Type I Protein Secretion—Deceptively Simple yet with a Wide Range of Mechanistic Variability across the Family

Abstract

A very large type I polypeptide begins to reel out from a ribosome; minutes later, the still unidentifiable polypeptide, largely lacking secondary structure, is now in some cases a thousand or more residues longer. Synthesis of the final hundred C-terminal residues commences. This includes the identity code, the secretion signal within the last 50 amino acids, designed to dock with a waiting ATP binding cassette (ABC) transporter. What happens next is the subject of this review, with the main, but not the only focus on hemolysin HlyA, an RTX protein toxin secreted by the type I system. Transport substrates range from small peptides to giant proteins produced by many pathogens. These molecules, without detectable cellular chaperones, overcome enormous barriers, crossing two membranes before final folding on the cell surface, involving a unique autocatalytic process.

Unfolded HlyA is extruded posttranslationally, C-terminal first. The transenvelope “tunnel” is formed by HlyB (ABC transporter), HlyD (membrane fusion protein) straddling the inner membrane and periplasm and TolC (outer membrane). We present a new evaluation of the C-terminal secretion code, and the structure function of HlyD and HlyB at the heart of this nanomachine. Surprisingly, key details of the secretion mechanism are remarkably variable in the many type I secretion system subtypes. These include alternative folding processes, an apparently distinctive secretion code for each type I subfamily, and alternative forms of the ABC transporter; most remarkably, the ABC protein probably transports peptides or polypeptides by quite different mechanisms. Finally, we suggest a putative structure for the Hly-translocon, HlyB, the multijointed HlyD, and the TolC exit.

INTRODUCTION: THE T1SS FAMILY

The broad family of ATP binding cassette (ABC) transporters found in all kingdoms of life was first detected in Escherichia coli as a curiously heterogeneous group of membrane proteins (1). Most of these were importers of a variety of small molecules, but surprisingly one was an exporter, HlyB (hemolysin B), that became the prototype of the type I secretion system (T1SS) for translocation of bacterial proteins. This is sometimes also referred to as the ABC-dependent export system. The T1SS can minimally be defined as secreting polypeptides carrying a C-terminal secretion signal and requiring an ABC transporter, with its characteristically highly conserved ATPase domain. As we shall discuss in detail in this review, the T1SS, at least in Gram-negative bacteria, in addition to an ABC transporter, requires a membrane fusion protein (MFP), also embedded in the inner membrane, but reaching across the periplasm to contact a specific outer membrane protein (OMP) to complete the translocon. A more strict definition of the T1SS, but still covering the great majority of known types, includes the requirement that the transport substrate, or allocrite as we shall interchangeably term it, contains multiple calcium ion binding sites. Thus, the most studied and the first group of these secreted proteins to be identified was the repeat in toxins (RTX) family (2–4), containing a varying number of very similar nonapeptide calcium ion binding repeats. These are implicated in the secretion process and located 100 to 200 residues upstream of the secretion signal. The RTX motifs form a unique β-roll structure with aspartate and glycine residues involved in calcium ion binding (5) and, as described in reference 6, over 1,000 RTX proteins are easily detected by a bioinformatic screen of databases.

For other recent general background reviews on different aspects of T1SS, see Linhartova et al. (6), Holland et al. (7), Thomas et al. (8), Lenders et al. (9), the review by Delepelaire, with a survey of a wide range of organisms having the T1SS (10), as well as a comprehensive review by Durand et al. (11) considering the possibilities of this and other protein secretion systems as antibacterial targets. Concerning the regulation of expression of the hly determinant and the mode of action of HlyA, there are recent reviews (8, 12), and these topics will not be considered here.

Pathogenicity Factors

RTX proteins include major pathogenicity factors carried by many important pathogens in mammals, insects, and plants. Urinary tract infections are the most common bacterial infections, particularly in females. Although generally easy to treat, severe infection can lead to pyelonephritis. Uropathogenic E. coli (UPEC) strains usually producing the prototype RTX protein, hemolysin A (hemolysin), are the most frequent cause of the disease. New exciting in vivo studies, for example, by the Hultgren group, using mouse models and human cell lines, are revealing how expression is regulated during infection by UPEC strains and how HlyA promotes urothelial cell death (13). Some of the other most studied type I proteins secreted by Gram-negative pathogens include, as well as the hemolysins, adenylate cyclase (CyaA [14]) secreted by Bordetella pertussis (whooping cough), proteases and lipases secreted by Pseudomonas and other Gram-negative species, and other toxins secreted by E. coli strains (EHEC [15]). In addition, the newly described T1SS members with giant allocrites (16), the multifunctional-autoprocessing repeats in toxins (MARTX family) are found in chromosomal islands of major human pathogens. This family of extremely large polypeptides (in the range 5,000 to 9,000 kDa) have been described primarily so far from the pathogenicity viewpoint, with very little known about secretion. These include the multitoxin MARTX Vc, encoded by rtxA_Vc from Vibrio cholerae (16–18), and a giant adhesin, LapA, an RTX protein from Pseudomonas fluorescens, recently described by Boyd et al. (19). More intriguing, as described later, another giant adhesin, SiiE, containing a completely unrelated set of Ca²⁺ repeats (20), is nevertheless secreted from Salmonella enterica via a T1SS. Very recently, a fascinating possibility has also emerged, indicating that the T1SS exists in the obligate parasitic family Rickettsiaceae. This T1SS apparently secretes several types of ankyrin proteins that appear to have no Ca²⁺ binding motifs at all (see below).

Peptide Transport

T1SS is also responsible for transport of a huge range of peptides up to approximately 10 kDa, widespread in bacteria (see reviews [21, 22]). However, while transport of these peptides depends on ABC transporters, the secretion signal is located at the N terminus. Secretion of such antibacterial peptides, microcins, bacteriocins, lantibiotics, and signaling molecules such as pheromones, has been reported in both Gram-negative and Gram-positive bacteria with frequencies of 33% and 44%, respectively. In Gram-positive strains, a stripped down form of the T1SS, lacking the MFP homologue, is employed. Regrettably, coverage of these fascinating transport systems in this review was not possible and we had also to limit ourselves primarily to T1SS associated with polypeptide transport in Gram-negative bacteria. However, the recently described crystal structures of ABC transporters for the microcin J25 (MccJ25) from E. coli and a peptide from Clostridium thermocellum constitute a major advance and, as we shall describe later, an important frame of reference in developing secretion models for the larger polypeptides.

Some General Characteristics of the Type I Proteins

Type I polypeptide-allocrites are also almost invariably characterized by having few if any cysteines, being very acidic, with pI values frequently below pH 5. In contrast, these proteins are distributed over an enormous range of sizes, extending from less than 100 to 9,000 amino acids. These latter include some adhesins and others packaging a lethal string of distinct toxins that penetrate host cells before the individual toxins are released by an autoproteolytic process (17, 23). However, all these proteins, at least in the Gram-negative bacteria discussed here, are secreted by an ABC/MFP/OMP translocon, the secretion signal is at the C terminus, and the great majority have RTX repeats of some kind. Some MARTX proteins at least are also characterized by a very large number of other types of repeats (not Ca²⁺ binding), especially toward the N terminus. These repeats are also relatively glycine rich, with an apparently conserved core consensus sequence. Recently, Kim et al. (24) showed that, while not required for secretion by the pathogen Vibrio vulnificus, these N-terminal repeats are required for subsequent translocation into the cytosol of the host cell. The MARTX proteins and other novel groups secreted via the T1SS will be discussed in more detail later.

Surprising Variation in the Details of the Secretion Process across the T1SS Family

In our experience, writing a review does not usually mean taking successive short time-outs to revisit some questions, even at the bench, to test new ideas as they arise. Doing this has certainly added stimulation and enjoyment, and we hope that it has produced a better review. Thus, a major theme that developed as we wrote the review is the remarkable way in which we now see the T1SS appearing in many different forms in the bacterial world, for reasons that are not always clear.

Bacterial Protein Secretion Is a Relatively Recent Discovery

In the period up to about 1980, no bacterial protein secretion system had been characterized, and how such proteins reached the exterior was a mystery. Toxins such as a hemolysin released by Gram-negative bacteria were known, but the mechanism of secretion had not been investigated. Studies of the biogenesis of bacterial membranes were largely restricted to the model laboratory strain E. coli K12 that in fact does not secrete any proteins to the medium. Thus, interest in protein translocation was largely confined to unlocking the puzzle of the distinctive partitioning of proteins to the compartments of the cell envelope, inner and outer membranes and periplasm, i.e., the process dependent on the Sec translocon and targeted by an N-terminal “export” or “signal sequence.”

Emergence of a Myriad Secretion Systems

The first clear example of bacterial protein secretion studied in any detail was the secretion of a hemolysin discovered around 1980. This is still the most studied, and it is considered the prototype system. Remarkably, however, by the later 2000s, an amazing variety of protein secretion systems (for a recent review, see reference 25), at least 15 in Gram-negative bacteria had been identified, together with six in Gram-positive bacteria. Four of these systems, designated T1SS, 3, 4, and 6, apparently involve a transenvelope “tunnel,” secreting proteins directly from the cytoplasm to the exterior. The great majority, however, involve an initial step to cross the inner membrane dependent on the Sec system, with the subsequent step to negotiate the outer membrane, surprisingly involving many alternative strategies (for an excellent review series, see reference 26).

The First Authentic Reports of Bacterial Protein Secretion: E. coli Hemolysin

These appeared in the early 1950s, notably by Robinson (27) working in the Seamen’s Hospital, Greenwich, United Kingdom, who identified an E. coli toxin that was cell associated (nonfilterable), heat labile with calcium ion dependence for activity. However, an apparently truly secreted hemolysin, released from late-exponential phase cells, was only identified in the 1960s (28, 29). Pioneering the modern phase of hemolysin studies, in 1979, the group of Werner Goebel in Würzburg, Germany, cloned a cluster of 3 genes found on a plasmid from a pathogenic E. coli strain, and transferred this into an E. coli K12 laboratory strain. Thus, Noegel et al. (30) identified two genes required for production of the active hemolysin molecule and at least one gene encoding a transport function. Springer and Goebel (31) studied the process of release of active hemolysin from E. coli K12. These authors used inhibitors of energy metabolism and protease processing, together with cell fractionation, to distinguish extra- and intracellular hemolysin. They concluded that HlyA was secreted in an energy-dependent process via the periplasm and apparently involving processing of the protoxin to a 55-kDa form able to cross the outer membrane (see also reference 32). However, subsequent studies, benefitting from pulse-chase radiolabeling experiments and more detailed cellular fractionation analysis, quickly demonstrated that HlyA (with a relative mobility equivalent to 107 kDa) was not processed and was secreted directly to the medium (2, 33–35).

T1SS, A NOVEL TRANSENVELOPE SECRETION PROCESS

Introduction

The first sequencing (3) of an hly operon by the group of Rod Welch (Fig. 1) from a T1SS genetic determinant located in the chromosome (as are the great majority) of an E. coli O4 serotype strain, revealed a putative toxin gene hlyA. However, surprisingly, the sequence indicated the absence of any classical N-terminal signal sequence able to target HlyA (calculated size 110 kDa) to the Sec translocon. Then several studies, including the analysis of another human chromosomal Hly determinant, LE2001 (34, 36, 37) studied by the Holland group, and the plasmid determinant, pHly152, studied first by W. Goebel, demonstrated that HlyA secretion did not involve a periplasmic intermediate (33, 38–40) and did not require SecA (41–43). These results indicated a novel protein translocation pathway and raised the intriguing question, how does this protein negotiate two membranes on its way to the medium?

Figure 1.

Organization of the hly, the hasA, and the slaA/lipA/prtA operons. The hly promoter and the binding site of the transcriptional regulators RfhA (213) or Fur are indicated. In the case of the has operon, the surface receptor HasR (gray), and, for the hly operon, the acyltransferase HlyC (green), are also encoded within the operon. The slaA gene encodes a surface protein, lipA a lipase, and prtA a metalloprotease. The allocrite or transport substrate genes are indicated in dark blue, the ABC transporters in brown, the MFPs in red, and the OMP, if present in the operon, in light blue. Please note that the outer membrane protein (TolC in the case of HlyA) is frequently not encoded in the corresponding T1SS operon. CHP in the rtx operon encodes an additional ABC transporter, but how these two transporters function independently or together is unknown (see the text on page 20).

Identifying the Genes Required for Type I Secretion

The first sequence of the hly operon also clearly showed that, in addition to the toxin, this encoded two probable membrane proteins (HlyB and HlyD) and a gene hlyC, subsequently shown to be required for activation of the HlyA toxin (44). Transposon mutagenesis and radiolabeling in mini and maxi cells, confirmed independently the identity of the hemolytically active product of the hlyA gene, as a nonprocessed polypeptide of 110 kDa. In addition, the products of the genes hlyB and D were shown to be located in the inner membrane and essential for translocation to the medium (2, 3, 36, 37, 45, 46). Then in 1990, a major advance by Wandersman and Delepelaire demonstrated that an E. coli outer membrane protein, TolC, was also essential for the secretion of HlyA (47). TolC was known to be involved in excretion of a wide range of molecules as well as the import of some bacteriocins, and apparently important in some way for the overall maintenance of the integrity of the outer membrane. The tolC gene is not linked to the hlyA operon, but the Wandersman group (48) showed that an operon in the phytopathogenic Erwinia chrysanthemi, encoding secretion genes for the metalloproteases B and PrtC, also encoded PrtF, a TolC homologue (Fig. 2). This addition of TolC to the putative type I secretion machinery was a crucial finding, providing a potential partner for the inner membrane proteins HlyB, D, and therefore the means to complete the transport pathway through the outer membrane to the medium.

Figure 2.

Summary of structural information for the hemolysin A T1SS. (Left) Homology model of dimeric HlyC based on the crystal structure of ApxC (53). The TAAT-specific insertion, which is unique and not present in the GNAT family, is highlighted in red. The Arg residue together with the catalytic triad (composed of a Ser, His, and Asn) that interacts with ACP is shown in ball-and-stick representation. (Right) The NMR structure of the CLD (54) and the crystal structure of the ATP-bound dimer of the NBDs (175) are shown in green and yellow, respectively. The TMD of HlyB and for HlyD are shown schematically as blue and red cylinders, respectively. The trimeric crystal structure of TolC (green/cyan/yellow) (138) is shown in cartoon representation. No structural information for the substrate, HlyA, is available. Please note that the presentation here for the structure, oligomeric state, and the extent of HlyD overlap with TolC is arbitrary. Note, while Koronakis et al. (145) suggested a trimeric arrangement for HlyD, more and more crystal structures and modeling evidence suggest that a closely packed hexameric state, as in other MFP analogues, is essential to maintain a tightly sealed structure. Evidence also suggests a tip-to-tip interaction or a small overlap between the ends of the MFP and TolC as in the AcrAB complex (169). The indicated contact between HlyB and TolC is arbitrary and remains controversial for the analogous tripartite AcrAB-TolC efflux pump (169). See the text for more details.

Importantly, these early studies also showed that HlyC, required to activate HlyA, and later shown by Koronakis and Hughes and colleagues (49, 50) to be a specific acyltransferase (see below), was not required for secretion (44). This is particularly relevant since HlyC with its cofactor, the cellular enzyme ACP (acyl carrier protein-dependent fatty acylation) is an unusual enzyme (51), unexpectedly synthesized in equimolar amounts to HlyA, with which it forms a stoichiometric complex. This led to the appealing hypothesis that HlyC could play a chaperone-like role to maintain HlyA competent for secretion. However, there is no evidence to support this and, in the absence of hlyC, there is apparently no effect on secretion of HlyA.

The Acyltransferase HlyC: Structure and Function

The hlyC gene encodes the acyltransferase HlyC. This enzyme, as indicated above, is not required for secretion of HlyA, but is required for modification of the unfolded HlyA (52) in the cytoplasm prior to transport. In concert with the endogenous E. coli acyl-ACP, HlyC transfers acyl groups on to two internal lysine residues (Lys564 and Lys690). The predominant length of the acyl chains is 14, 15, or 17 carbon atoms. Such acyltransferases, including the most studied Gcn5-like N-acyltransferase (GNAT) family, are found in all kingdoms of life.

Recently, the Koronakis group (53) reported the crystal structure of an HlyC homologue from Actinobacillus pleuropneumoniae, ApxC (Fig. 2). This protein shares 70% amino acid identity with HlyC and can replace E. coli HlyC in vivo to activate pro-HlyA. The structure of ApxC revealed a dimer, which was supported by analysis of the oligomeric state of ApxC in solution. The monomer is composed of a five-stranded β-sheet flanked by six helices. More intriguing is a deep cleft between the third and fourth strand. This cleft is also present in structures of acyltransferases of the GNAT family and thus appears to present a conserved structural feature of these enzymes. Combined with mutagenesis studies, active site residues important for catalysis were mapped to the deep cleft in the central β-sheet and residues important for ACP interaction were also identified. This structure therefore represents an important advance in our understanding of the mechanism of activation of HlyA-like molecules transported by T1SS and will open up new ways to suppress the activity of these toxins.

