res 61:351-362.

Kyte J, Doolittle RF. 1982. A simple method for displaying the hydropathy character of a protein. J Mol Biol 157:105-132.

Le Cahérec F, Deschamps S, Delamarche C, Pellerin I, Bonnec G, Guillam MT, Thomas D, Gouranton J, Hubert JF. 1996. Molecular cloning and characterization of an insect aquaporin. Eur J Biochem 241:707-715.

Lee JW, Zhang Y, Weaver CD, Shomer NH, Louis CF, Roberts DM. 1995. Phosphorylation of nodulin 26 on serine 262 affects its voltage-sensitive channel activity in planar lipid bilayers. J Biol Chem 270:27051-27057.

Livingstone CD, Barton GJ. 1993. Protein sequence alignments: A strategy for the hierarchical analysis of residue conservation. Comput Applic Biosc 9:745-756.

Luyten K, Albertyn J, Skibbe WF, Prior BA, Ramos J, Thevelein JM, Hohmann S. 1995. Fps1, a yeast member of the MIP family of channel proteins, is a facilitator for glycerol uptake and efflux and is inactive under osmotic stress. EMBO J 14:1360-1371.

Maurel C, Reizer J, Schroeder JI, Chrispeels MJ. 1993. The vacuolar membrane protein gamma-TIP creates water specific channels in Xenopus oocytes. EMBO J 12:2241-2247.

Maurel C, Tacnet F, Guclu J, Guern J, Ripoche P. 1997. Purified vesicles of tobacco cell vacuolar and plasma membranes exhibit dramatically different water permeability and water channel activity. Proc Natl Acad Sci USA 94:7103-7108.

Modesto E, Lampe PD, Ribeiro MC, Spray DC, Campos de Carvalho AC. 1996. Properties of chicken lens MIP channels reconstituted into planar lipid bilayers. J Membr Biol 154:239-249.

Mulders SM, Knoers NVAM, Van Lieburg AF, Monnens LAH, Leumann E, Wühl E, Schober E, Rijss JPL, Van Os CH, Deen PMT. 1996. New mutations in the AQP2 gene in nephrogenetic diabetes insipidus resulting in functional but misrouted water channels. J Am Soc Nephrol 8:242-248.

Mulders SM, Preston GM, Deen PMT, Guggino WB, Van Os CH, Agre P. 1995. Water channel properties of major intrinsic protein of lens. J Biol Chem 270:9010-9016.

Pao GM, Wu LF, Johnson KD, Höfte H, Chrispeels MJ, Sweet G, Sandal NN, Saier MH Jr. 1991. Evolution of the MIP family of integral membrane transport proteins. Mol Microbiol 5:33-37.

Park JH, Saier HM Jr. 1996. Phylogenetic characterization of the MIP family of transmembrane channel proteins. J Membrane Biol 153:171-180.

Persson B, Argos P. 1994. Prediction of transmembrane segments in proteins utilising multiple sequence alignments. J Mol Biol 237:182-192.

Preston GM, Carrol TP, Guggino WB, Agre P. 1992. Appearance of water channels in Xenopus oocytes expressing red cell CHIP28 water channel. Sciences 256:385-387.

Preston GM, Jung JS, Guggino WB, Agre P. 1994. Membrane topology of aquaporin CHIP. J Biol Chem 269:1668-1673.

Rao JKM, Argos P. 1986. A conformational preference parameter to predict helices in integral membrane proteins. Biochim Biophys Acta 869:197-214.

Reizer J, Reizer A, Saier MH Jr. 1993. The MIP family of integral membrane channel proteins: Sequence comparisons, evolutionary relationships, reconstructed pathway of evolution, and proposed functional differentiation of the two repeated halves of the proteins. Crit Rev Biochem Mol 28:235-257.

Richey DP, Lin ECC. 1972. Importance of facilitated diffusion for effective utilization of glycerol by Escherichia coli. J Bacteriol 112:784-790.

Rivers RL, Dean RM, Chandy G, Hall JE, Roberts DM, Zeidel ML. 1997. Functional analysis of nodulin 26, an aquaporin in soybean root nodule symbiosomes. J Biol Chem 272:16256-16261.

Sabolic I, Brown D. 1994. Water transport in renal tubules is mediated by aquaporins. Clin Investig 72:698-700.

Saier MH Jr. 1994. Computer-aided analysis of transport protein sequences: Gleaning evidence concerning function, structure, biogenesis and evolution. Microbio Rev 58:71-93.

Sanders OI, Rensing S, Kuroda M, Mitra B, Rosen BP. 1997. Antimonite is accumulated by the glycerol facilitator GlpF in Escherichia coli. J Bacteriol 179:3365-3367.

Thompson JD, Higgins DG, Gibson TJ. 1994. CLUSTAL W: Improving the sensitivity of progressive multiple sequence alignment through sequence weighting, positions-specific gap penalties and weight matrix choice. Nucleic Acids Res 22:4673-4680.

Weaver CD, Shomer NH, Louis CF, Roberts DM. 1994. Nodulin 26, a nodule-specific symbiosome membrane protein from soybean, is an ion channel. J Biol Chem 269:17858-17862.

Weig A, Deswarte C, Chrispeels MJ. 1997. The major intrinsic protein family of Arabidopsis has 23 members that form three distinct groups with functional aquaporins in each group. Plant Physiol 114:1347-1357.

Wistow G, Pisano MM, Chepelinsky AB. 1991. Tandem sequence repeats in transmembrane channel proteins. Trends Biochem Sci 16:170-171.

Yang B, Verkman AS. 1997. Water and glycerol permeabilities of aquaporins 1-5 and MIP determined quantitatively by expression of epitope-tagged constructs in Xenopus oocytes. J Biol Chem 272:16140-16146.