The aberrant introduction of splice site junctions within TM segments would likely lead to serious disruption of the encoded membrane protein structure. Indeed, computational studies assessing the likelihood of an intronic division within a TM region have found that that the probability of a TM not being divided by an intron was slightly higher than the expected probability for a random 22-mer amino acid sequence (42). This difference was most pronounced in single-pass MPs, where TM regions were encoded on single exons 84.6% of the time, in comparison with the random sequence expectation of 58.5%. Thus, there seems to be evolutionary pressure for MPs, especially single pass MPs, to keep introns out of regions coding TMs within pre-mRNA.

Rh blood group antigens are important in blood-typing for blood transfusion and organ transplantation. These antigens are typically predicted to have 12 TM segments (48). The RhCcEe proteins represent an example of MPs whose topologies are altered due to AS (49). The characterization of Rh cDNA clones led to the conclusion that at least four different transcripts are produced from the RhCcEe gene by AS (Figure 3). Exon removal in one of the isoforms (RhVIII) results in the removal of residues 163–313, which form five TM segments. As a result, three of the helices are predicted to cross the membrane in reverse orientation relative to the original full-length transcript. Inversion of TM segment topology also occurs in isoform RhVI, which, as a result of exon removal, is missing residues 163–267, which form three TM segments. The reversal alters the residues that are surface-exposed and may alter antigenicity (see Figure 3A). Additionally, RhVI has another deletion (corresponding to residues 359–384) due to exon removal. This deletion causes a frameshift and results in a different C-terminal sequence with a premature stop codon. The Rh4 isoform is missing amino acid residues 314 to 358 as a result of exon removal; this exon deletion event induces a frameshift resulting in a different C-terminal sequence and a premature stop codon at the same location as in RhVI. Interestingly, the new C-terminal sequence generated by the frameshift in RhVI and Rh4 encodes the same 14 amino acids, indicating that this sequence is likely important, although its function is not yet known (see Figure 3).

Another example of altered TM topology is found in the V2 vasopressin receptor, a G-protein coupled receptor (GPCR). Its two isoforms, V2a and V2b, were found to have different topological states and different levels of stability (50) (Figure 3B). V2a is a stably expressed seven-TM segment (7-TM) protein found at the plasma membrane. Conversely, V2b differs in sequence from V2a in the C-terminal region after the sixth TM segment. This altered C-terminal tail results from the use of an alternate 3' splice site 76 base pairs downstream of the canonical 3' splice site, which results in a frameshift. The V2b protein isoform was experimentally shown to significantly populate two different topologies: a 7-TM topology similar to canonical GPCRs with the C-terminus located intracellularly, and a 6-TM topology with the C-terminus oriented to the extracellular matrix (Figure 3B). The V2b receptor also differs in that it is localized to the ER and Golgi apparatus rather than trafficking to the plasma membrane. Retention of the protein in the ER and Golgi could be due either to the splicing-based deletion of C-terminal forwards-trafficking motifs or may reflect sequestration of the truncated protein by MP folding quality control.

A third protein whose TM topology is affected by alternative splicing is the glycine receptor (51), which assembles into heteropentameric (α₂β₃) ion-conducting pores and acts to mediate neurotransmissions in the central nervous system. It was found that the β subunit of this receptor was alternatively spliced to remove exon 7 (βΔ7). The long isoform of the β subunit including exon 7 possesses a predicted four-TM topology while the βΔ7 variant possesses a predicted two-TM segment topology due to the removal of TM segment 1, the first intracellular loop and a portion of TM segment 2 (see Figure 3C). Interestingly, the extracellular domain is kept intact, the protein is stably expressed in the cerebral cortex and can hetero-oligomerize with the glycine receptor α-subunits or with the glycine receptor-anchoring protein gephyrin. Because the TM segment 2 (which is shortened in βΔ7, where it is instead predicted to form the tenth beta strand in the large extracellular domain) is part of the ion conduction pathway, the functional properties of glycine receptors containing βΔ7 may be different. Indeed, experiments showed that the α₁βΔ7 complexes did not have resistance to the channel blocking agent picrotoxin while α₁β complexes display this property (51), suggesting that the pore is formed only by the α-subunits.