Structural Organization of HlyA Including the RTX Motifs

Studies of hemolysin secretion have primarily involved two genetic determinants, both isolated from a human host, the plasmid derivative pHly152, and the chromosomal derivative LE2001 (30, 34). The two determinants show small deviations in amino acid sequence; confusingly, this includes the presence of an additional residue early in the hlyA gene in plasmid isolates, thus encoding 1,024 residues, not 1,023 as found with the chromosomal determinants like LE2001. The large N-terminal domain of HlyA contains the hydrophobic regions involved in pore formation, together with lysine residues 563 and 689 (numbering according to the chromosomal borne determinant; 564 and 690, in the plasmid determinant), the sites of acylation by HlyC. Importantly, the sequence of the hly operon (reviewed by Welch [4]) also revealed the presence of many glycine- and aspartate-rich nona-repeats in HlyA, designated as repeats in toxins (RTX) and located toward the C terminus of the protein. The RTX motif usually begins with GG; however, there is no strict consensus, although a sequence considered able to bind a calcium ion is adhered to. Thus, the number of these repeats in E. coli hemolysin has variously been reported to be as few as 6 (54) or 11 to 17 (55), depending on the strictness of the consensus applied. The original consensus indicated by the Rod Welch group was, in fact, LXGGXGND, and 13 repeats were identified on this basis. In contrast, Lecher et al. (54), based on the strict consensus sequence, GGXGXDXUX (where U is a large or a hydrophobic amino acid), containing the critical G and D residues directly involved in binding Ca²⁺, identified 6 repeats in two rather closely linked clusters, terminating 173 residues from the C terminus of HlyA. In contrast, for example, the proteases PrtG, PrtB, and PrtC, all from E. chrysanthemi, have a single tight cluster of 3 and 4 RTX repeats, respectively, ending approximately 90 amino acids from the C terminus (56, 57).

Structural studies have shown that two nonapeptides bind one Ca²⁺ ion (Fig. 3). The main coordination occurs via the two aspartic acid side chains, but the backbone and side chain interaction of other amino acids within the repeat also contribute to Ca²⁺ coordination. It is important to stress that two consecutive repeats bind the same ion. This arrangement creates a so-called parallel β-roll or parallel β-helix. Within such a motif, a strand is present between the two repeats in the Ca²⁺-bound state so that the strands back against each other as first described in alkaline protease (5).

Figure 3.

Known structures of substrates of T1SS. Ca²⁺ ions are highlighted as blue spheres and proteins are displayed in cartoon representation. From top to bottom, a fragment of the giant adhesion SiiE that has non-RTX Ca²⁺ binding sites (121), alkaline protease from P. aeruginosa (5), T1SS RTX lipase LipA (113), and HasA (110), which lacks a Ca²⁺ binding domain. The β-rolls are highlighted. The N and C termini are indicated. Please note that only a single domain of SiiE is shown. The domain architecture is provided above the structure. SiiE contains 53 so-called Big domains shown as black boxes. The N-terminal coiled coil and the C-terminal insertion are shown as white, rectangular boxes. The three Big domains of SiiE that have been crystallized are highlighted by a yellow box. The molecular architecture of a calcium binding site is summarized in the central black box. One Ca²⁺ ion (blue sphere) is coordinated by two GG repeats through interactions of the Ca²⁺ ion with the carboxylate side chain and two carbonyl oxygens of the peptide bond per GG repeat.

A C-TERMINAL SIGNAL FOR TYPE I SECRETION

Evidence for a C-Terminal Secretion Signal and Determining the Minimal Size

In some early studies, it is noteworthy that type I secretion (apparently first so designated in reference 58) was often described in publications as a “signal-independent” secretion system. This was misleading, since it implies no secretion signal is required, while the proponents meant lacking an N-terminal signal as is the case for targeting the SecAYEG translocon. Many subsequent studies of the large number of known RTX proteins, where these have been tested, have confirmed the presence of a specific C-terminal secretion signal, i.e., an essential region carrying some form of code specific for docking to the T1SS translocon. Although detailed evidence is still surprisingly limited, we presume that the essential role for the secretion signal is recognition by and docking at least with the cognate ABC protein. However, at least for HlyA, as discussed in detail later, the secretion signal likely also specifically binds the MFP component of the translocon.

A potential signal required for type I secretion was first identified in 1985 by Gray et al., who showed that deletion of the C-terminal 27 residues of HlyA completely blocked secretion (40). Confirming the importance of this novel C-terminal signal, fragments constituting the terminal 218 (23 kDa) or 113 residues (12 kDa) of HlyA were shown by the Holland group to be autonomous for secretion, dependent on HlyBD (43, 59). Subsequently, a HlyA autonomous fragment of only 62 residues was described (60). Finally, using a strategy effectively of upstream internal deletions, Koronakis et al. (33) localized the HlyA secretion signal to the terminal 50 or so residues.

Another autonomous type I secretion targeting region of 50 residues was identified for the alkaline protease from Pseudomonas aeruginosa by Duong et al. (61). For the secretion of the protease PrtG from E. chrysanthemi (expressed in E. coli), Ghigo and Wandersman (56) showed that the C-terminal peptide of 56 residues could be secreted autonomously. Then by constructing internal deletions close to the C terminus, these authors found that retention of only the terminal 29 amino acids was sufficient to promote secretion (albeit at 50% of wild-type [WT] level).

However, the secretion signal for metalloprotease B, also from E. chrysanthemi (62), was located within the terminal 40 residues, and, more recently for a lipase, TliA secreted from P. fluorescens, a secretion signal, was located within the C-terminal 105 residues. This region includes three tightly packed RTX, apparently only separated from the likely secretion signal by around 20 residues (63). Similarly, Angkawidjaja et al. (64) described the efficient secretion of alkaline phosphatase fused to the C-terminal 98 residues of the lipase PML (I.3) from Pseudomonas. This includes 5 closely packed RTX motifs upstream of the secretion signal. For the 1,706-residue adenylate cyclase toxin (CyaA) the secretion signal has only been mapped by C-terminal deletion, and an approximate location was indicated within the C-terminal 75 residues. However, this deletion analysis by Sebo and Ladant (65) was complicated by the finding that two alternative or secondary secretion signal regions could also be detected further upstream. Finally, the secretion signal for HasA, the small hemophore from Serratia marcescens, was reported to be within the 56 C-terminal amino acids (66).

Posttranslational Secretion but Where Are the Chaperones?

The early studies of type I secretion, demonstrating the role for a C-terminal targeting signal, evidently showed that this must be a posttranslational process. This was exciting but raised further questions. For example, how could a very large polypeptide like HlyA, presumably requiring approximately 70 seconds at 37°C to complete its synthesis (assuming a synthesis rate of 15 amino acids per second), be maintained in a secretion competent form, neither aggregating nor being degraded, until the C-terminal docking signal was available? Possible explanations might include extended retention by the ribosome with coupling to transcription-translation of the transport proteins (as recently demonstrated for assembly of the luciferase complex from Vibrio harveyi [67]); early tethering to the translocator; segregation to a particular (protected) cellular compartment, capture by chaperone(s); or, less likely, the inherent stability of the unfolded or partially folded state of HlyA and other type I proteins. All these are plausible but most are predicated on the supposition that an additional signal near the N-terminal region would be required for early recognition of nascent forms of type I proteins. However, there is no evidence for such signals. This is especially puzzling regarding the many type I proteins now known to be composed of thousands of residues. Moreover, as discussed below, an enormous variety of heterologous passengers can be secreted when fused upstream of the C-terminal of type I proteins. In the exceptional case of the non-RTX HasA protein, the Sec-system chaperone SecB is essential for secretion (68) and one study has indicated that the general chaperone GroES is required for HlyA secretion (J. Whitehead and J. M. Pratt, University of Liverpool, unpublished results), but this has not been confirmed. Moreover, no chaperones have been implicated in the secretion of any other type I proteins. Therefore, it remains unclear how nascent type I polypeptides remain secretion competent long enough to make a successful docking with the translocon.

A plausible alternative scenario envisaged in a number of previous studies that avoid the need for chaperones is that type I proteins initially fold up rapidly, with the ABC protein involved in coordinating the subsequent unfolding and insertion of the protein into the transport pathway. The recent studies by Bakkes et al. (69) to be discussed below, however, have ruled out such a prefolded state, at least for HlyA.

The C Terminus of HlyA Promotes Secretion of a Wide Variety of Unrelated Proteins

Clear confirmation that a specific C-terminal region signal was necessary and sufficient to promote secretion was obtained by Mackman et al. (43). They fused the C-terminal 218 residues of HlyA (now called HlyA1, containing 3 RTX; Fig. 4) or by fusing the 102 C-terminal amino acids of HlyA (HlyA3, no RTX) to the C terminus of E. coli porin OmpF, lacking its normal N-terminal signal. In addition, Gentschev et al. (41) and Hess et al. (70) later reported that alkaline phosphatase, normally an E. coli periplasmic protein, was secreted to the medium when its C terminus was fused to an even smaller C-terminal fragment of HlyA containing only 60 terminal residues. However, a recurring caveat here is that, in this and many other reports, the efficiency of secretion per molecule synthesized was not measured, rendering the significance difficult to evaluate.

Figure 4.

Cartoon representation of the HlyA constructs mentioned in the text. The length of the constructs is scaled to their number of amino acids, which are provided for each construct. The major pore-forming domain of HlyA is shown in red, and the secretion signal is in blue. The individual nona repeats in the RTX domain are shown as vertical green bars. N and C termini are indicated. The fragment of HlyA, A1 contains 208 amino acids, A2 contains 160, and A3 contains 102.

Subsequently, a very large number of heterologous proteins, mostly fused with the larger (218 residues) C-terminal region of HlyA that contains three RTX motifs, were successfully demonstrated. Moreover, a wide variety of such heterologous passengers formed from cytoplasmic or Sec-dependent, exported proteins were shown to have functional activity, indicating normal folding. The examples include streptokinase (71), beta-lactamase (38), maltose binding protein (69), mammalian intestinal fatty acid binding protein (72), green fluorescent protein (GFP), and alkaline phosphatase (64, 73).

THE RTX MOTIFS: STRUCTURE AND FUNCTION

RTX Are Essential for Secretion of Many but Not All Type I Proteins

The RTX family is now known to be very large and includes, for example, a wide range of enzymes and adhesins in addition to toxins. Nevertheless, the RTX terminology, however defined, still rather tightly describes a large family of prokaryotic proteins whose secretion is dependent on a C-terminal signal and an ABC transporter. Many studies have also shown that RTX motifs play a role in secretion to the medium of many type I proteins, in particular, very large proteins. One of the best examples is provided by studies of the secretion of a Pseudomonas I.3 lipase. Thus, secretion levels were clearly shown to be proportional to the number of RTX motifs retained in different engineered constructs, with barely detectable levels of secreted lipase when 11 or 12 of all the repeats in the WT were removed (74).

In some contrast, it is equally clear that the RTX motifs are dispensable for secretion in some contexts. Thus, type I substrates like the small HasA protein and the bacteriocin colicin V completely lack the RTX motif (Fig. 3). HasA, although having a C-terminal secretion signal, has the special feature that secretion requires the dedicated chaperone SecB, which interacts with the N terminus of HasA. More surprisingly, although secretion is still dependent on an ABC transporter, the 39 residues of ColV, constituting the signal sequence, are located at the N terminus of the allocrite. Moreover, unlike HlyA and other large type I polypeptides, the ColV signal sequence is cleaved and the signal removed by a cysteine protease domain found at the N terminus of the cognate ABC transporter (75, 76). As discussed later, a protease domain in the ABC protein is not required for secretion of polypeptides. Instead, fascinatingly, an inactive relic of the protease is essential for secretion of HlyA and some other type I proteins. Such variation in the details of the process among different type I subgroups will recur frequently in this review.

In another specific exception to the rule, the minimal autonomous fragment of HlyA, the C-terminal 60 residues, completely lacks RTX but is still apparently secreted quite efficiently. In addition, when certain heterologous “passenger” polypeptides, for example, E. coli porin OmpF (43) or alkaline phosphatase (41), were fused to the C-terminal fragment HlyA3 that lacks RTX repeats (Fig. 4), secretion was still obtained. However, these particular passengers have the inherent properties of being normally exported across the inner membrane, and like WT HasA, require the SecB chaperone for normal secretion. They also share another common characteristic, relatively small size. In such cases, the role performed by the RTX, to facilitate calcium ion-dependent folding of substrates emerging from the translocon on to the cell surface, may simply be redundant. Indeed, as suggested by Letoffe and Wandersman (57), many of these results can reasonably be explained, if beyond a maximum size, around 30 kDa, the RTX motifs become essential for secretion (and folding).

Fusion of heterologous passengers to other T1SS secretion signals has also demonstrated successful secretion in the absence of RTX motifs (57, 64, 76). In some cases, the passenger enzyme in the fusion was shown to be active, despite the lack of RTX. However, in the case of a GFP fusion to the signal region of lipase TliA from P. fluorescens, the RTX motifs were apparently required for correct folding of GFP (63). Finally, as described below, some naturally occurring type I proteins completely lack RTX, but evolution has thrown up some fascinating alternatives.

Calcium Ion Regulation of Protein Folding: Role of RTX Motifs in Type I Secretion

In vitro experiments with HlyA and fragments thereof, ranging from those with the complete number of RTX repeats down to truncations having only three repeats, demonstrated that Ca²⁺ is essential for the folding of these proteins (77–79). Folding only occurs in the presence of Ca²⁺ or other divalent ions such as Sr²⁺ or Ba²⁺ but not with Mg²⁺. Interestingly, folding studies with purified HlyA following its secretion and subsequent unfolding by treatment with EDTA and 8 M urea, indicated that Ca²⁺-induced folding is not only restricted to the RTX domain but, more importantly, extends to the entire protein (79, 80).

This latter observation is supported by folding studies in vitro of maltose binding protein (MBP) or MBP slow folding mutants fused to the C-terminal fragment of HlyA, HlyA1 (Fig. 4 and reference 69), where passengers are folded to give biological activity. As expected, the purified mutant MBP proteins themselves displayed a decreased folding rate, while the velocity of unfolding was not affected. This was also observed with the MBP-HlyA1 fusions. Importantly, in the absence of Ca²⁺, essentially simulating the E. coli cytosol, the slow folding mutants displayed an even slower folding rate when fused to the C-terminal fragment of HlyA. This suggested that long-range interactions were propagated between the MBP and the HlyA moiety. Moreover, the slower the folding of a particular mutant, the greater the influence of Ca²⁺ ions on the folding of the entire fusion protein. This result raised the surprising possibility that the RTX in the absence of Ca²⁺ (as in vivo), rather than accelerating folding, can actually slow down the process. Thus, facilitating maintenance of the unfolded state until translocation and surface emergence of passenger domains, their consequent folding and release into the medium is completed. In this sense the RTX exert a negative effect on intracellular folding contrasting with its positive driving of extracellular Ca²⁺-dependent folding.

The Mechanism of Folding of Adenyl Cyclase Toxin

Detailed studies of calcium ion-dependent folding of the RTX motifs into the characteristic β-roll structures (first identified by reference 5), in an elegant series of experiments concerning the adenylate cyclase toxin (CyaA), have provided fascinating insights into the mechanism. Adenylate cyclase toxin secreted by B. pertussis contains around 40 RTX motifs (allowing for some relaxation of the strict consensus) distributed into 5 blocks within the terminal 700 amino acids of the 1,706-residue toxin. This so-called RD (RTX domain) region is required for binding of the toxin to its integrin receptor on target cells and also contains the approximately 75-residue secretion signal. The last RTX block, finishing around residue 1620 (81), appears to be separated from the secretion signal by a maximum of only 30 amino acids.

Ladant and Chenal and colleagues have shown, using spectroscopic techniques, biochemical and hydrodynamic measurements that the entire RD domain of CyaA, including the secretion signal, in the absence of calcium ions adopts a premolten globule or intrinsically disordered structure. Notably, this contains small amounts of unstable β-sheet (80). Addition of Ca²⁺ (10 mM, mimicking physiological extracellular levels) results in cooperative binding of approximately 1 calcium ion per repeat to form a compact β-roll structure. The folded structure has a reduced net charge, suggested by the authors to reflect neutralization of internal electrostatic repulsions of the aspartates by calcium ions. More detailed subsequent studies, concentrating on in vitro folding of RTX block V (9 RTX, approximately residues 1529 to 1620) and its C-terminal flanking region (CTF), demonstrated that this subdomain folded up the entire RD region in the presence of Ca²⁺ (summarized in reference 82). However, remarkably, when separated from the CTF, defined as the residues extending to residue 1680 (reference 81 and Fig. 1), the repeats were unable to bind calcium ions and did not fold. Even with calcium ions present, there was no folding of the RTX without the CTF. This clearly revealed a key role for some downstream amino acids in the initiation or stabilization of the folding process by a mechanism that remains to be elucidated.

Blenner et al. (83) also investigated the mechanism of folding of the block V RTX motifs of CyaA and the role of a similar CTF region. The results also showed that truncation of the CTF resulted in reduced affinity for calcium ions and β-roll formation. Importantly, the authors noted that the CTF itself is rich in aspartic acid residues, and, in fact, replacing this region with a completely unrelated sequence, but nevertheless able to bind calcium ions, restored the ability of the upstream RTX to fold in the presence of calcium ions. It was therefore suggested that the stimulating effect of the CTF was dependent on an entropic stabilization effect.

Largely similar results for calcium ion-dependent cooperative folding of the RTX region of C-terminal fragments of alkaline protease secreted from P. aeruginosa were also obtained by Zhang et al. (84). This study interestingly also demonstrated that the folded RTX domain was directly involved in chaperoning the folding of the N-terminal protease domain of this protein.

BIOTECHNOLOGICAL APPLICATIONS OF THE TYPE I SECRETION PROCESS

Introduction

A large number of pro- and eukaryote proteins or fragments of proteins have now been successfully secreted as fusions to C-terminal regions of HlyA and, in several cases, shown to fold correctly (69, 85–88). We recently developed a two-vector system coding for the inner membrane components (HlyB/HlyD) on one plasmid and the fusion protein (gene of interest fused to the N terminus of HlyA1, see Fig. 4) on the other plasmid. Here, successful secretion of slow folding mutants of MBP and an internal fatty acid binding protein was demonstrated (69, 72). Yields were approximately 15 mg/liter of cell culture for the slowest folding mutant of MBP and approximately 1 mg/liter of cell culture for intestinal fatty acid binding protein (IFABP).