Signal peptides are important for membrane targeting and determination of protein subcellular localization and are often altered as a result of AS. In a computational analysis of the mouse transcriptome, 40% of transcriptional units (sets of transcripts derived from the same gene) were shown to have signal peptide variation, with the majority of transcript variation being the consequence of AS (52). These alterations are important for the targeting soluble proteins to various subcellular compartments (lumen of endoplasmic reticulum vs. cytosol, for example). Signal peptides also play an important role for targeting and integration of proteins into membrane, where they help determine membrane protein topology. The same study investigated a high-confidence set of 782 transcriptional units, where the presence of alternative transcription initiation start and termination sites in transcripts were independently confirmed, with variations in membrane organization. While this set most likely excludes most examples of AS based on the selection criteria, it is important to note that the AS may accomplish a similar task in cells based on alternative initial exons. In the same set of transcripts, there were 41 instances of signal peptide removal that resulted in a switch from Type I (C-terminal cytosolic) TM topology to Type II (N-terminal) TM topology in single-pass TM proteins. It is likely that use of alternative initial exons, which make up 58% of the mechanisms of variation among the set of transcriptional units with signal peptide variation investigated in this study, could result in a similar total inversion of topology. It seems feasible that AS could also result in complete inversion of topology of multi-span membrane proteins. There are MPs of this class that are known to populate opposite topologies (53), although splicing has not so far been established as the driving mechanism.

Signal peptide removal by AS has been documented in the parathyroid hormone (PTH)/PTH-related peptide (PTHrP) receptor (54). The gene for this GPCR can be transcribed from two different transcription start sites. Use of a novel promoter in kidney results in the generation of three alternatively spliced transcripts (type I-type III), of which the signal peptide is replaced with a hydrophilic sequence in the type III splice variant. This splice variant, which was shown to retain reactivity, was expressed only minimally on the cell surface, with the majority being localized to intracellular compartments, likely as a result of replacement of the signal peptide. The type III receptor could be reactive to an intracellular form of PTHrP or may simply represent the outcome of a downregulatory mechanism.

In addition to altered TM topology, alternative splicing can eliminate TM domains from transcripts that would otherwise encode MPs. This results in soluble protein isoforms that may have vastly different roles in (or out) of the cell. A survey (55) of a portion of the human MP transcriptome revealed that nearly 40% of the 464 surveyed alternatively spliced single-pass MP genes have isoforms that completely lack a TM domain (Figure 4A). For example, the alternatively spliced soluble isoform of the syndecan-4 receptor (SCD-4) is very similar to the ectodomain of the canonical isoform. Syndecan receptors normally function as co-receptors by concentrating and presenting a variety of ligands (e.g. extracellular matrix proteins, growth factors, cytokines, and cell adhesion molecules) to other cell surface receptors, or they act to internalize ligand, sequestering them away from other receptors, thereby modulating signaling. The ectodomain of SCD-4 can be proteolytically cleaved, an event that is regulated by multiple signaling pathways, and this solubilized domain can still bind ligand, affecting signaling by competing with the membrane-bound SCD-4 counterpart (56). It is possible that the AS soluble isoform may play a similar ligand-sequestration role as that played by the ectodomain generated by proteolytic cleavage.

Another example of production of a soluble isoform from a MP transcript is cadherin-7, a cell-adhesion TM glycoprotein. In developing chickens, it was found that a soluble form of cadherin-7 was expressed (57) and that this isoform can interact with the corresponding domain in the TM variant of this same protein, resulting in inhibition of cell-cell adhesion by interfering with homomeric interactions of the TM form.

Ions cannot cross the hydrophobic lipid bilayer via diffusion. Cells therefore express numerous types of channels, both to regulate cytosolic ion content and, in the case of specialized cells such as neurons and muscle cells, for signaling purposes. Ion channel activity can be regulated by AS. For example, transient receptor potential (TRP) potassium channels are heavily regulated by AS (58). The functional impact can range from altered conductance to changes in selectivity or activation. More drastic changes include the generation of dominant negative isoforms that prevent native channel function through heteromultimerization with the canonical isoforms. Consider the cold/menthol sensitive TRPM8 channel. Bidaux et al (59) have described the activity of two newly discovered short isoforms of TRPM8 (dubbed short TRPM8α and short TRPM8β). These two short isoforms contain only the cytosolic N-terminal domain and are thus are not capable of forming functional channels, but bind to and regulate canonical long-form TRPM8 channel tetramers (Figure 4B). Specifically, these variants stabilize the closed form of the channel, reducing channel activity and sensitivity to cold.

The TRPM3 channel is alternatively spliced to produce different sequences within the presumed pore domain, generating different cation selectivities. For instance, TRPM3α1 and TRPM3α2 are both outwardly rectifying cation channel isoforms, meaning they allow higher current flux out of the cell than in, whereas other TRPM3 variants are inwardly rectifying. TRPM3α2 is highly permeable to divalent cations while TRPM3α1 favors monovalent conductance. Additionally, monovalent cations block currents through TRPM3α2 but not TRPM3α1. These TRPM3 variants clearly demonstrate a role for AS in regulating channel ion permeability, selectivity, and regulation (60).

Transporters facilitate the passage or flip-flop of solutes across the membrane, often in an energy-consuming manner against the TM concentration gradient (active transport). Transporter activity is typically tightly regulated, with AS being one mechanism of regulation.

An important family of regulated transporters are the P-type ATPases. This family includes the plasma membrane calcium pump, which is present in all eukaryotic cells and helps to maintain appropriate Ca²⁺ levels within the cytosol by moving calcium ions from the cytosol to the extracellular matrix. The calcium pump is alternatively spliced such that its calmodulin-binding regulatory domain is altered to reduce its net positive charge, resulting in a lower calmodulin affinity (61). Reduced affinity of the ATPase for calmodulin results in reduced transporter affinity for cytosolic Ca²⁺ ions.

Another example is the glutamate transporter family. After release of neurotransmitter into the synaptic cleft, transporters are responsible for transmitter reuptake by the cell and regulation of the volume of these vital molecules at the synapse. Levels of glutamate, the major excitatory neurotransmitter in the brain, are regulated by glutamate transporters, the most prominent of which is GLT1 (EEAT2 in humans). The three alternatively spliced isoforms of this protein (GLT1a, GLT1b, GLT1c) differ in their C-termini (62), with both GLT1b and GLT1c containing a PDZ-recognition motif. GLT1a lacks this sequence (Figure 4C). Inclusion of a PDZ-binding sequence likely has functional consequences for the different GLT1 isoforms, since they may then recruit PDZ domain-containing proteins that modulate transporter function, trafficking, or associated signaling. GLT1a and GLT1b have similar expression profiles, and are primarily expressed in the brain, while GLT1c is expressed in the retina.

G protein-coupled receptors (GPCRs) are an essential subset of MP, the target of almost 40% of the drugs on the market today (63) and represent an interesting example in terms of AS. Although many GPCR-encoding genes do not contain introns, some GPCR genes do (64–66). Besides the V2R and PTHrP receptors discussed earlier, classic examples include the D₂ dopamine receptor and the rat type I pituitary adenylate cyclase-activating peptide receptor (PAC₁ receptor). The D₂ receptor is present in one of two isoforms in the central nervous system (CNS): short (D_2S) or long (D_2L). The D_2S receptor lacks 29 residues within the third intracellular loop, likely resulting in different signaling properties due to the role of the intracellular loops in interaction with its cognate G-protein (64). The PAC₁ receptor is expressed in five different isoforms, all with variations in the third intracellular loop resulting in different signaling properties. The alternative forms of this loop differentially regulate activation of the downstream effectors adenylate cyclase and phospholipase C (64), again, most likely by altering interactions and specificity between the receptor and heterotrimeric G proteins.