Similar results were obtained more recently using C-terminal signals from other type I proteins. In particular, a minimal C-terminal fragment composed of 103 residues, including 4 RTX, from the P. fluorescens Tli lipase was used for secretion of epidermal growth factor (EGF) or GFP (63). In addition, the terminal 60 residues of another pseudomonad lipase, PML although lacking RTX, were also able to support secretion in a fusion with alkaline phosphatase as passenger (64). A new vector for use in P. fluorescens was described more recently by Ryu et al. (89) that harbors genes for the ABC transporter and the MFP. In this case, genes of interest are fused to the C terminus of the lipase transporter recognition domain (LARD) from the lipase TliA. Successful secretion was shown for GFP and alkaline phosphatase. These examples demonstrate that new approaches are also underway in other laboratories to employ T1SS as a platform for protein production and purification.

Optimizing Secretion of Fusion Proteins: the Rate of Intracellular Folding of HlyA Dictates Secretion Efficiency

It is important to note that, although many different heterologous proteins have been secreted using the C-terminal region of RTX proteins, the great majority of such studies have taken a simple empirical, trial-and-error approach: are particular fusions secreted or not? In addition, in most such experiments, protein levels were not measured quantitatively and no attempt was made to determine secretion efficiency, i.e., the number of molecules secreted compared with the number synthesized. Moreover, when measured, yields of secreted fusions were found to be very low, restricted to the high microgram per liter range at best.

Recognizing the limitations of the empirical approach, Bakkes and colleagues recently examined how at least one key factor, fast folding of a passenger domain, might compromise efficient secretion of a given fusion (69). Debarbieux and Wandersman (90) had previously shown that, if HasA (albeit an atypical non-RTX type I protein) was allowed to fold in the cytoplasm before inducing the synthesis of the translocator proteins, it could not be secreted. This result, in fact, appeared to contradict earlier suggestions that type I allocrites might require unwinding (by the ABC protein) prior to secretion. Bakkes et al. (69), taking up this idea, conducted a detailed study of the prototype Hly-T1SS using folding variants of the maltose binding protein (MalE), fused upstream of the C-terminal secretion signal of HlyA via the HlyA1 fusion vector described in the next section. The MalE mutants varied only in their folding properties, with no effect on the unfolding rate. The results showed clearly that the slower the folding rate, the higher the secretion level of the fusion protein. Thus, the folding rate dictates secretion efficiency. Importantly, with the slow-folding mutations that promoted efficient secretion, the activity of MalE in the fusion was not compromised.

These results clearly ruled out the possibility that HlyA folds in the cytoplasm before being secreted, and also demonstrated for the first time experimentally that transport substrates of a T1SS only fold after secretion. More importantly, these studies (69) provided a rational platform for the construction of T1SS fusion proteins. If these are secreted poorly, decreasing the folding rate by introducing point mutations is recommended. This strategy was supported by secretion studies on a fusion with intestinal fatty acid binding protein (IFABP) . Here, mutations that slowed the folding rate of IFABP increased secretion substantially, again without apparently compromising activity (72). However, mutations that decrease folding of a certain protein cannot be predicted per se but have to be determined experimentally. Thus, error-prone PCR or DNA shuffling could be the best approach, combined with screening for increased secretion levels.

In addition to the importance of the folding rate of the passenger, the expression levels of the inner membrane components of the HlyA T1SS (HlyB and HlyD) are also of prime importance for high yields of secreted protein. Only under conditions in which both membrane proteins were highly expressed was efficient secretion observed (69, 72). However, efficient secretion of heterologous fusion proteins, in general, is likely to be a multifactorial process dependent on several factors, in addition to folding rate and expression levels. Nevertheless, these studies provided an important starting point and, in our hands, the Hly T1SS is a very powerful system for efficient, high-yield secretion.

Generation of Novel Hly Fusion Vectors

A vector encoding the 23-kDa C-terminal of HlyA (HlyA1) was constructed in plasmid pSU. This was designed for easy insertion of coding sequences for different passenger proteins N-terminal to the secretion signal, plus engineered cleavage sites for subsequent enzymatic release of the passenger domain. pSU was combined with plasmid pLG575, encoding both HlyB and D, to promote secretion of different heterologous passengers (72). This HlyA1, 208-residue fragment (see Fig. 4), contains only the distal 3 RTX repeats of WT HlyA plus the approximately 55-residue secretion signal. In addition, this C-terminal fragment of HlyA also contains, upstream of the signal sequence, the binding site for the C39 peptidase-like domain (CLD) of HlyB, as discussed later (Fig. 2). This is the specialized cytoplasmic N-terminal domain that is required for secretion of HlyA, but is also required for secretion of heterologous proteins fused to HlyA1 and smaller fragments from the HlyA C terminus.

ANALYSIS OF THE STRUCTURE-FUNCTION OF THE SIGNAL: AN ENIGMATIC CODE

Type I C-Terminal Secretion Signals, Highly Conserved in Closely Related Transport Substrates, but There Is No Universal Code

The presence of specific C-terminal secretion signals for a wide range of type I proteins is now well established to lie within the terminal 50 to 70 residues. However, identifying the precise function of the signal, in particular, the coding information required for docking with the translocon, has proved difficult.

We previously (91) compared the primary sequence of the C-terminal 60 residues of HlyA and a wide range of functionally quite different type I proteins, including a representative protease, a lipase, degrading enzymes for different surface polymers, S-layer proteins, iron-regulated proteins, different cytotoxins, and a surface protein required for gliding motility. Remarkably, this analysis revealed no obvious amino acid conservation of these proteins either with HlyA or with each other. In contrast, in our recent survey of the last 60 amino acids of 40 closely related hemolysins, we found a pattern of highly conserved residues (see Fig. 5C). Similarly, confirming earlier published reports, we have observed in a smaller sampling the conservation of a different set of C-terminal residues among closely related proteases or lipases. These results, and the more detailed studies discussed below, clearly indicate that there is no single code specific for all type I proteins. Rather, each subgroup appears to retain a largely unique code.

Figure 5.

Models for the nature of the secretion signal code. (A) Predicted secondary structure of the C-terminal secretion sequence of HlyA. (B) The Ling model derived from directed combinatorial mutagenesis (104) emphasizes the functional importance of the proximal helix I plus the extreme terminal residues. (C) The linear code model (100, 101) emphasizing individual key residues (highlighted in red) essential for function in the hemolysin subfamily, as derived from single-substitution mutagenesis. (D) Secondary structure predictions for the C-terminal region containing the secretion signal sequence of representatives from other subfamilies of T1SS transport substrates. Note in (D) that β-sheet or α-helical regions are predicted for the C-terminal signal region with little conservation of the primary sequence.

Early studies indeed suggested that at least two subtypes, hemolysin-like and PrtG protease-like groups, can be distinguished at the extreme C terminus, respectively, by a short motif of hydroxylated residues or by three hydrophobic amino acids preceded by an acidic residue, DVLA (56, 92, 93). Nevertheless, the hydroxylated tail is not present in all toxins described as hemolysins, for example, LktA from Pasteurella haemolytica (Fig. 5D). Interestingly, a C-terminal “motif” of hydrophobic amino acids preceded by an acidic residue appears conserved not only in proteases, but also in lipases, for example, LipA_SM (93). However, other type I substrates, such as adenyl cyclase toxin or the newly studied ankyrins (described below), have C termini lacking either of the hydroxylated or hydrophobic motifs.

Although based on only very limited studies, type I allocrites and components of their translocon also appear to be interchangeable only between systems with very similar transport substrates (see, for example, references 48, 61, 62, 75, 94, and 95). In most cases, this appears to parallel the relative conservation of the inner membrane transporter components, for example, more than 80% conservation within the distinct HlyA or the protease-lipase type I systems, but only around 25% between subfamilies (see, for example, reference 96). More puzzlingly, the C-terminal 70 amino acids of the leukotoxin, LktA from P. haemolytica), although seemingly with no obvious similarity to the HlyA signal sequence (Fig. 5C and D), is functionally able to replace the C-terminal 58 amino acids of HlyA (97). There have been other claims of the HlyA translocon appearing relatively promiscuous, able to secrete, to some degree, some very unexpected proteins. These include the CyaA toxin (65, 95), protease B from E. chrysanthemi (62) and especially the non-RTX protein, bacteriocin ColV (98) with its completely unrelated N-terminal signal. However, apparently none of these heterologous systems are able to reciprocate and secrete HlyA. Moreover, with a note of caution these supposed examples of relaxed specificity, often detected only with antibodies and not quantified, might in reality simply reflect very inefficient secretion.

HlyA C-Terminal Signal: Individual Residues Dispersed along the Signal Region Are Important for Function

The great majority of attempts to elucidate the nature of the code carried by type I secretion signals have concerned the prototype HlyA and most of the following sections will be focused primarily on this system. Note that, as indicated above, most studies have involved two hly determinants, one located in the chromosome from a pathogenic strain LE2001 (34, 35) and a second encoded on a plasmid, pHly152 (30), with total lengths of 1,023 and 1,024 amino acids, respectively. To avoid confusion in the following discussion involving studies in different laboratories, residues will therefore be identified with respect to their position from the C terminus.

Early random mutagenesis studies from different groups (results summarized in reference 100) immediately suggested that many residues within the signal region were redundant, with many single-residue substitutions having very little or only moderate effects on HlyA protein secretion levels. However, as highlighted in Fig. 5C, 60% to 70% reductions in secretion were obtained in 8 mutants carrying substitutions at positions −46 (Glu), −43 (Lys), −39 (Ala), −35 (Phe), at −45 (Ile), and reductions around 50% at −15 (Asp), −10 (Arg), −5 (Leu). Importantly, construction of different combinations of substitutions at residues E-K (−45) F-L (−35) D-R (−15) gave additive effects, maximally reducing secretion to less than 2% in the triple mutant, as measured by hemolytic activity in liquid cultures and confirmed by SDS-PAGE. Overall, these latter studies also placed the N-terminal boundary of the signal to between position −46 and −53 (100).

A Structural Code or a Linear Code of Individual Residues

Although we are only concerned here with HlyA, it is important to note that, if there is a structural code based on helical secondary structure (therefore able to pass through TolC) it is not conserved in the protease-lipase subfamily, since these proteins appear to have β-sheet rather than helices in the C-terminal region.

As illustrated in Fig. 5, simple inspection of the sequence of the HlyA signal region indicates several characteristics first considered as potential coding information. These are a charged cluster covering residues −29 to −32; a 16-residue “aspartate” box (−30 to −15) flanked by acidic residues that appeared to be conserved in some other T1SS transport substrates (101), a block of 13 uncharged residues, distal to the charged cluster and overlapping the aspartate box, and, as mentioned above, a cluster of relatively hydrophobic, mostly hydroxylated residues at the extreme terminus (33, 102). However, several mutagenesis experiments to test the possible role of various motifs, in fact, found no functional role for the “aspartate box,” and the charged cluster could also be mutated or even deleted without significant loss of function (97). Similarly, mutagenesis experiments have apparently found no functional role for the 13-residue uncharged region, while mutation or deletion of the extreme C-terminal 6 or 7 residues, as discussed below, affects posttranslational folding rather than secretion per se.

There is also confusion in the literature regarding the precise position of two predicted α-helices in the signal region (43, 102, 103, 104), reflecting how the precision of algorithms has markedly improved. For the purpose of this review, we will use the prediction shown in Fig. 5A.

In line with most subsequently published reports, these α-helices will be referred to as the most proximal (−37 to −47), i.e., close to the N terminus of the signal, and a downstream helix (hereafter referred to as helix 2, −21 to −26). However, most studies subsequently concentrated on the possible coding role for potential amphipathic helices, overlapping the proximal predicted helix (see Fig. 5A). Thus, Koronakis et al. (33) first postulated a potential 18-residue amphipathic (helix 1) located between amino acids −34 and −51, centered on the sequence EISK (−42 to −45). The identification of an amphipathic helix was based primarily on presentation of the 18 residues on a Schiffer-Edmunson helical wheel and, for simplicity, we will call this helix 1 (or H1) from now on. Then, Stanley et al. (102) provided some mutagenesis data and a helical wheel analysis supporting a functional role for a larger 26-residue amphipathic helix (residues −23 to −48) that we will refer to as helix 1′ (H1′). Importantly, this includes the original amphipathic helix 1 but extends to include both the charged cluster as part of helix 2.

In brief, Stanley et al. (102), using internal deletions or random mutagenesis, including the use of degenerate oligonucleotides, identified mutants carrying multiple as well as some single substitutions in the C-terminal signal region, consistent with H1′ having a functional role in secretion of HlyA. Notably, other studies (92, 100, 101), using random-site-directed and -saturated mutagenesis of specific parts of the signal region, found a number of point mutants either consistent or equally clearly not consistent with such an amphipathic helix. Importantly, these latter studies included the demonstration, from experiments where the WT and a poorly secreting mutant were coexpressed in the same cell that secretion of the WT was not affected. That is, the mutations were recessive, indicating that the mutant signals were not engaging the translocon. We wish to emphasize here that, for nonsecreting mutants, without determining the location of HlyA, it is essential to determine the location of HlyA. Is it in the cytoplasm (recessive mutants), stuck in the translocon (dominant mutants), or secreted to the medium, but then degraded or aggregated? Unless this analysis is done, the conclusions that can be drawn regarding the mechanism of secretion, including the role of the signal sequence, are limited. Unfortunately, such further investigation has usually not been pursued.

Ling and colleagues (104–106) used a different approach to the coding problem, a combinatorial analysis inserting large numbers of random peptide sequences into specific regions of the signal region. The results obtained were consistent with a functional role for a proposed 10-residue amphiphatic helical structure (−36 to −49) encompassing the “secretion hot spot” EISK (92, 100) that we will return to later. In contrast, a combinatorial library targeted on the terminal 5 amino acids of helix 1′ (and a further 6 residues downstream) revealed no functional role, and similarly for the combinatorial analysis of residues −8 to −16 (see Fig. 5 for the residue map). In many of these experiments, hemolytic colony halo size was as a measure of secretion efficiency, with controls to establish the quantitative nature of the procedure. However, in our experience, a large reduction in the amount of secreted protein is required before any obvious change in halo size is evident. Thus, this is a useful but a rough guide to secretion changes. However, for screening signal sequence mutants, the method is limited to analyzing the proximal region of the signal since, as described, later the most distal region is involved in folding of HlyA and therefore activity.

Other studies involving internal deletion analysis specifically examined the role of the charged cluster, DVKEER (−28 to −33) within helix 1′, a crucial contributor to amphipathicity. However, these experiments gave contradictory results. Thus, while reference 102 reported that deletion of 11 residues −27 to −37 abolished secretion, the deletion of only the charged cluster still permitted 71% of the WT secretion level of HlyA (97). These results suggest rather that, in the larger deletion, a residue just outside the charged cluster is important for secretion and an obvious candidate is the phenylalanine at position −35, which is crucial for high-level secretion of HlyA (92, 100).

In summary, from the extensive mutagenesis experiments described above, most possible coding motifs visible in the primary sequence could seemingly be eliminated. Moreover, overall, the mutagenesis results do not support a functional role for the large amphipathic helix 1′. However, the results of the Ling studies remain compatible with a short potential amphipathic helix (residues −38 to −49) having a functional role. On the other hand, mutagenesis experiments have identified 8 individual amino acids critical for function, including four within H1 and a new amphiphatic helix defined by the Ling group, the −37 to −44 region. We will return to this important point concerning a linear code rather than a code based on secondary structure.

Physical Techniques in the Search for Secondary Structure in the Secretion Signal

Is there a linear amino acid code, particular secondary structure, a mixture of both, or something entirely different that determines recognition of the signal sequence by the type I translocator? First, it is important to keep in mind that any secondary structural component of the signal code in intracellular HlyA must be able to transit the TolC channel, whose maximum diameter is likely restricted to 20 Å (107). This therefore limits secondary structure during transit to an α-helix.

Nuclear magnetic resonance (NMR) studies of isolated type I signal peptides have been described, but these have not provided definitive evidence for any specific structure. The earliest studies for HasA and PrtG (108, 109) found no secondary structures in aqueous solution. In membrane mimetic environments, with the somewhat ambiguous inclusion of the helix-promoting agent trifluoroethanol, some helical content was detected in the proximal region of the signal, while the rest appeared unstructured. Essentially, similar results were obtained with intact HasA in detergent micelles (110). A circular dichroism (CD) spectrum analysis of the C termini of HlyA (61 residues) and LktA (70 residues) also indicated unstructured peptides in aqueous solution with some helical secondary structure in detergent (111). Similar results were obtained with an NMR analysis of the HlyA C-terminal secretion signal (61 amino acids) in SDS (106). These results were interpreted, although with a number of caveats, to indicate the presence of two short helices (see Fig. 5A). Importantly, all the experiments involving detergents were apparently predicated on the idea that in vivo the signal-ABC protein interaction takes place in a membrane environment, although this has not been substantiated experimentally for any type I protein.

Notably, in vitro structural analyses in an aqueous medium of the C termini of the type I proteins, adenylate cyclase toxin (80–82) and the AprA protease from P. aeruginosa (84), indicated that, in the absence of calcium ions, the C terminus is largely disordered. The region analyzed in these experiments included the RTX Ca²⁺-dependent folding domain and the proximal region of the secretion signal, at least up to residue −26 from the terminus (81).

Finally, β-strand structures at the C terminus can be predicted for some proteases and lipases (see Fig. 5C). Moreover, the few available crystal structures for such proteins have antiparallel β-strand sheets in that region (5, 113). However, it is noteworthy that β-strand structures, if formed in vivo, are not known to have amphiphilic properties and are too bulky to transit TolC.

In summary, amphiphilic helices are not widely conserved in signal sequences of type I allocrites, and overall the physical evidence for helices in HlyA and HasA is weak, while the CyaA C terminus in vitro indeed appears disordered. Therefore, together with inconsistent genetic evidence for a specific functional helix in the HlyA signal, this suggests that the type I secretion signal in vivo lacks any functional secondary structure.

The C-Terminal Signal of Type I Allocrites May Have a Dual Role

Type I secretion signal regions appear to have at least two functions, a proximal recognition region (perhaps extending from around −15 to −46 in HlyA), while a few residues at the extreme terminus appear to provide a distinctive function, affecting in some way posttranslocation folding. Thus, end deletion constructs, measuring secretion efficiency based on hemolytic activity in liquid cultures (33, 102), reported that removing the C-terminal 7 or 8 amino acids of HlyA reduced secretion by 50% to 70%. Importantly however, later studies showed that removal of the 6 terminal residues or multiple mutations in this region of HlyA had minimal effects when levels of HlyA protein in the medium were analyzed. We concluded that such deletions had a major effect on folding of HlyA to the native, active form, as indicated by hypersensitivity to trypsin and tryptophan fluorescence spectroscopy (92, 103). These results therefore suggested an important functional role near the terminus of the signal, affecting folding in some way, rather than translocon recognition, as required in the model of Stanley et al. (102). Interestingly, mutations within the terminal 4 residues of certain lipases, a motif also relatively well conserved in secreted proteases (115), resulted in reduced stability of the secreted protein. In addition, the deletion of the C-terminal 14 residues of HasA stalled translocation in some unknown way, indicating that this region is not involved in recognition of the translocon (114). Finally, in vitro studies with CyaA have indicated that at least the proximal part of the secretion signal may be implicated in RTX-driven, posttranslocational folding. However, alternatively, we might speculate that a few residues at the extreme C termini of type I proteins are required for anchoring the emerging polypeptide to some surface component.

Secretion Signal of HasA

The minimal HasA C-terminal fragment capable of autonomous secretion was found within the last 56 residues (66). Subsequently, Cescau et al. (114) constructed a nonsecreted variant of HasA with the C-terminal 14 amino acids deleted. The truncate was able to fold in the cytoplasm (and therefore bind heme) but was still able to engage the ABC and MFP proteins and to assemble a complete, but stalled translocator that included TolC (shown by affinity purification of the complex). Indeed, with the high level of production of the ABC and MFP proteins employed in these experiments, the majority of cellular TolC was titrated, rendering the cells sensitive to SDS, presumably by depletion of the AcrAB drug and detergent transport pump. As described below, SDS sensitivity can be a useful indicator of stalled translocons.

In addition, Cescau et al. (114) provided preliminary evidence that the truncated signal fragment bound the ABC protein HasD in vitro. These results, unfortunately not apparently followed up, suggested that the terminal 14 residues of the HasA secretion signal are not required for docking but might have a role, for example, in stabilization of the secreted protein.

Possible Coding Motifs in Lipase and Protease Secretion Signals

Other more limited studies have been performed to identify possible conserved motifs and essential functions in the secretion signals for lipases and proteases from different Gram-negative species. As also illustrated in Fig. 5 (with representative examples only), the C-terminal sequences and putative secretion signals for proteases and lipases reveal some apparently distinctive features, although these appear quite different from those of both HlyA and HasA.

Ghigo and Wandersman (56) first suggested that the extreme C-terminal sequence DLVL was a secretion ‘motif’ in both proteases and lipases. Indeed, for the protease PrtG, deletion of this sequence prevented its secretion. However, subsequent studies found that a similar motif, although present, was not required for secretion of lipase LipA from S. marcescens (93). In a similar study of the I.3 (PML) lipase by Kuwarhara et al. (115), the hydrophobic residues at the extreme C-terminal motif were also found not to be required for secretion. In addition, these authors showed that the extreme C-terminal motif in the I.3 lipase was required for heat stability of the protein rather than for secretion per se.

S. marcescens also secretes the heme binding protein HasA and, intriguingly, the dedicated HasA T1SS is able to secrete, to some extent, PrtA and LipA from this same host. However, reciprocal secretion of HasA_SM was not observed. Comparison of the two HasA sequences in fact identified a short motif around residue −17 from the C terminus present only in the Pseudomonas HasA. Site-directed mutagenesis confirmed that residues −15 to −19, VTLIG (forming a β-sheet structure in the fully folded proteins) were indeed critical for secretion (93). Based on the testing of hybrid translocators from LipA and HasA translocons, the authors suggested that the motif was specifically involved in targeting the ABC transporter. Moreover, reference 115 confirmed the importance of the conserved VTLIG for secretion of a lipase (PLM) from Pseudomonas. It would be important now to test directly the binding of such a motif to the relevant ABC proteins.

Screening Tests for Functional States of the Type I Translocator

The article by Cescau et al. (114), exploiting SDS sensitivity of cells to detect a frozen complex, highlights the use of simple plate tests for easy monitoring of certain states of the translocon. Thus, Blight et al. (116) observed an increase in vancomycin sensitivity (but not to another antibiotic, fusidic acid) in E. coli K12 expressing hlyBD and hlyA. However, vancomycin resistance was restored when a highly defective, secretion signal mutant of HlyA was present, indicating that assembly of the whole translocon and active translocation of HlyA increased envelope permeability to vancomycin in some way. Surprisingly, a similar study (117), however, found that the presence of HlyB and D was sufficient to sensitize E. coli to the antibiotic, and the reason for these differing results is unclear. In any case, changes in vancomycin sensitivity can be used to screen mutants or the effects of suppressors to test for changes in the function of type I translocons.

Competition experiments, coexpressing WT and mutant forms of type I proteins in the same cell and in the presence of the transport functions, are also useful to distinguish mutants (recessive) blocked in binding to the translocon and mutants (dominant) that engage the translocator, but fail to proceed to the next stage, and therefore titrate available translocons, consequently inhibiting secretion of the WT transport substrate. Combining this with testing for sensitivity to SDS or vancomycin should provide a quick screen for classifying mutants.

THE HlyA SECRETION SIGNAL: PERHAPS NOT SO ENIGMATIC AFTER ALL

Alternative Functional Models

As described above, comparative analysis of C-terminal sequences for T1SS proteins clearly rules out a universal code and the following discussion is restricted to the most studied system, E. coli hemolysin secretion. The Hughes-Koronakis group (102) proposed a complicated, but very specific model with particular importance ascribed to the presumed large amphipathic helix (helix 1′, Fig. 5). The model proposed that the initial step was insertion of H1′ into the membrane as a loop (independently of the transporter proteins, HlyB and HlyD), seemingly with the implication that this permits partitioning of HlyA into the membrane prior to docking with the translocator. Then, the C terminus of the signal, covering the last 8 or so residues of HlyA (including the hydroxylated “tail”), was suggested to dock with HlyB. However, the main tenets of the model have not been tested, in particular, the insertion of the signal region into the membrane or whether specific regions of the signal bind directly to the translocon. In fact, now it would generally be supposed that T1SS proteins enter the translocator via an aqueous chamber exposed to the cytoplasm. The model also appeared to suggest, although did not explicitly state, that the C terminus of HlyA was translocated last. This is clearly at odds with the now generally accepted hypothesis that the C terminus should be the first to reach the surface, and thus facilitate Ca²⁺-dependent folding of HlyA. Indeed, this has now recently been demonstrated experimentally by the Schmitt group (118). As described below, an eGFP-HlyA-fusion protein forms a stuck intermediate extending across the envelope with the C terminus exposed to the exterior.

Concerning the large helix in the Hughes-Koronakis model, in other studies the terminal 11 residues of the proposed helix 1′, including the charged cluster, were found to be dispensable for secretion. Furthermore, based on their combinatorial mutagenesis experiments, the Ling group (104) proposed a second alternative model for a structural code for the signal sequence. As represented in Fig. 5B, the model suggests that domain I, largely encompassing the predicted helical region shown in Fig. 5C (residues −37 to −47, covering the hotspot EISK), is essential for secretion. Domain I in this model (presumably assumed to dock with the translocon), is followed by a long downstream “connector” (residues −9 to −28). This region, as also found by others (100–102, 104), appears to contain little specific coding information. Finally, a relatively hydrophobic region of 9 residues extending to the terminus was identified, termed functional domain II, similar to that described in the Hughes-Koronakis model, also apparently essential for secretion. However, this region, as also discussed above, rather appears to be required for posttranslocational folding and/or stability.

Closing in on a Linear Secretion Code for the HlyA Subfamily

Both the two models above propose that the signal code depends on a helical structure forming in vivo prior to partitioning into the membrane or is induced by association with the membrane. In contrast, as described above, based on the identification of individual amino acids essential for secretion of HlyA, and in the absence of strong corroborating evidence that functional helical secondary structures actually exist in the HlyA signal sequence in vivo, Kenny et al. (100, 101) proposed a different model, a linear code constituted by up to eight key individual residues, distributed within the region −45 to −15 (see Fig. 5C). These were suggested to act cooperatively for specific docking with the transport proteins HlyB and HlyD. Notably, these key residues include the EISK motif. Therefore, if we discard the suggested helical nature of domain I, the Ling model would be compatible with this linear model. Indeed, Hui and Ling (104) specifically keep this possibility open, while also considering the possibility that certain key residues might only be functional when presented on helical structure. However, looking at all the evidence available, we clearly favor the simpler linear code for translocon docking, at least for the hemolysins.

Finally, recent studies indeed provide strong support for a conserved specific pattern of several individual amino acids in the hemolysins, with particular prominence of the motif EISK in the proximal region of the signal sequence (L. Schmitt, S. Pehersdorfer, K. Kanonenberg, and I.B. Holland, unpublished data).

T1SS TRANSLOCATION IN SALMONELLA, RICKETTSIA, AND OTHER BACTERIA: YET MORE VARIETY IN SECRETION SIGNALS AND SECRETION WITHOUT RTX

The MARTX Transporters

Some very recent findings are bringing surprises and new dimensions to T1SS. The group of Karla Satchell has recently pioneered the study of toxins secreted by the type I system in many of the most important pathogenic bacteria. These “giant” multifunctional-autoprocessing repeats in toxins (MARTX), up to 9,000 residues long, frequently contain strings of several distinct toxins. Many studies have concentrated on the structural organization, mode of penetration into host cells, and mechanism of action of these toxins (17). In fact, at least some of these proteins have a slightly variant form of the classical RTX repeat (consensus, GGXGXDXUX) located close to and upstream of the C-terminal secretion signal. However, calcium ion binding and the 3D structure of these MARTX repeats (consensus GGX(N/D)DXHX) has not been described, and, as yet, few details of the actual secretion mechanism have appeared. These ABC transporters, as described later, however, like HlyB, have an additional N-terminal domain, CLD (see Fig. 2). Intriguingly, Boardman and Satchell (119) showed that secretion of RtxA_VC, and other closely related toxins, requires a 4- rather than a 3-component translocon, with a second gene encoding an additional ABC transporter. The ATPase activity of both ABC proteins is required for secretion, and the authors propose that this may take the form of a heterodimer (see also review in references 24 and 120).

A Novel T1SS in Salmonella

Several other more recent studies have now identified a novel T1SS in Salmonella enterica with a number of unexpected features. This involves another giant type I transport substrate, (with over 5,000 residues) constituting a nonprocessed, nonfimbrial adhesin, SiiE, that was likely missed in the screen described by Linhartova et al. (6) for RTX proteins. Thus, this type I protein (121) contains 53 closely packed blocks of repeats constituting approximately 90% of the adhesin. However, these are not RTX repeats; nevertheless, each block (shown in Fig. 3) corresponds to about 90 amino acids, with highly conserved sets of dispersed Asp residues that bind Ca²⁺ ions (see also review [20]). Moreover, secretion still apparently depends on an approximately 60-residue secretion signal at the C terminus, with a translocon formed by ABC, MFP, and OMP components, all encoded in the same operon (122). Surprisingly, the siiE operon also contains two additional, proximal genes encoding membrane proteins, one with some similarity to the ExbB/TolQ membrane proteins (implicated in harnessing the proton motive force in bacteria for import through the outer membrane), and the other possibly similar to MotB (part of the proton-driven motor controlling flagellum rotation). The authors (123) suggested that these proteins have a novel accessory function possibly to form a proton-conducting channel, linked in some way to secretion or surface fixation of the transported polypeptide. It will be fascinating to discover how indeed these accessory proteins work and whether such partners are required for secretion of other giant type I proteins.

Novel Type I Proteins in Rickettsia and Bacteroidetes

Rickettsia form a large genus of aerobic, Gram-negative bacteria, living as obligate intracellular parasites in both invertebrate and vertebrate hosts and causing many diseases in humans, including typhus. Organisms such as Rickettsia typhi, have reduced genomes of fewer than 900 genes, relying on the host for biosynthesis, for example, of amino acids and nucleosides. The organism is frequently transmitted to humans by fleas, lice, and mites through bites, allowing the bacteria to access the vascular system and to establish residence in endothelial cells of the skin and major organs. So-called scrub typhus is endemic in the Asia-Pacific region with huge numbers of people dying annually of this disease.

Rickettsia and other intracellular bacteria produce several copies of ankyrin proteins as pathogenicity factors. These proteins characteristically contain a varying number of tandem repeats (mostly of 33 amino acids) of degenerated sequence but a conserved secondary structure, allowing stacking of the repeats upon each other (124). This presents a platform, maximizing opportunities for protein-protein interactions. Such broad specificity for protein interactions allows the modulation of the action of a wide variety of host proteins and therefore facilitates intracellular survival. One of the first groups to identify potential type I secreted proteins in Rickettsia was Wilson et al. (125), using a proteomics approach. In another recent review, Gillespie et al. (126) used a bioinformatics and literature search to identify likely secreted proteins and components of presumed translocation systems in Rickettsia. The results revealed at least 19 candidate secreted proteins, including some ankyrins, a near-classical Sec system, and genes encoding at least three secretion pathways, TSS 1, 4, and 5. In parallel, a number of groups, in particular (127), have provided intriguing evidence that many of the 47 ankyrins in the scrub typhus pathogen, Orientia tsutsugamushi (member of the Rickettsiales), are indeed type I transport substrates.

Indications of 7 putative type I protein translocation systems, as defined by the presence of linked genes encoding an ABC, MFP, and a TolC-like homologue, have been found in the Gram-negative, obligate anaerobe, Bacteroides fragilis (125). B. fragilis is a member of the phylum Bacteroidetes, the largest group of bacteria in the human microbiota, having interesting differences in the lipid composition of membranes and peptidoglycan, compared with the Proteobacteria. So far, no potential type I allocrite has been identified, nor sequences encoding RTX motifs in the B. fragilis genome, thus pointing to another interesting variation on how to fold up type I proteins following translocation.

FUNCTIONAL ANALYSIS AND ASSEMBLY OF THE TRIPARTITE TRANSLOCON

TolC the Outer Membrane Exit

Multifunctional properties of TolC: a rather nonspecific channel

While the MFP and ABC proteins are specifically and uniquely dedicated to the transport of type I substrates, TolC is an extraordinarily multifunctional and, under some conditions, apparently, an essential protein (128, 129). TolC in E. coli can act as an uptake site for colicins but more importantly forms the outer membrane exit of the tripartite pump with AcrA and AcrB (and other homologues), to efflux a wide range of toxic molecules. These include many cationic dyes, a wide range of antibiotics such as penicillins, detergents such as Triton X-100, SDS, bile acids, and even simple organic solvents (130), fatty acids (131), or cyclic AMP (132). Essentially, TolC appears to act as a rather nonspecific duct, maximizing its range of transported molecules, by associating with a variety of MFPs and membrane transporters. In the absence of TolC, therefore, E. coli becomes sensitive to many different molecules.

Although the number of TolC molecules, about 1500, is relatively low (133), interestingly, Krishnamoorthy et al. (134) recently demonstrated that apparently less than 10% of these are sufficient to maintain cells free of toxic levels of an antibiotic like vancomycin. This indicates that TolC is normally present in significant excess, contributing to its overall capacity to participate in the efflux of a wide range of molecules.

For secretion of HlyA, TolC interacts with the ABC transporter, HlyB, and the MFP, HlyD, in the inner membrane. In contrast, when TolC is associated with AcrAB to extrude small molecules, the ABC transporter is replaced by a resistance-nodulation-division (RND) protein (135). This is a secondary active transporter that utilizes the proton gradient across the inner membrane to energize transport. The structures and mechanism of action of such drug transporters is now understood in some detail and, as we will discuss below, this provides an instructive template for the organization of the T1SS translocator.

Thanabalu et al. (136) showed that HlyB and HlyD only formed a detectable complex with TolC upon interaction with the unfolded substrate, HlyA, and only remains associated with TolC during the actual secretion process. However, as we will also argue later, it remains a possibility that T1SS transport substrates simply stabilize normally transient interactions between TolC and its different partners.

Structure of TolC

A structure for TolC was initially obtained by two-dimensional crystallization in 1997 (137) at a resolution of 13 Å. In 2000, a seminal article by the Koronakis laboratory described the crystal structure of TolC at a resolution of 2.1 Å (138) and that the functional unit is a trimer embedded in the outer membrane of E. coli. TolC forms a β-barrel, constituted by a 12-stranded antiparallel β-sheet, by which each monomer contributes four β-strands to the final 12-stranded β-barrel. In contrast to the ligand-specific porins such as maltoporin, this barrel appears to be always open to the extracellular space because the structure lacks a potential plug, corresponding to the loop that forms a plug in the E. coli ferrichrome-iron transporter FhuA. Furthermore, and in contrast to classical porins, TolC extends very far (100 Å) into the periplasmic space (Fig. 6). This extended part of TolC adopts an entirely α-helical fold, composed of 12 α-helices, four from each monomer. At the periplasmic end of the structure, the helical bundle has a maximal diameter of only 3.9 Å (139). This indicates that the structure is the closed state, since this is obviously too small to allow transit of even α-helices or small solutes (see Fig. 6).

Figure 6.

Crystal structures of TolC in (A) the closed (PDB entry 1EKP, [138]) and (B) the open state (PDB entry 2XMN [107]). Each monomer of the TolC trimer is colored differently. The length of the periplasmic helices is highlighted. The maximal opening of the closed and open states of TolC is also indicated below the figure.