In the case of the cannabinoid receptor hCB1, two splice variants have been identified (hCB1a and hCB1b) in addition to the canonical receptor. These variants have altered amino termini; hCB1a has an altered N-terminal sequence and hCB1b is missing the first 33 N-terminal amino acid residues. In both cases AS occurs through use of different splice site donor/acceptor sites. hCB1a and hCB1b display altered pharmacological properties in relation to the canonical hCB1 receptor (Figure 4E). Their affinity for the native ligand anandamide is dramatically reduced and their affinity for 2-arachidonoylglycerol is somewhat decreased. Additionally, it was found that instead of acting as an agonist, as it does in the hCB1, 2-arachidonoylglycerol functioned as an inverse agonist when interacting with hCB1a and hCB1b (67). Because the N-termini are not in the ligand-binding site, this example points to a function of AS in allosteric regulation of receptors.

TM enzymes are also affected by AS. Carbonic anhydrase XII (CA XII) is an enzyme whose alternative splicing is associated with cancer (68). There are two alternative transcripts of CA XII, the longer of which encodes 11 additional amino acid residues immediately preceding the predicted TM domain. Both transcripts are found in healthy tissue, but the longer isoform normally dominates. However, in astrocytomas expression of the shorter isoform is greatly increased and the longer isoform was rarely detected. Why the shorter isoform is more prevalent in cancer cells is not well understood; although, it is likely that the shorter isoform of CA XII may have difficulty in forming its normal quaternary structure. This is because a portion (GXXX) of the GXXXG motif, important for homodimerization of the enzyme, is missing due to alternative splicing (Figure 4D). Thus quaternary structure disruption likely contributes to catalytic dysfunction. For additional examples of enzymes affected by AS, see the Alzheimer’s disease section.

The proteolipid protein (PLP) is the most abundant protein of central nervous system (CNS) myelin and is thought to maintain structural integrity of myelin and support myelin compaction, although it may have additional roles in signaling (69). PLP is a tetraspan MP with an intracellular “principal loop” connecting TM helices II and III. DM-20 is a splice variant of PLP and lacks a large portion (34 amino acids) of the principal loop (70), including two palmitoylation sites (71) (Figure 4F). It is likely that this region allows for differential regulation of the functions of PLP and DM-20, altering their trafficking, structure, and/or signaling functions. It is interesting to note that AS-based elimination of two palmitoylation sites may also reduce the propensity of DM-20 to interact with cholesterol-rich membrane domains often referred to as “lipid rafts” or other specialized membrane domains in myelin membranes.

Differences in alternatively spliced isoforms can impact the ability of the protein product to interact with partners. As discussed above, the GLT1 transporter isoforms differ in their possession of a PDZ-binding motif, a motif likely used to bind protein partners containing PDZ-binding domains. Another example is the N-methyl-D-asparate (NMDA) receptor. Yotiao, an A-kinase anchoring protein, scaffolds to the NR1A splice variant of the NR1 subunit of the NMDA receptor but not to the NR1C splice variant (72). This interaction is modulated by either the presence or absence of a subdomain encoded by an alternatively spliced C1 exon cassette that is included in NR1A but not the NR1C splice isoform. Additional research suggests that Yotiao may mediate a ternary complex between itself, the NMDA receptor and cAMP-dependent protein kinase II. This interaction is postulated to provide a mechanism for cAMP to modulate NMDA-receptor signaling (73). Similarly, calmodulin interacts avidly with the receptor through binding of NR1A to the region encoded by the very same C1 exon cassette, thereby inhibiting NMDA receptor function (74, 75).