Noting a series of salt bridges and hydrogen bonds present at the periplasmic entrance of TolC, Koronakis and colleagues tested the functional implications of this by designing an elegant set of experiments, using site-directed mutagenesis to disrupt these interactions (140, 141). Indeed, this generated a TolC variant with a maximum diameter of 22 Å. This variant, as also shown in Fig. 6, apparently represents the open state compatible with translocation of α-helices of type I proteins or small molecules in the tripartite drug efflux systems. Thus, as proposed by Koronakis et al. (33), the data fit beautifully into a model whereby, during translocation, a molecule like HlyA could induce an “iris-like” opening of the periplasmic “tunnel” of TolC, involving outward sliding of the helices. However, one has to add that this now generally accepted mechanism was challenged by the crystal structure of the TolC homologue VceC in V. cholerae. This is required for transport of a range of xenobiotics and small molecules (142). This structure revealed that the key residues apparently responsible for closure of the TolC entrance are not conserved in VceC, raising the interesting possibility that the “iris-like” mechanism proposed for TolC was not necessarily used in all other T1SS or small molecule efflux pumps.

The MFP HlyD

Little is known about the structure and functional roles of other type I ABC and MFPs and the following sections will therefore consider only HlyD and HlyB. HlyD from E. coli is assigned to the so-called family of bacterial “membrane fusion proteins” or MFPs.

Topology of HlyD

Wang et al. (46), using fusions of beta-lactamase to different sites in HlyD, identified regions rendering cells sensitive or resistant to the antibiotic, depending on the exposure of the inserted enzyme to the cytoplasm or periplasm, respectively. This and cellular fractionation analysis demonstrated that HlyD has a single transmembrane domain (approximately 60 residues) for insertion into the inner membrane. A subsequent detailed analysis using alkaline phosphatase fusions confirmed this topology (143).

Mutational analysis of HlyD, essential for secretion

Genetic analysis of HlyD function has been surprisingly limited; nevertheless, deletion and mutagenesis analysis has indicated several possible functional roles for HlyD, including specific association with HlyB; recognition of HlyA; providing a tightly sealed transenvelope channel; ensuring the correct folding state of HlyA; and specific interactions with TolC.

As shown in Fig. 7, the N-terminal extension of HlyD contains several obvious features, a predicted 25-residue amphiphilic helix, a box of five charged amino acids close to the center, and three proline residues toward the terminus. In addition, three positively charged amino acids in the most distal region apparently adjacent to the transmembrane domain (TMD) are likely required for stable association with the membrane bilayer (144).

Figure 7.

Structural and functional features of EmrA and its homologue HlyD. (Top) The rectangles represent the distinct conserved regions of EmrA from E. coli (involved in multidrug transport) and HlyD from E. coli, indicating the location of the cytosolic domains, the transmembrane helices, the coiled-coil, the lipoyl and β-barrel domains. Scaling is based on the number of amino acids that contribute to the individual parts. The transmembrane helices are shown in green. The two helical regions of the coiled-coil domain are highlighted by orange boxes and labeled 1 and 2. The two parts of the lipoyl domain are indicated by blue boxes labeled, respectively, N and C for the N- and C-terminal parts of this domain, while the β-barrel domain is represented by brown boxes. The asterisk marks the position of the C terminus of EmrA from A. aeolicus, because this protein has no TMD. The zoom-in shows the sequences of the N-terminal cytoplasmic domains of HlyD, and other MFPs involved in type I secretion, PrtE (protease secretion), LipC (lipase), and LktD (another hemolysin). The putative amphipathic helix (letters in red) and the charged cluster (letters in blue) in HlyD, implicated in interaction with HlyA and consequent recruitment of TolC (145), are highlighted. The positions of the domains of EmrA are derived from the crystal structure obtained for EmrA from A. aeolicus. The corresponding positions in HlyD are estimated from the predicted secondary structures (including the coiled coil). (Bottom) Cartoon representation of the crystal structure of monomeric EmrA from A. aeolicus, which lacks a TMD and an N-terminal cytoplasmic extension (154), with a zoom into the compact form of the associated lipoyl subdomains. The recently determined crystal structure of parts of HlyD (152) is superimposed on the EmrA structure to emphasize the similarity of both proteins (cyan). Black residues in the α-helical hairpin represent the predicted position of the heptad repeats important for coiled-coil formation. In the lipoyl domain, blue residues are identical amino acids in more than 50% of the sequences analyzed in reference 155, while red residues are similar amino acids in more than 50% of the sequences analyzed. Note the closely adjacent N and C termini made possible by the flexible nature of the α-helical hairpin in EmrA. This would also allow movement to facilitate interaction with the cognate outer membrane protein, TolC.

Pimenta et al. first showed that deletion of the first 40 amino acids of HlyD blocks secretion completely, while the truncated HlyD remains normally expressed and stably localized to the membrane (45). Surprisingly, in competition experiments, this HlyD deletion mutant, when coproduced with WT HlyD (and HlyB) does not affect secretion of HlyA, but greatly reduces its hemolytic activity. This suggests the aberrant packing of WT and mutant HlyD molecules in a mixed oligomer, resulting in secretion of misfolded hemolysin (J. Young, A. Pimenta, and I.B. Holland, unpublished results). Mutants with such properties are a recurring finding as indicated in the next section. Indeed, as detailed in the next section, Balakrishnan et al. (145) confirmed that the cytoplasmic N-terminal extension of HlyD, apparently highly conserved in T1SS hemolysin-like toxins, plays a key role in the recognition of HlyA and then recruitment of TolC into a fully competent translocon.

Schulein et al. (146) observed some conservation of the C-terminal amino acids of HlyD (especially the last 35 residues) based on the limited range of HlyD-like sequences available at that time. When the 10 C-terminal residues of HlyD were deleted, secretion of HlyA was completely blocked. Similarly, site-directed mutagenesis of the terminal, Leu, Glu, or Arg residues of HlyD demonstrated that these were essential for secretion (Fig. 8). In competition experiments with expression of both WT HlyD and the truncated HlyD, secretion of HlyA was greatly reduced, indicating that the structure of the mixed oligomer was rendered defective in some way. Curiously, these authors also identified a region of 44 amino acids in HlyD (residues 127 to 170, the proximal part of the helical hairpin; see Fig. 7), showing a surprising 47% identity with residues 233 to 274 in a β-barrel region of TolC. Deletion of this region also completely blocked HlyA secretion, but unfortunately, analysis of the specific role of these residues for HlyD function has not subsequently been followed up.

Figure 8.

Mapping the position of mutations on a topology map of HlyD. The topology is based on membrane fractionation experiments and beta-lactamase insertions giving antibiotic resistance (46, 147). Each 50th residue is marked by a black box. Residues 150 to 246 and 251 to 327 define the approximate position of the coiled coils, residues 97 to 128 and 328 to 360 for the two half-lipoyl domains, while the major β-barrel occupies the C-terminal domain. The lipoyl domains, in particular, are likely to be involved in tightly packing the HlyD protomers, while the β-barrel by analogy with other MFPs could be involved in interaction with HlyB. Orange residues encompass the position (close to the middle of the helical hairpin) of the reportedly conserved RLT motif proposed to interact with TolC (214). At the N terminus, key regions required for binding HlyA and consequent triggering of TolC recruitment were defined (145) by the N-terminal deletions discussed in the text: deletion 1 removes the first 20 residues (the majority of the putative amphiphilic helix); deletion 2, the first 40 residues; deletion 3, the first 45 residues; and the internal deletion 4 removes the charged cluster, which is especially important for recruitment of TolC. Blue residues indicate secretion-defective mutations from different groups. * marks mutations blocking secretion, conditional on high calcium ion concentrations in the medium (147). These include mutations having possible effects on HlyD packing and map mainly in the lipoyl domain and the β-barrel.

In a different approach (147), random mutagenesis of hlyD in E. coli LE2001 was used to obtain several mutants located in the HlyD-periplasmic domain, whose expression, stability, and membrane association were unaffected. The mutants showed a range of novel properties and mapped to different domains of HlyD (Fig. 8), compatible with defects in different steps in the secretion process. Three mutants, T85I, K404E, and D411N, gave greatly reduced halos on blood plates and secreted little or no HlyA protein. In addition, in T85I, HlyA was not detectable in cell envelopes, suggesting a block early in initiation of secretion. In competition experiments, the D411N and K404E mutations (both mapping to the terminal β-domain) were dominant, apparently also because of formation of a structurally defective mixed HlyD oligomer. Moreover, in this mutant, the secreted HlyA protein had substantially reduced hemolytic activity (Young, Pimenta. and Holland, unpublished). Remarkably, in the same study (147), the four mutants, T85I (mapping just beyond the TMD), L165Q (proximal coiled-coil region), and V334I and V349I (both in the distal lipoyl region), showed both greatly reduced secretion levels and hemolytic activity of HlyA, but only in the presence of a high calcium ion concentration. Thus, with 10 mM Ca²⁺, no colony halos formed on blood plates, while, with 1 mM Ca²⁺, halos formed as seen with WT HlyD. From trypsin treatment of intact cells or isolated envelopes, it appeared that secretion from T85I is blocked by Ca²⁺ at an early step, with evident cytoplasmic accumulation of HlyA. In contrast, with L165Q, HlyA clearly accumulated in the envelope, apparently forming a stuck intermediate, while the two lipoyl mutants, V334I and especially V349I, showed accumulation of HlyA primarily on the cell surface. Moreover, with both lipoyl mutants, HlyA recovered from the medium was shown to be misfolded, since the HlyA protein was hypersensitive to trypsin and the low specific activity was restored to near-WT level after urea denaturation and renaturation. Interestingly, Vakharia et al. (148) described mutations affecting amino acids in the periplasmic domain of TolC, similarly resulting in reduced activity of the secreted HlyA.

Previous studies have shown that, in E. coli, the periplasmic calcium ion concentration increases substantially as the extracellular concentration is increased, bringing periplasmic levels up to at least micromolar, compared with nanomolar in the cytosol (149). We speculate, therefore, that in these Ca²⁺-sensitive HlyD mutants, to varying degrees, the packing of HlyD protomers or the sealing of interfaces of HlyD with HlyB or TolC are distorted sufficiently to allow penetration of the translocon by calcium ions. We suggest that these bind to the RTX during transit, triggering some premature structural changes, causing slowing of translocation and ultimately misfolding on the surface or, in the case of the mutant L165Q, actual arrest in the translocator.

The cytoplasmic domain of HlyD is required for HlyA-dependent TolC recruitment

MFPs that participate in ABC-dependent (T1SS) protein translocation invariably, to our knowledge, appear to have a single TMD anchor and a short N-terminal extension into the cytoplasm. However, the length and nature of the extension can vary markedly (Fig. 7). In the case of HlyD, the TMD is preceded by an approximate 59-residue N-terminal extension.

Balakrishnan et al. (145), confirming the importance of the C terminus for secretion described in the previous section by using an N-terminal 45-residue deletion of HlyD to block secretion, demonstrated that the mutant HlyD, in the presence of HlyA,B, was unable to “recruit” TolC into the transenvelope complex. Nevertheless, oligomerization of HlyD and its interaction with HlyB and HlyA, were apparently all retained. Notably, however, in the absence of HlyB, the truncated HlyD failed to bind HlyA, perhaps indicating that HlyB and D bind to overlapping sites in HlyA. Curiously, using a similar cross-linking approach, we found that oligomerization of HlyD, in contrast, was abolished when the N-terminal 40 residues of HlyD were deleted. We have no explanation for this discrepancy (Young, Pimenta, and Holland, unpublished).

As shown in Fig. 7, the first 45 amino acids of HlyD contain a predicted 25-residue amphipathic helix and a downstream cluster of five charged residues. Balakrishnan et al. (145), using other deletion mutants subject to cross-linking in vivo, showed that, while both the helical region and the charged cluster were required for secretion, only the charged cluster was necessary for recruiting TolC. However, no clear role was established for the helical region. Nevertheless, the authors were able to propose that the cytoplasmic N terminus of HlyD (apparently with the help of the ABC protein), when sensing the presence of HlyA, mediates transduction of a conformational signal to the periplasmic domain that allows recruitment of TolC. Importantly, however, from more recent comparative genomic analysis, it is clear, although surprising, that this mechanism for recruiting the OMP is not conserved. For secretion of other type I proteins such as the proteases and lipases, the MFP homologue has a very short, quite different N-terminal sequence compared with that of HlyD. Recruitment of the OMP in these cases must therefore require a different mechanism. For example, in some bacteria, the relevant OMPs and MFPs may inherently be able to form a complex without any special recruitment mechanism. Thus, for the analogous MacAB, TolC-macrolide transporter (153), and similarly for AcrAB, TolC (150), assembly of the entire complex apparently occurs even in the absence of an allocrite.

HlyD shares many structural features with MFP analogues

MFP is a misnomer since these proteins do not promote membrane fusion in the classical sense. Rather, they provide the physical and functional coupling of proteins in transenvelope complexes that straddle the inner and outer membranes; an alternative term also used is “adaptor proteins.” Symmons et al. (151) in an excellent review now introduce a new synonym, periplasmic adaptor proteins (PAPs).

The importance of HlyD in type I transport was, to some extent, initially underestimated, assumed simply to fulfil the role of connecting the inner membrane ABC protein with the outer membrane TolC exit to the medium. Until recently, no HlyD structural data were available, but now a partial structure of HlyD, 277 residues out of 478 and lacking both the cytoplasmic extension and the C terminus, has been published (152). However, many mutational studies and, in particular, crystal structures of functional analogues have brought important insights into what appears to be a quite complex role for HlyD. These homologues include the extensively studied multidrug transporter MFPs, AcrA, EmrA, MacA, and MexA, that form three protein component translocons with AcrB/TolC, EmrB/TolC, MacB/TolC, and MexB/MexX, respectively, in Gram-negative bacteria.

HlyD, as well as MacA, which interestingly is also coupled to an ABC transporter (153), and EmrA (154), linked with a major facilitator superfamily (MFS) secondary transporter to provide energy, all dock with TolC in the outer membrane. These proteins, like HlyD, have a single TMD peptide anchor close to the N terminus, embedded in the inner membrane. In the case of HlyD, this TMD is preceded by the cytoplasmic extension of about 60 amino acids, required for recruiting TolC, as discussed in the previous section. AcrA, in contrast, is anchored in the membrane from the periplasmic side by a lipid modification of the N terminus. This AcrA architecture presumably facilitates capture of drug molecules from the bilayer or periplasm, while the TMD and the N-terminal extension in HlyD permits direct contact with the HlyB TMD and with HlyA in the cytoplasm.

Although HlyD, as well as AcrA, MexA, and other family members share less than 25% identity, Johnson and Church (155) identified a number of well-conserved structural motifs in the large periplasmic domains of all these proteins (see also reference 151). As shown in Fig. 7, there is an extensive helical hairpin, strongly predicted to form coiled coils, toward the N terminus, for example, in EmrA. HlyD is also predicted to have a long helical region with two blocks of likely coiled coil and, as confirmed by the partial structure of HlyD, remarkably similar to that of EmrA (Fig. 7). Importantly, the hairpin domain in HlyD and EmrA is flanked at each end by a short half-“lipoyl” domain. This widely conserved lipoyl module is known to activate a number of enzymes, involving a switch in architecture from two well-separated halves to a fused globule of two lipoyl domains. However, in the functional form of MFP proteins, the lipoyl domains are always apparently associated in a compact structure. Other predicted structures conserved to varying degrees in MFP proteins are some β-sheet, N-terminal to the helical hairpin, an extended β-barrel domain toward the C terminus, followed by a disordered C-terminal domain (151, 155).

Importantly, MFPs appear to be very flexible molecules, with, in particular, flexible “joints” at the borders of the distinct domains. This presumably facilitates the acrobatics required to achieve stable contacts with proteins across two membranes and its oligomerization to form a tightly sealed palisade around its inner and outer membrane partners (156, 157). Moreover, the folding of the helical hairpin, formed by the coiled-coil region of AcrA, EmrA, and therefore very probably HlyD (Fig. 7), brings both the N- and C-terminal regions together, close to the inner membrane, available for interaction with the associated inner membrane energy transducer. Interestingly, exciting new structural models of a number of drug efflux pumps can provide important pointers to the positioning of HlyD in the translocon.

HlyB the ABC Transporter

Topology of HlyB

The topology of HlyB, an ABC half-size transporter, was deduced experimentally in 1991 (46) from the construction of beta-lactamase fusions throughout the length of the protein. The results, where an externally exposed beta-lactamase confers penicillin resistance on strains carrying a fusion, indicated six transmembrane domains (TMDs) in reasonable agreement with those predicted by algorithms. The fusion data also appeared to indicate two additional domains near the N terminus, but these were poorly predicted. In fact, sequence data indicated strong homology in this region with a cysteine protease, although with defective catalytic site (91) likely to prevent activity. Interestingly, Gentschev and Goebel (158) subsequently reported a topology analysis of HlyB, using beta-galactosidase and alkaline phosphatase fusions, showing differences in the precise position of the transmembrane helices (TMHs), but also identifying eight possible TMHs, including two in the N-terminal region. However, as described later, the N-terminal CLD domain (Fig. 2) was recently shown by Lecher et al. (54) to constitute an ancient but inactive C39 cysteine protease, nevertheless essential for secretion of HlyA. The topology of HlyB shown in Fig. 9, is a composite cartoon combining all the fusion data from Wang et al. (46) and Gentschev and Goebel (158). In addition, an arbitrarily fixed TMD length of 25 residues was adopted, with the exception of TM2 where the fusion data were otherwise more difficult to accommodate in the model. Importantly, in view of the recent demonstration that the CLD binds to the C terminus of HlyA, this region is now reassigned to the cytoplasm. The model, in particular, indicates a single large periplasmic loop P1 (a possible interaction site for HlyD), and two very small loops, P2 and P3, with two relatively large cytoplasmic loops, C1 and C2. In addition, the model predicts two regions of approximately 25 and 36 amino acids, with perhaps important functional roles, separating the TMDs from the CLD and nucleotide binding (NBD) domains.

Figure 9.

Mutations localized to the HlyB topology model. This is a composite based on beta-lactamase, beta-galactosidase, and alkaline phosphatase insertions, as described in the text. Each 50th residue in the sequence is marked by the black box. The crystal structures of the CLD and NBD were derived from purified fragments, residue 1 to 130 and residue 467 to the terminus, respectively. The conserved motifs Walker A, C-loop, Walker B, and the histidine of the H-loop, all in the NBD, are highlighted in blue, red, green, and brown, respectively. Letters indicate mutations identified in different genetic screens. Green letters highlight those mutations that affected the level of secretion, including three temperature sensitive mutations marked *, one of which (marked **) also grew poorly at 42°C. Letters in blue boxes are mutations in HlyB suppressing one or other of two deletion mutations in the HlyA secretion signal. Residues additionally marked by a # affected the oligomerization of HlyD, while triangles are secretion-defective insertions. Note the “hotspot” for mutations in the relatively well-conserved region, predicted to be periplasmic domain P3. In the case of ambiguities, for the periplasm the residue substituted is to the right. See text for other details.

Mutational analysis of HlyB

Surprisingly, few mutagenesis studies targeting the membrane spanning half of HlyB have been reported. Unfortunately, therefore, this very important approach has been seriously neglected. However, several mutants obtained for the NBD domain in a variety of ABC transporters, including HlyB, have considerably enlightened the analysis of the catalytic action of the NBDs by confirming the important role of many of the highly conserved residues. However, since these mutations have already been studied in great detail in many different ABC proteins (159), they will not be considered further here.

As summarized in Fig. 9, a few HlyB mutations linked to the membrane domain give intriguing phenotypes but, so far, have been much less instructive in linking structure to function than comparable HlyD studies. Several mutants were isolated by random mutagenesis by Blight et al. (116); two of these (G10R, and P624L) showed novel temperature-sensitive secretion defects. Residue G10 was located close to the N terminus, indicating that this region that we now call the CLD might have a specific role in the secretion process. The mutation P624L is located in the P-loop of the NBD. This is a highly conserved residue, and its implication in HlyA secretion is intriguing since this amino acid is highly conserved in both pro- and eukaryote ABC transporters, forming part of the Pro-loop linking the two domains of the NBD (160), and just downstream of the “signature” motif (LSGG). Another temperature-sensitive secretion mutant, G408D, was located in the deduced periplasmic loop P3 between TMDs 5 and 6. The corresponding region, in some cases covering up to 12 to 15 residues, we found was quite well conserved in a wide range of bacterial ABC transporters, but also in the eukaryotic ABC proteins, Pgp, Pfmdr, and CFTR. Saturation mutagenesis of the region, residues 399 to 412 (see Fig. 9), yielded mutations, I401T and D404G, giving no hemolytic colonies and approximately 20% of WT levels of HlyA hemolytic activity when measured in liquid cultures. In addition, a double mutant S402P, D404K completely lacking secretion of HlyA was obtained by site-directed mutagenesis. Therefore, in this region of 8 residues, at least four amino acids are essential for HlyB function.

Figure 9 also illustrates the location of the suppressor mutations in HlyB (161, and next section) that restored, at least to low levels, the secretion of HlyA deleted for the proximal or the distal halves of the 50- to 60-residue HlyA secretion signal. These suppressor mutations are widely dispersed throughout HlyB, with the authors suggesting that they might define the binding site for HlyA. However, these mutations are not site specific, and thus it appears more likely that they represent possibly unrelated structural changes, somehow bypassing the need for the full signal.

Two additional secretion mutants (Fig. 9), E256K and S279L, were found in cross-linking experiments to be defective in oligomer formation of HlyD in vivo (Pimenta, Young, and Holland, unpublished). This suggests that these are possible regions involved in formation of the HlyB, D complex.

HlyB interacts with the HlyA secretion signal

Current evidence indicates that the C-terminal region of HlyA interacts with HlyB, not only via the N-terminal CLD, but also with the NBDs. It is not yet clear if these binding sites in HyA do or do not overlap and whether they are close to the HlyA binding site of the HlyD N-terminal extension discussed above.

Different laboratories, using various chimeric translocons formed from combinations of cognate and noncognate ABC, MFP, and OMP proteins, have shown that selectivity controlling secretion of type I protein HasA and a lipase resides in the cognate ABC protein (162). As far as we are aware, however, a similar analysis has not been done for the Hly system.

However, the Ling group (summarized in Zhang et al. [161]) isolated HlyB mutations suppressing the effect of either of two HlyA secretion signal mutants. One mutation (del 1) deleted the proximal half of the presumed 58-residue signal (including helix I and the charged cluster) leaving residual secretion at about 5% of WT. The other mutation deleted essentially the distal half of the signal sequence, residual secretion being reduced to less than 1%. However, again as noted above, effects of the distal deletion would include misfolding of secreted molecules and therefore an apparent secretion defect.

Putative suppressors, obtained following random mutagenesis of hlyB, were screened for increases in hemolytic colony halo sizes. The secretion levels of the finally purified suppressors were quantified by hemolytic assay in liquid cultures. In view of the reservations above regarding hemolytic colonies, not surprisingly perhaps, levels of secretion by the suppressors of the HlyA terminal deletion were poor, with most increasing only 2% to 5%. However, levels 10% to 20% of WT were obtained with suppressors of the proximal deleted signal. In total, 21 suppressors were obtained, with one mutant, A629V selected independently by both deletions.

The majority of the suppressors map around the membrane domains and flanking cytosolic part of HlyB, as predicted by the topology analysis presented in Fig. 9. The authors concluded that the suppressors, in general, defined specific residues directly forming the binding site for the HlyA C-terminal secretion signal. However, while it cannot be ruled out that some of these HlyB mutations, particularly those selected against the N-terminal deletion, might affect an HlyA binding site, deletions cannot be suppressed allele specifically. Alternatively, we suggest that the results may reflect more indirect mechanisms of suppression that, nevertheless, if exploited further, might provide useful insights into the mechanism of action of HlyB. Finally, Zhang et al. (161) remarkably also showed that the mutants restored some level of secretion to HlyA deleted for the entire terminal 58 amino acids. This is a puzzling if not a worrying result, being very difficult to explain.

Surface plasmon resonance studies of the isolated NBD of HlyB have also demonstrated directly an interaction with a C-terminal fragment of HlyA (163). This interaction (with an affinity of 4 μM) was strictly dependent on the presence of the HlyA secretion sequence, since its deletion completely abolished the interaction. Interestingly, the dissociation rate of the interaction was accelerated in the presence of nucleotides, ATP or ADP, suggesting that the interaction is fine-tuned in the biological context. Interestingly, since the C terminus of HlyA is now known to be clearly translocated first, binding of the signal region to HlyB must be readily and rapidly reversible.

With a different T1SS, Delepelaire (164) also demonstrated that the ABC-ATPase activity of a partially purified PrtD from E. chrysanthemi was regulated in vitro by the cognate signal sequence. Thus, the natural C-terminal secretion sequence inhibited ATPase activity almost completely, although with perhaps a surprisingly low K_i of approximately 0.2 μM, indicating tight binding. These results for both HlyB and PrtD point to a common interaction between the secretion sequence of the substrate and the ABC transporter. This, plus the additional evidence for binding of the CLD upstream of the signal, within the distal RTX region, contributes to an emerging molecular picture of the sequence of events leading to initiation of HlyA secretion.

Coassociation of HlyB and HlyD

A number of studies looking at the ability of chimeric translocons to secrete different allocrites have indicated that specific interactions between the MFP and ABC proteins, and between MFP and OMPs, are required for specificity, i.e., secretion of a particular polypeptide. However, there is as no evidence yet of interactions between the OMP and the ABC protein. In the analogous tripartite drug transporters, it is still not completely clear when, if, and precisely how the analogous inner and outer membrane components interact (see reference 156).

HlyB and D form a constitutive complex: stability of HlyD depends on HlyB and TolC

A number of genetic studies have indicated that the ABC and MFP proteins of the Hly, Prt (protease), and Cva (colicin V) translocons interact. More directly, evidence for interactions has been obtained by using copurification with an affinity tag or in cross-linking experiments (136, 165, 166). Other support for an interaction between the ABC and MFP proteins comes from measuring their stabilities in cells. For the Hly T1SS, both HlyD (45) and HlyB (Young, Pimenta, and Holland, unpublished) are synthesized constitutively throughout the growth phase with both the laboratory strain and the pathogenic strain, LE2001. Notably, this expression is not coupled to the synthesis of HlyA, whose production (at least in laboratory cultures) is usually restricted to a relatively brief window during late-exponential phase (35, 45, 103). Moreover, in the presence of TolC, but in the absence of HlyB, HlyD becomes less stable, with a half-life of approximately 100 min at 37°C, compared with 5 h when both HlyB and TolC are present. However, in the absence of TolC, the presence of HlyB actually renders D highly unstable, with a half-life of 36 min (45). All these results are consistent with conformational changes perhaps required in all three proteins to achieve assembly of a functional complex. However, this has not been demonstrated directly, although there is strong evidence for this in the analogous bacterial drug transport complexes (see reference 156).

Preformed transporter complexes in other T1SS

Interestingly, the CvaB (ABC) and the CvaA (MFP) proteins of the colicin V translocator were also found to depend on each other for stability, while both are very unstable in the absence of TolC (165). In addition, interaction of TolC with CvaA appeared to induce structural changes in the latter, enhancing its sensitivity to intracellular proteases when TolC was absent. Notably in these studies, although the ABC and MFP were present, the transport substrate was absent, thus suggesting that all three proteins of the translocon can form a complex, independently of the allocrite. Surprisingly, a major gap in our knowledge is the absence of any reports, as far as we are aware, on the structure of these ABC MFP complexes, the stoichiometry, or the detailed nature of the protein-protein interfaces. However, useful information is forthcoming from different approaches to identify regions of the analogous MFPs, AcrA, MacA, EmrA, and CusB involved in drug efflux with the cognate transporter. These have indicated that the C-terminal membrane proximal domain, β-barrel, and lipoyl domains can in some cases be involved in interactions with the cognate energy transducer (see the recent review [156]). Of particular relevance, perhaps, are studies with the Mac-efflux pump in E. coli, where MacA (MFP) forms a complex with its partner MacB, an ABC transporter, also involving possible interactions between the respective TMDs (167). Moreover, with this Mac-macrolide efflux system, the periplasmic domain of MacA (the MFP) not only stabilizes the ATP-bound form of the associated MacB, ABC, but also stimulates its ATPase activity (153, 167). This suggests a greater degree of coordinated activity between the MFP and the energizing component than encountered so far.

Isolation of an assembled translocon for HasA

Letoffe et al. (166) first showed that, in cells producing the transport substrate, an entire translocon containing the ABC, MFP, and OMP could be isolated from cells as a single complex present in the detergent-solubilized membrane protein fraction. This was obtained by copurification through specific affinity binding to the tagged allocrite, but without cross-linking to stabilize protein interactions. Instead, a strategy was used to maintain protein interactions that involved frozen transport complexes in which the allocrite is “engaged” but secretion is blocked. Two systems were studied for the secretion of protease C from Erwinia and HasA from S. marcescens, but we shall only consider HasA. For such studies the hemophore HasA is a good choice, since, when folded, this can easily be detected by its specific affinity for heme. HasA as the tagged allocrite, was coexpressed with the heterologous Prt-transport proteins from E. chrysanthemi. This T1SS secretes several proteases but does not secrete HasA. However, HasA nevertheless blocked the translocon in some way. Consequently, all three Prt-transport proteins, ABC, MFP, and OMP, could be specifically purified together with the heme-affinity-purified HasA. In addition, all three transport proteins were still copurified when a translocon was used that does allow secretion of HasA, i.e., a chimera with the cognate ABC protein (HasD) coexpressed with the MFP and OMP from the Prt system. This indicates that the active transport complex somewhat surprisingly appeared very stable, even during detergent extraction.

Interestingly, but also surprisingly, when fully folded HasA (purified from culture supernatants) was added in vitro to proteins solubilized from cells expressing the Prt transport proteins, HasA still successfully interacted and copurified with the ABC protein. However, there was no interaction with the MFP and TolC proteins. When purified HasA was added to solubilized membrane proteins from cells previously primed to assemble the Prt translocon in vivo, by the presence of a cognate transport substrate (protease B), all three transport proteins were again affinity purified with HasA. This, albeit indirect evidence showed that functional assembly of the complete Prt complex required the in vivo presence of the cognate allocrite to form a sufficiently tight association to survive the extraction procedure. Hwang et al. (165) concluded that the MFP, OMP complex in these experiments, unlike HlyB with HlyD, did not form a strong association in the absence of transport substrate.

When only a single transport protein (ABC, MFP, or OMP) was expressed in cells together with the allocrite, Letoffe et al. (166) also confirmed that the ABC protein (that is PrtD) could be detected in association with HasA. The authors concluded that, for this particular T1SS translocon, assembly is initiated by the ABC-allocrite association (with no detectable direct docking of allocrite and the MFP). This is followed by an ABC and MFP interaction that triggers recruitment of the OMP to complete assembly of the complex. A subsequent study from the same group (114) in fact showed that a specific region of the signal sequence of HasA, via an interaction with the ABC transporter, promoted recruitment of TolC.

These studies were an important advance, and the results overall provided coherent conclusions. Nevertheless, the experimental setup had some limitations, particularly the assumption that all the possible protein interactions would resist detergent solubilization. The study also showed that folded HasA was able to bind the ABC protein in vitro with HlyA, while the CLD interacts with an unfolded HlyA, independent of the secretion sequence, the isolated NBD interacted with folded HlyA1. This was strictly dependent on the secretion signal but it is not excluded that this region was unfolded.

Isolation of the Hly translocon

In a subsequent and somewhat simpler system, Thanabalu et al. (136) analyzed possible interactions involving HlyB or HlyD using these proteins carrying an N-terminal histidine affinity tag. Importantly, putative complexes in this case were “frozen” by cross-linking in vivo with a reversible cross-linker. An HlyB, HlyD complex was detected even in the absence of HlyA and TolC. However, all three transporter proteins could be copurified in a complex when coexpressed with HlyA, clearly showing that the transport substrate was required to recruit the outer membrane component TolC. Importantly, tagged HlyD and, in particular, HlyB, when produced separately, could still be copurified together with HlyA, indicating that this interaction can occur before the complete translocon is established. Moreover, the authors did not apparently rule out the idea that both HlyB and HlyD might bind HlyA in a coordinated manner. However, the results indicated that HlyD but not HlyB could be cross-linked to TolC confirming that HlyD formed the “bridge” to the OMP. Using an HlyB mutant defective in ATPase activity to block secretion, these authors also found that ATP hydrolysis was not required for assembly of the whole complex, oligomerization of HlyD (detected as cross-linked trimers), the interaction of HlyB and HlyD, or their binding to HlyA. This therefore is also a “frozen complex.” It is frustrating, however, as in virtually all other similar experiments in the literature, that no attempts were made to determine at which stage secretion was blocked in the frozen complex, is the allocrite simply bound to the translocon or is it able subsequently to enter the “channel” to form a stalled intermediate.

Notably, in these studies in contrast to the analysis of HasA secretion discussed above, Thanabalu et al. identified preformed HlyBD complexes and, in particular, an important early role for the MFP in recognition of HlyA (136). This appeared to represent a fundamental difference with assembly of the Prt complex. Consequently, it is important to consider whether interaction of the allocrite with PrtE (MFP) was missed by Letoffe et al. (166) because of insufficient stability in the absence of cross-linking. On the contrary, it now seems more likely that the role of the MFP in the two systems is indeed fundamentally different, since an allocrite binding site in PrtE, analogous to that in HlyD, simply does not exist (Fig. 7). However, the Thanabalu et al. experiments gave no indication of which regions of HlyD (or B) and HlyA interact, and this still remains unknown. However, as described above, a subsequent analysis by Balakrishnan et al. (145) identified the HlyD binding site for HlyA and how this plays a key role by coupling binding of HlyA to assembly of the translocon. Finally, Thanabalu et al. sketched out a speculative model of the secretion process (136). This suggested that HlyA binds to HlyB and D, sequentially or simultaneously, to the same or different sites in HlyA. This triggers conformational changes in HlyD and recruitment of TolC, concomitant with opening of the transport channel, composed of HlyD and TolC multimers, but not the ABC protein itself. A ratcheted transit of HlyA through the envelope was then envisaged, involving the proton motive force.

Envisaging a possible structure for the Hly translocon

High-resolution structures for the N termini and C termini of HlyB, the CLD, and NBD, respectively, have been obtained, but no structure yet for the crucially important membrane domain, likely to form the initial pore leading to the HlyD channel. In addition, a partial structure for HlyD has recently been described (152), while only the RTX β-roll structure is known for the hemolysin itself. In contrast, several recent structural studies, combined with model building suggest how the known crystal structures of, for example, AcrA, AcrB, and TolC are organized in the complete pump. These studies have involved much debate, not yet concluded, concerning the precise contribution of each subunit to the putative transenvelope transport pathway. This includes whether the inner membrane protein (AcrB or MexB) contacts the OMP, and, especially contentious, how does the long (approximately 140 Å) strikingly funnel-shaped MFP at its narrowest point make contact with the outer membrane protein. For the construction of pseudo-atomic structural models of such drug pumps, the crystal structures of three partners were superimposed on low-resolution cryo-electron microscopy structures. These have confirmed the formation of a contiguous structure from cytoplasm to the exterior with AcrA and probably AcrB forming the walls of the periplasmic channel (see figure 6 of Du et al. [168] for this beautiful structure). Notably, the model of Du et al. (169) seems to confirm that AcrB and TolC do not interact. The Du et al. model also indicates that the termini of the long TolC helices protrude into the upper reaches of the AcrA funnel sufficient to give quite a small wraparound of TolC by AcrA. A similar study by Kim et al. (170) described essentially the same overall organization with the difference that the respective tips of AcrA and TolC intermesh with no real overlapping.

Borrowing from these models gives a rather good idea of the overall organization of the distal half or more of the HlyA translocon. Thus, we suggest that the TolC trimer and an HlyD hexamer interact to form a tightly sealed upper chamber with its central transport channel. Similarly, the drug pump models can be used to predict some of the interfaces that might form the lower chamber. In particular, in line with the structural similarities with EmrA, HlyD should form tightly sealed HlyD protomers, involving lipoyl:lipoyl interactions. Interactions between HlyD and HlyB, by analogy with the drug transporters, would probably involve periplasmic domains of the ABC protein, with the C-terminal β-barrel domain of HlyD protomers enwrapping the HlyB TMDs to form the lower part of the translocon channel. Finally, TolC itself is probably a relatively nonspecific passageway, not only for small molecules, but evidently also for polypeptides since many heterologous passenger proteins fused to the HlyA secretion signal have been successfully transported through TolC. However, beyond that, the drug pump models are not helpful. AcrB, for example, in contrast to HlyB, has an extremely large periplasmic domain, functions in a completely different way as a proton antiporter, and allows drug molecules access to the translocon from the membrane or periplasm rather than from the cytoplasm. Clearly, only the structure of an entire HlyB interacting with HlyD can solve this problem. Finally, while small drug molecules accumulating in the MFP-OMP chambers could be expected ultimately to diffuse outward to the exterior, this cannot explain the extrusion of huge polypeptides that presumably require propulsive force of some kind, a problem that remains for the future.

STRUCTURE FUNCTION OF ABC TRANSPORTERS

A major goal of research in this T1SS field is to understand the function of a unique member of this very large and ubiquitous ABC transporter family that is involved in polypeptide transport. These ancient transporters are quite remarkable in terms of the enormous range in size and complexity of the transport substrates, while relying on the same basic energy-transducing unit. This implies an inherent plasticity to facilitate coupling of the transport and energizing domains as these have evolved to encompass more and more allocrites. Transporting gigantic protein polymers would seem to be the ultimate in this evolutionary process. To understand this fully, not only detailed knowledge of the translocation pathway at the atomic level, but also the accompanying mechanism of catalysis of ATP and its coupling to allocrite movement will be required. Before coming to the specific properties of the HlyB transporter, it is appropriate, therefore, to review here the available information concerning the structure-function of the wider ABC family of transporters.

The Structure of the ABC ATPase

The first crystal structure of a purified ABC NBD domain was reported in 1998 by Hung et al. (171). This structure of HisP, the ATPase component of the histidine importer complex from Salmonella entericia serovar Typhimurium, revealed a monomer with ATP bound between the conserved Walker A and B motifs. In addition to the conserved RecA or F1 ATPase fold, two ABC protein-specific subdomains were identified. These are, respectively, a β-sheet domain that harbors an aromatic residue interacting with the adenine moiety of bound nucleotides, and an entirely helical subdomain that contains the conserved C-loop motif, the hallmark of ABC transporters. No other interactions between the adenine ring and the NBD were detected in the HisP structure and all subsequently determined structures. This explains why ABC transporters do not possess true nucleotide specificity, and ATPase activity can also be energized by different nucleotides such as GTP, CTP, or UTP.

Several biochemical analyses of HisP, other isolated NBDs, or full-length transporters have revealed cooperativity in ATP hydrolysis. This by definition indicates the presence of more than one ATP binding site during a single catalytic cycle. Therefore, the initial HisP monomeric structure was unable to explain the ATPase activity and to solve the puzzle, various dimeric arrangements of the NBDs were subsequently suggested. The correct architecture of the NBD dimer was finally proposed by Jones and George (172), based on simulations. The elegantly simplistic solution they arrived at was subsequently verified by the crystal structure of MJ0796, an isolated NBD from Methanocaldococcus janaschii (173). As shown in Fig. 10B, in the HlyB dimer, an ATP molecule is sandwiched between the Walker A of one monomer and the C-loop (the hallmark of ABC transporters, consensus sequence LSGGQ/R) of the opposing monomer, the so-called head-to-tail arrangement of NBDs. Thus, ATP acts as molecular glue, and the dimer is stable in the presence of ATP, but not ADP, since the C-loop of the opposing NBD interacts only with the γ-phosphate moiety of ATP. This fundamental arrangement has now been observed in all structures of isolated NBDs in the presence of ATP, with more than 40 structures deposited in the protein data bank. These include the NBD of HlyB (174, 175), but also, more recently, the structures of several full-length ABC transporters determined at resolutions ranging from 2.5 to 4.5 Å (see reviews [176, 177]).

Figure 10.

(A) ATP/Mg²⁺ bound H662A structure for the head-to-tail dimer of the HlyB-NBD (174). The RecA core of the NBD is shown in green and pale green, respectively, while the α-helical subdomain in yellow/pale yellow and the β-subdomain in gray. The conserved motifs Walker A, Walker B, and C-loop are highlighted in blue, magenta, and red, respectively. Conserved residues interacting with the nucleotide (arrowed) sandwiched between the two monomers are highlighted in ball-and-stick representation and labeled. The bound cofactor Mg²⁺ is shown as a green sphere. (B) Zoom into the ATP binding site. Please note that S606 (from the conserved C-loop) resides in the opposite monomer.

The Mechanism of ATP Hydrolysis by ABC Proteins Is Still Controversial

Results of experiments designed to uncover the mechanism of ATP hydrolysis, including those for HlyB, interestingly point to subtle differences in the mode of action of individual ABC transporters. Mutational studies have identified two amino acid residues crucial for ATP hydrolysis: one is the glutamate adjacent to the classic Walker B motif present in all P-loop NTPases (178), and the other is the histidine of the H-loop, another universally conserved sequence motif of ABC transporters (179). Replacement of this glutamate by glutamine-abolished ATPase activity in vitro in, for example, MalK, the NBD of the maltose importer (180); BmrA, a bacterial drug transporter (181); or MJ0796, the NBD of an ABC transporter with an unknown transport substrate (182). This mutation was actually employed to crystallize the ATP-bound, dimer state of the MJ0796 NBD (173). However, in the isolated NBD of HlyB, the same substitution E to Q, resulted in a residual ATPase activity of approximately 10% (183), while the isolated, similarly mutated NBD of the yeast mitochondrial ABC transporter Mdl1 displayed a very low but measurable level of ATPase activity (184). However, an approximately 20% residual activity was identified for the E to Q mutant form of GlcV, an NBD from a thermophilic ABC transporter (185). In contrast, mutation of the conserved histidine of HlyB resulted in a complete loss of ATPase activity. Similar results were obtained for the maltose and histidine importers (186, 187), while, in the case of the yeast ABC transporter Pdr5, ATPase activity was completely unaffected when the histidine was substituted. Intriguingly, however, this mutated transporter displayed a changed spectrum of transported substrates (188), surprisingly emphasizing the importance of the histidine residue in determining allocrite specificity. Together these results might suggest different mechanisms of ATP hydrolysis in certain ABC transporters, and extrapolations from one system to the other should be done with caution.

As discussed above, two particularly important amino acids for catalytic activity have been identified in different NBDs, the glutamate of the Walker B motif, involved in binding the magnesium ion, and the histidine of the conserved H-loop. In MJ0796, as indicated above, mutation of the glutamate to glutamine abolished ATPase activity completely and the term general “catalytic base” was coined (182). However,, as also indicated above, substitution of the histidine in HlyB by alanine abolished ATPase activity completely, while residual activity was observed in the case of the E/Q mutant (174, 183). Moreover, in a detailed biochemical analysis of ATP hydrolysis in vitro, several lines of evidence obtained were more consistent with a mechanism of substrate-assisted catalysis, emphasizing the role of the histidine, rather than general base catalysis giving the key role to glutamate. Therefore, the term “linchpin” (174) was also coined, and a critical role for a catalytic dyad composed of the histidine and the glutamate was postulated, in which the side chain of the glutamate interacts with the imidazole side chain of the histidine that, in turn, stabilizes the attacking water.

Subsequently, Oldham et al. (189) described structures of the maltose importer in the ground and transition states. For the latter, transition state analogues, i.e., vanadate and metallofluorides, were used. In these structures, the attacking water molecule, which has to be in line with the bond to be broken, was unambiguously identified as the glutamate, while the histidine was not in direct contact with the attacking water. However, the conclusion of Senior (190) that ABC transporters therefore operate by using general base catalysis does not take into account the limitations of crystal structures. The term “general base catalysis” refers to the kinetics of a reaction and operates if, and only if, proton abstraction is the rate-limiting step of the reaction. This certainly cannot be deduced from a static crystal structure. Rather, for example, isotope experiments should be used to demonstrate general base catalysis, where the reaction is performed in D₂O instead of H₂O or in a mixture of both. Since the mass of D₂O is larger than that of H₂O, more energy is required to abstract a deuterium compared with a proton. If proton abstraction is the rate-limiting step, the overall reaction velocity will be slower in D₂O. Such experiments have been performed so far only for the isolated NBD of HlyB (174) and the isolated NBD of Mdl1 (191). In both cases, no evidence for “general base catalysis” was observed. Thus, the precise catalytic mechanism for ATP hydrolysis is still not finally resolved and will require further investigation. Moreover, the possibility that different ABC-ATPases employ different mechanisms to fuel allocrite translocation through the hydrolysis of ATP is not excluded.

Mechanism of Translocation of Small Molecules by ABC Transporters: Current Model

The first crystal structure of a full-length ABC transporter in 2002 (192) was for the vitamin B₁₂ importer from E. coli (BtuCD). Now, there are more than 15 structures of ABC import and export systems deposited in the protein data bank at various stages of their transport and catalytic cycles. However, the most complete system, in terms of structural information for the entire transport cycle, is that for maltose import in E. coli (180, 189, 193). Moreover, this is also supplemented by a wealth of functional data. From all these studies (recently reviewed by Jones and George [176]), an atomic resolution picture is emerging of how, at least for small molecules, the chemical energy stored in ATP is coupled to the movement of an transport substrate across a biological membrane. Without going into great detail, the “two-side-access” model proposed in 1966 by Oleg Jardetszky (194) to explain the mode of action of membrane pumps appears to be valid for many ABC transporters. In the resting state of the transporter, the allocrite binding site is accessible from the exterior (importers) or from the cytosol (exporters). Binding of ATP leads to dimerization of the NBDs, and the effect of the resulting conformational changes is transmitted to the TMDs, via the so-called coupling helices. These are present at the terminus of particular long helices extending directly from the TMDs into the cytoplasm. Thus, each TMD of an ABC exporter is in constant contact with both NBDs, while one TMD contacts only one NBD in ABC import systems. The coupling helices first defined in structures of a putative bacterial drug transporter, Sav1866 (195), have been confirmed by structures of other ABC transporters (196–202). The coupling helices are roughly oriented parallel to the membrane plane and form crucial contact points between TMDs and NBDs, communicating the status of the membrane domain and the energy source to drive transport.

In current models, largely constructed to explain the translocation of small molecules such as anticancer drugs or nutrients, the consequence of NBD dimerization switches the accessibility of a binding site for a given molecule, from the exterior to the cytosol (importers) or from the cytosol to the exterior (exporters), resulting finally in release of the molecule. Subsequent ATP hydrolysis resets the system, and the resting state is restored. One has to keep in mind that, in addition to the two important alternating conformational states (inward-facing and outward-facing), an intervening occluded state must be adopted during the transport cycle, in which both sides of the transporter must be sealed. Otherwise, the transporter would adopt a conformation during its transport cycle, where both sides are accessible, a scenario producing an immediate, lethal dissipation of the proton motive force across the membrane.

Small Peptides Are Secreted by an “Alternating Access” Pathway but Large Polypeptides Require a Different Mechanism

Recently, Choudhury et al. (197) reported the crystal structure of an E. coli ABC peptide transporter, McjD, in the presence of a nonhydrolyzable ATP analogue. The structure revealed a novel outward-occluded state. The structure includes a large cavity (approximately 5900 ų, 40 × 21 × 10 Å), facing the cytoplasm, apparently able to bind the microcin, MccJ25, a 21-amino-acid, uniquely lasso-shaped, antimicrobial peptide. The outward-occluded conformation and other highly conserved structural features strongly suggest that McjD uses the alternating access mechanism to couple ATP utilization to microcin transport across the inner membrane of E. coli. Final release of this microcin to the medium is apparently dependent on TolC (203), another example of this amazing general purpose duct. However, there appears to be no information available concerning the presumed requirement for an HlyD homologue in microcin release.

More recently, highly important and exciting new structural information was also obtained from the crystals of an ABC transporter (PCAT) apparently secreting a peptide from the Gram-positive C. thermocellum. Although the natural transport substrate for this transporter has not been characterized, one gene of the operon encodes a 90-amino-acid protein that likely represents a bacteriocin. The structure of PCAT was obtained both for the nucleotide-free and a nonhydrolyzable ATP-bound form at 3.6 Å and 5.5 Å resolution, respectively (204). Based on the two conformations, open to the cytosol in the nucleotide-free state and closed on both sides of the membrane in the ATP-bound state, a large cavity facing the cytoplasm was visible in the absence of ATP. This cavity is sufficient to accommodate an entire small folded protein, such as a bacteriocin, after removal of the leader sequence. The PCAT transporter contains a C39 peptidase domain very similar in structure to the HlyB CLD described below (see also Fig. 2). This cleaves after the double-glycine motif of the N-terminal leader prior to translocation across the cell membrane. Interestingly, the C39 domain was only visible in the electron density of the nucleotide-free state of the transporter, while invisible in the ATP-bound state. This indicated a large degree of flexibility for the C39 domain in the latter conformation and the authors suggested that the two C39 subunits effectively move in a coordinated way from the periphery of the homodimer to deliver the peptide directly to the translocation site in the large cavity. The authors proposed, therefore, that the classical alternating access model was able to explain translocation of this peptide in a single step across the membrane, and, in this Gram-positive organism with no outer membrane, an additional adaptor protein is not required.

However, in marked contrast to microcins and bacteriocins other T1SS transport substrates range from 20 kDa to huge, close to 900-kDa polypeptides, and the available evidence suggests that all of them are transported in the unfolded state (69). Thus, as also concluded by Lin et al. (204), one has to question whether the “simple” alternating access mechanism applies to most T1SS ABC transporters, since the size of the unfolded substrate is evidently too large to be transported in one step. Moreover, if type I secretion involves ratchet-like translocation through a narrow pore as seen with the classical Sec system, it is difficult, if not impossible, to envision an occluded state. If not alternating access, HlyB could perhaps adopt a structure more related to the Sec translocon and simply deliver HlyA across the inner membrane into the HlyD channel. Alternatively, similar to the action of the ABC protein SUR that regulates the opening of the Kir_ATP potassium channel, HlyB might control the opening of an HlyD, TolC channel for extrusion of HlyA to the exterior, without directly contributing to the channel for HlyA. Finally, the most recent structural analysis of an ABC transporter, PglK (205), revealed that the lipid-linked oligosaccharide from Campylobacter jejuni, in order to inverse its orientation in the bilayer, is “flipped” along a novel pathway formed by PglK. The process requires an outward-facing conformation of PglK, but in this case there is no inward-facing conformation involved. This further emphasizes the plasticity of the ABC protein, and we can anticipate a fascinating solution to the puzzle of polypeptide transport by the T1SS from future structural data.

STRUCTURE FUNCTION OF THE PROTOTYPE TYPE I ABC TRANSPORTER HLYB

Structural Insights into the Catalytic Cycle of the HlyB NBD

The isolated NBD of HlyB has been investigated in great detail in terms of biochemistry and structure (160, 163, 174, 175, 206). Crystal structures of all states of the catalytic cycle have been determined for the wild-type or mutant NBDs. These structures, combined with determination of biochemical parameters, revealed much about the mechanism of ATP hydrolysis (see below). In addition, structural analysis was essential to identify residues of the NBD responsible for cooperativity in ATP hydrolysis and, for example, to define a role for the D-loop, a highly conserved motif of ABC transporters but without apparent function hitherto (175).

As in many other ABC-ATPases, the isolated NBDs of HlyB display cooperativity with a Hill coefficient of approximately 1.4 (183). In other words, the two ATP binding sites are not equal; that is, one of the many events leading to hydrolysis of bound ATP, binding, dimerization of the NBDs, ADP and/or phosphate release and dimer dissociation, is differently timed in the two sites. Otherwise, classic Michaelis-Menten kinetics would be observed. In fact, cooperativity requires asymmetry between the two ATP binding sites, and the structures of the ATP-bound states of ABC dimers did not at first reveal any obvious asymmetry. However, after analysis of the HlyB crystal structure and a subsequent sequence analysis of more than 10,000 NBD sequences, two residues apparently of particular significance with respect to asymmetry were identified for site-directed mutagenesis (175). These residues (R611 and D551 in the HlyB-NBD) are located one residue downstream of the C-loop and Q-loop, respectively. In the ATP/Mg²⁺-bound form, captured in the crystal structure of the HlyB NBD dimer shown in Fig. 10, these two residues are able to form a salt bridge but only within one monomer. This is because, in the other monomer, the distance between the two residues is too great. Importantly, in the absence of the bridge, a tunnel opens, reaching from the position of the γ-phosphate to the exterior, while the tunnel is closed in the other monomer by the salt bridge. These important findings represent a first glimpse of asymmetry in an ABC-NBD dimer.

Zaitseva et al. excitingly also showed that the change of only one of these bridge residues to alanine (R/D to R/A or A/D) drastically reduced ATPase activity, but, more importantly, abolished cooperativity. Changing the R-D interaction to K-E resulted in reduced ATPase activity, but restored cooperativity (175). These results indicate that the R-D interaction is at the heart of cooperativity because it regulates phosphate release from one of the two ATP binding sites. In the case of the alanine mutations, the tunnel would be open at both sites and the two phosphates released simultaneously, rather than sequentially. The importance of the R611-D551 interaction was subsequently confirmed for the antigen (peptide) transporter (TAP) (207).

The role of the D-loop, a highly conserved motif very useful for the identification of ABC proteins, has been elusive. In the crystal structure of the NBD of HlyB, the aspartate of the D-loop interacts with a serine residue of the Walker A motif of the opposing NBD (Fig. 10B). This is another example of an interdomain interaction between the two NBDs involving highly conserved motifs. In this case indicating an important role for the D-loop in sensing the presence of ATP in the opposing ATP binding site. Recent elegant studies by the Tampé laboratory have also emphasized the importance of the D-loop in the transport of immunopeptides by the TAP transporter (208). Similar conclusions were obtained from the analysis of other ABC transporters such as the vitamin B₁₂ importer (209).

The Extinct Peptidase Domain CLD in HlyB Interacts with the HlyA C Terminus: NMR Structural Analysis

Early studies noted that HlyB, compared with many other bacterial ABC transporters, has an extended N-terminal domain, but its role has been a mystery. Curiously, this N-terminal domain, covering approximately 130 amino acid residues, was identified from its sequence as a member of the cysteine-dependent C39 peptidase family. These particular peptidases are only found so far in active form in bacterial ABC transporters involved in the secretion of small peptides, including bacteriocins. In these molecules, the corresponding protease cleavage sites are found in the N-terminal secretion signal, C-terminal to a double-glycine motif. Obviously, HlyA does not contain an N-terminal targeting signal and does not undergo any proteolytic processing. In fact, the catalytically essential cysteine of the C39 peptidase domain in HlyB is replaced by a tyrosine, resulting in loss of protease activity. Thus, the N-terminal domain of HlyB was termed a “C39-like peptidase domain” (CLD).

In a major advance in 2012, Lecher et al. (54) revealed an important function for the CLD, since its deletion completely abolished secretion of HlyA. In further detailed studies, the purified CLD was shown to interact with the unfolded but not with folded HlyA or the truncate HlyA1. However, the 160-residue HlyA2, C-terminal fragment that lacks the C-terminal 58 residues encompassing the secretion signal (see Fig. 4), still binds to the CLD. This clearly showed that the binding site was located upstream of the secretion signal close to 3 RTX repeats (characterized, coincidently or otherwise by double-GG motifs), somewhere between residues 805 and 965.

To explore the structural details of the CLD and possible residues involved in binding to HlyA, the solution structure shown in Fig. 2 was determined by NMR. First, this revealed no obvious structural differences between the CLD and an authentic C39 peptidase domain (ComA-PEP [210]). However, subtle but important differences in the respective active sites were detected. While the aspartate residue of the catalytic triad of the C39 peptidase domains was found in an identical conformation in the CLD, the histidine residue was flipped by nearly 180° out of its canonical conformation. This novel position was stabilized by π-π interactions with a tryptophan residue. Interestingly, pairwise sequence comparison of many CLD domains in other ABC transporters revealed that in CLDs associated with T1SS, the combination of the absence of cysteine and the flipped histidine positioning was always present, while, in the authentic peptidase domains, the expected histidine and cysteine residues were also found and located at the same position, compatible with catalytic activity. This analysis might allow a robust and easy identification of new RTX-transporting systems in the future, based on the sequence of the CLD.

Finally, exciting results were obtained when an NMR analysis was used to determine the localization of residues in the CLD specifically engaged in binding to the HlyA C terminus. First, it is important to note that the published binding site for the normal C39 peptidase substrate, the GG cleavage motif, does not coincide with the site of HlyA binding in the CLD. However, intriguingly, the CLD residues binding to HlyA were found to be on the opposite side of the molecule to that shown to bind the classical GG motif. However, the significance of this finding may have to await determination of the molecular details of the CLD binding site within the RTX domain of HlyA. Notwithstanding this, the current data strongly suggest that the CLD participates in a very early step of the secretion process. Thus, Lecher et al. (54) proposed that the role of the CLD is to act as a receptor to position the unfolded HlyA in the close vicinity of the T1SS. Such a role could provide an additional initial tethering of HlyA (together with HlyD and the NBDs of HlyB) in coordinating its entry into the transport channel. Indeed, significant support for this idea comes from the recent structural analysis of a bacteriocin transporter PCAT discussed above.

An eGFP-HlyA–Fusion Protein Forms a Stuck Intermediate Extending across the Envelope with the C Terminus Exposed to the Exterior

Clear proof that HlyA is indeed extruded progressively through the envelope from the cytoplasm to the exterior was demonstrated very recently by Lenders et al. (118), who engineered the C-terminal HlyA1 fragment of HlyA (termed HlyAc in that publication), or full-length HlyA in a fusion to the C terminus of the fast-folding eGFP protein. The latter passenger rapidly folds in the cytoplasm and stalls further translocation. Employment of antibodies to HlyA1 or HlyA and measurement of fluorescence from eGFP showed conclusively that the N and the C termini of the fusion were in the cytoplasm and external medium, respectively. Importantly, the results in this paper also settled any doubts that the C terminus is secreted first, and, consequently, that the RTX region is immediately available for calcium ion-dependent folding of HlyA as the unfolded molecule emerges on the cell surface.

A DEVELOPING MODEL FOR HLYA TRANSLOCATION

We have tried to develop a model that portrays in a simple way what in reality is an intricate series of sequential intermolecular interactions between the C terminus of HlyA and the ABC and MFP proteins, and the consequent interaction with the OMP TolC. These interactions are likely to involve reciprocal short- and long-range conformational changes in the transport proteins that result in their assembly as a contiguous transenvelope structure. Equally, apparently multiple interactions between the HlyA C terminus and HlyB then prepare for the coordinated insertion of the HlyA C terminus into the transport channel. Progressive extrusion of the largely unstructured HlyA then proceeds across the periplasm, through the peptidoglycan mesh, through TolC, and finally autocatalytic folding of the emerging protein on the cell surface. We divide this complex process into hypothetical steps simply for convenience.

HlyB dimers and probably HlyD hexamers form a stable complex, independently of HlyA. We propose that the TMDs of HlyD encircle the membrane domains of the HlyB dimer, and, in this ground state, we assume that ATP is already bound to HlyB, but the NBDs are not tightly closed. In step 1, when the synthesis of HlyA is complete or nearly complete, the secretion signal, and one or more GG residues in the preceding RTX, bind to the NBDs and CLD of HlyB, respectively. Sequentially or simultaneously, the cytoplasmic extension of HlyD also has to bind to HlyA (location unknown). Multiple events are then somehow coordinated the tethering of both HlyD and HlyB to HlyA, the promotion of long-range “reorganization” of the periplasmic domains of the HlyD hexamer to recruit TolC into the fully assembled translocon. At this stage, the TolC exit may be open to the exterior, as has been suggested for the assembled pseudoatomic AcrAB/TolC complex, but other parts in the pathway may remain closed.

We suggest, in step 2, the C-terminal region of HlyA is delivered by the CLDs to a specific cytoplasmic part of the HlyB domain; binding of the HlyA signal to HlyD, CLD, and NBDs is reversed. These coordinated events, we speculate, constitute the trigger that creates the “translocation” site, a narrow pore, in the membrane domain of HlyB. We further speculate that this allows HlyA direct access to the contiguous channel formed first by HlyD and then TolC. In the purely hypothetical step 3, with the energetics still completely unclear, we could envisage that translocation occurs with sequential insertion of successive segments of HlyA into the transenvelope channel, thus resembling the threading of polypeptides through the Sec translocon. Notably, among the many transport processers dependent on ABC proteins, T1SS is seemingly unique in requiring continued consumption of energy during a protracted segmental transit of the polypeptide. In step 4, folding of HlyA in the calcium ion-rich exterior should commence as soon as the C-terminal RTX domain emerges from TolC. Whether, or to what extent, extracellular folding per se contributes energetically to the threading mechanism remains unclear. Similarly, still unknown is the timing of additional key steps such as induced dimerization of the ATP-loaded NBDs; hydrolysis of ATP and the disassembly of the translocon remain to be elucidated.

CONCLUSIONS AND PERSPECTIVES, FOCUSING PRIMARILY ON THE HLY SYSTEM

Novel Features

HlyA secretion depends upon a novel C-terminal secretion signal recognized by both MFP and ABC components of the translocon. At the heart of the translocon, the ABC transporter itself has novel characteristics, distinguishing it from the majority of this super family. HlyA secretion also has other features not shared by some other type I secretion systems. Moreover, from very recent findings, including our own revisiting of databases for unpublished material, it became clear that several aspects of the mechanism of type I secretion manifests itself in many forms, possibly related to allocrite type or the nature of the producing species.

Much Still to Learn about the Utilization of ATP

Despite the availability of several crystal structures for NBDs and entire ABC transporters, the specific step promoted by binding of the allocrite, binding of ATP, or precisely how and why, and at what stage, ATP hydrolysis is finally triggered, still remain relatively unclear for most ABC transporters, including HlyB. Moreover, a major fundamental step, central to our understanding of the action of ABC transporters, often sidestepped (even in recent reviews), is whether binding of ATP is spontaneous or triggered by the binding of the transport substrate. The literature is confusing, with some models presented with ATP already bound, while others show binding of the transport substrate triggering ATP binding. Encouragingly, recent studies of the effect of allocrites on the ATPase cycle with purified proteins, for example, the maltose transporter or the peptide transporter by TAP (involved in adaptive immunity) incorporated into proteoliposomes or nanodiscs, are specifically addressing these important issues (208, 211). In both cases, binding of ATP and the allocrites were clearly shown to be quite independent events. In addition, a recent beautiful structural analysis of different forms of a lipid flippase, the ABC transporter PglK, Perez et al. (205) proposed that the ground state of the transporter is the ATP-bound form. In fact, with the cellular ATP concentration at 3.5 mM, severalfold higher than the ATP affinity constant of ABC transporters, ABC proteins should always be present with ATP bound. Thus, normally, we expect that the ATP-loaded NBDs are somehow constrained and do not dimerize; rather, they are poised to close on the appearance of the transport substrate.

HlyB Function, Fascinating So Far but Surely Much More to Come

HlyB, interestingly, binds to two sites in HlyA. The NBDs bind to the secretion signal, while the N-terminal CLD (inactive C39 protease domain) binds (or tethers) HlyA close to the GG-rich RTX, just upstream of the signal sequence. Notably, however, the corresponding binding site in CLD, as determined by NMR, intriguingly is distinct from the binding site for GG repeats in the ancestral protease. The precise role of the CLD in HlyA secretion remains unclear, while a CLD is not required for secretion of small to “middle sized” proteins such as the proteases and lipases. Interestingly, the recent PCAT structure of a peptide transporter discussed above prompts the idea that a CLD picks up the C-terminal end of a type I protein, which is then guided to the translocation site. However, the HlyA secretion signal binds to the NBDs of HlyB, suggesting that the ABC protein may be implicated in coordinating an early initiation event with regulation of ATPase activity.

High-resolution structures of HlyB and studies of the precise stage, allocrite recognition, translocon assembly, or something else that blocks secretion when the CLD is absent, should illuminate the specific role of the HlyB CLD.

Different Type I Protein Families Appear to Have a “Personalized” Secretion Code

The hypothesis, based on available evidence, is that for HlyA (and related hemolysins), the secretion code is composed of a dispersed pattern of specific individual residues centered on the apparently conserved cluster EISK (100, 101) rather than on structural motifs. Such a signal code could provide a “lock and key” mechanism for docking with either or both HlyB and HlyD. However, a limited alignment analysis of the C termini of other type I proteins suggests that different subgroups have a distinctive code. The signal sequence also appears to have an additional role, with the extreme few residues, again apparently different sets for different subgroups, somehow affecting posttranslocational folding or stability. However, the nature of this phenomenon remains a mystery.

Translocon Specificity

Translocon selectivity in T1SS, i.e., translocon specificity for a given polypeptide, appears to reside (in the different subfamilies) in the secretion code and its docking target, the ABC protein. However, in distinction from the CLD and NBDs, if there is a specific binding site for HlyA in the membrane domain of HlyB, it is not yet identified. Curiously, for HlyA, specificity may also be determined by binding to HlyD, but it is not clear how many other type I proteins share this property.

Type I Proteins: Calcium Ion-Dependent Autonomous Folding on the Cell Surface

Many type I proteins, including HlyA, are secreted into potentially hostile environments where chaperones are absent. For HlyA, evolution has neatly solved the chaperone problem by providing RTX repeats (or their analogues, in some cases) to promote autonomous folding in the environment replete with the catalyst Ca²⁺. Moreover, several authors have presented the persuasive idea that calcium ion/RTX-dependent folding could also provide some or most of the required energy to “pull” large type I RTX proteins through the translocon. Until recently, this proposition had not been tested experimentally. Now, Bumba et al. (212) with the adenylate cyclase have shown that Ca²⁺-dependent folding is not essential, but rather appears to accelerate CyaA translocation. It remains to be seen if this role in augmenting energy requirement for translocation is conserved. In addition, these findings will reawaken the debate on what are the fundamental energy requirements for transport.

ABC Transporters Secreting Large Polypeptides Utilize a Novel Mechanism

The exciting first structures of T1SS ABC transporters secreting small peptides have revealed an overall structure consistent with the generally agreed mechanism for the export of small molecules by a very wide range of ABC transporters, that is, the alternating access model (194). In contrast, it is clear that HlyB and other T1SS homologues cannot transport a large unfolded polypeptide to shift the whole molecule to the exterior in a single step, as required for the alternating access mechanism, and therefore must function in a fundamentally different way. Notably, the recent structure of the bacterial lipid-oligosaccharide flippase (205) suggests a novel mechanism for an ABC exporter different from alternating access. This involves the formation of a transport pathway or “pore” within the membrane domain of the ABC protein. This system offers a simpler alternative concept for HlyA translocation that requires HlyA-induced reorganization of the membrane domain in HlyB to create the transport pore that connects directly to the HlyD-TolC channel.

Envisioning the Structure of the Translocon and the Role of HlyD

Homologues of HlyD involved in drug transport, such as AcrA, appear to form a sealed channel crossing the periplasm, and it seems very probable that HlyD plays the same role in formation of the Hly translocon. Notably, also among the related drug transporters, MacA forms a complex with an ABC transporter and AcrA, MacA (E. coli), and EmrA (from Aquifex aeolicus) all dock with TolC. In addition, like HlyD, MacA and EmrA also have a single N-terminal TMD embedded in the inner membrane. These shared properties of HlyD mutants lead to the prediction that 6 HlyD protomers, together with HlyB and TolC, form a contiguous transenvelope channel.

The flexible nature of the periplasmic domain of such MFPs presumably is also is a key factor in sealing and then stabilizing the fragile translocation “tunnel” across the gel-like, volume-variable periplasm. Reflecting this, HlyD, in some way not yet clear, appears important for the smooth transit of HlyA to the surface and its subsequent folding. Finally, in view of the recent exciting pseudoatomic models of the tripartite E. coli multidrug drug efflux pumps (169), which include both TolC and structural homologues of HlyD, we are now able to build a picture of the possible structure of the Hly translocon.

Although crucial details are still lacking, the N terminus of HlyD clearly plays a crucial role in the early engagement of HlyA and the consequent recruitment of TolC into the translocon “on demand.” However, surprisingly, this function is not conserved in many other T1SS, for both large and small allocrites. Consequently, how such a regulated assembly of the translocon is managed in other type I systems remains unclear. Conceivably, transient associations of these translocon components may simply be stabilized by interaction with the transport substrate.

T1SS Subtypes Display a Remarkable Variety of Mechanistic Detail

A striking realization during the preparation of this review was the surprisingly large number of variations between different subgroups of type I proteins regarding important details of the secretion mechanism. One of the most dramatic but puzzling variations is the likelihood that different subgroups have their own secretion signal code, presumably requiring corresponding sequence or structural variations of the cytosolic part of the cognate ABC transporter. Another major puzzle is that, while the N-terminal cytoplasmic domain of HlyD has a fundamental role in the secretion process, this is not conserved in many other T1SS. Type I proteins include giants up to close to 9,000 amino acids, and growing evidence indicates a range of variations to accommodate secretion of some of these proteins: the accessory CLD domain, a variety of novel auxiliary translocon components and a possible contribution of extracellular folding to the energetics of transport for some proteins. Similarly, the emerging evidence that, while small peptides can be secreted via an alternating access mechanism facilitated by the ABC transporter, larger proteins are presumably extruded through some kind of “pore.” This would mean yet-to-be-discovered sequence and structural variation in the corresponding transport domain for the T1SS ABC transporters. Moreover, for the giant type I proteins, likely with conceivably extremely long transit times, modifications to meet the greater energy needs are inevitable.

There are also variations in the T1SS, although less clear-cut, in posttranslocational folding mechanisms. The extreme C-terminal motifs that may be involved do not appear to be conserved and, although in many cases this folding depends upon RTX motifs and extracellular Ca²⁺ ions, a recently discovered exception in S. enterica indicates a likely independent evolutionary event that has created an alternative calcium ion binding site.

While the recruitment of CLDs to the ABC protein appears to be an expedient linked to allocrite size, justification for the sophisticated structure-function of the HlyD N terminus is more difficult to comprehend. Interestingly, however, as first pointed out by Rod Welch (3), several characteristics of the sequence of the hly determinant, including the low GC content, indicated that this was a recent acquisition by E. coli from an unrelated species. This unfamiliar neighborhood might have selected for the secondary recruitment of a novel system for linking the MFP into a stable complex with TolC. In fact, frequent lateral transmission of pathogenicity factors followed by subsequent adaptation to local physiology could explain other forms of T1SS variations and, indeed, similar phenomena might reasonably be expected in other bacterial secretion systems. The prevalence of such a variety of forms of T1SS emphasizes that a mechanistic paradigm with one type I protein is no guarantee that this will be conserved widely.