Aberrant AS of the insulin receptor (InsR) results in the insulin resistance phenotype observed in DM1 patients. The insulin receptor functions as a heterotetramer with two α- and two β-subunits. In DM1, there is a switch from the normal production in skeletal muscle of the IR-B splice variant of the α subunit, which includes a 12 residue C-terminus, to production of the IR-A variant lacking this terminus (88, 89). Omission of the C-terminus results in reduced responsiveness of the InsR to insulin and reduced kinase activity (and hence, signaling) for the IR-A isoform that is expressed in DM1 (90, 91).

One of the most common symptoms of DM-1 is myotonia, a delay in the relaxation of muscle after contraction. Hyperexcitability is likely caused by loss of function of chloride channels. It is thought that this is due to aberrant alternative splicing of the muscle-specific chloride channel (ClC-1), which results in decreased levels of the channel (92). Three aberrant splice variants of ClC-1 were found in the skeletal muscle of DM-1 patients. All three of these disease-generated isoforms contain premature stop codons introduced by the splicing changes and the transcripts are likely degraded by nonsense-mediated decay. This results in the loss of functional translated ClC-1 in skeletal muscle and thus a decrease in chloride conductance across the membrane, causing myotonia (93).

Perhaps the most debilitating symptom of DM-1 is muscle wasting and weakness. It has been shown in other disorders that this symptom can result from changes to calcium homeostasis in muscle cells that increase intracellular Ca²⁺ levels (94). Two proteins essential to maintaining calcium homeostasis are the ryanodine receptor (RyR1) and the sarcoplasmic endoplasmic reticulum Ca²⁺ ATPase (SERCA). The RyR1 is responsible for the release of Ca²⁺ in storage in the sarcoplasmic reticulum during muscle contraction and SERCA transports Ca²⁺ back across the ER membrane for storage. Research has shown that the transcripts for both of these proteins undergo AS in DM-1 patients (95).

RyR1 has 2 variants seen in DM1: the neonatal form of RyR1 lacks exon 70 (residues 3481–3485), which corresponds to five residues found in the receptor modulatory region, (96) and the adult form RyR1 includes exon 70. Adult DM-1 patients have been found to have elevated levels of the neonatal RyR1 isoform in their skeletal muscle. This form of RyR1 has lower channel activity than the adult isoform, which would result in increased depolarization-dependent Ca²⁺ release (95).

Two SERCA genes, SERCA1 and SERCA2 are also implicated in DM-1. SERCA1 has two splice variants: SERCA1a is the adult version of the protein and contains exon 22, which encodes 7 amino acids at the C-terminus. The neonatal form, SERCA1b lacks exon 22, and its C-terminus has an eight-residue, highly-charged C-terminal tail. In DM-1, a switch occurs, where the major isoform produced is the neonatal form (SERCA1b) as opposed to the adult form. It was also found that SERCA2 has a novel variant (SERCA2d) with lowered expression in DM-1. SERCA2d contains intron 19, which results in the addition of 27 amino acids, a frameshift, and a premature stop codon in exon 20. The aberrantly expressed SERCA isoforms both have an altered C-terminus, but how this impacts overall pump function is not known (95).

The voltage-dependent Ca_v1.1 calcium channel plays a central and important role in excitation/contraction coupling in skeletal muscle (97). This channel also undergoes aberrant splicing that is associated with DM-1. AS of Ca_v1.1 in DM patients results in omission of exon 29 (ΔE29), removing a portion of the extracellular loop near the voltage sensor. As a consequence, the ΔE29 isoform exhibits altered calcium channel gating, which increases Ca²⁺ influx. These defects result in Ca²⁺ overload, likely contributing to muscle weakness and wasting.

Three major splice isoforms of APP have been documented: APP695 (lacking exons 7 and 8; the predominant neuronal isoform), APP751 (lacking exon 8 and expressed abundantly in non-neuronal CNS tissue and peripheral tissue), and APP770 (containing both exons 7 and 8 and expressed predominantly in peripheral tissues and at only low levels in the CNS) (108, 109). Exon 7 encodes a Kunitz protease inhibitor (KPI) domain in the ectodomain of APP that is possibly involved in blocking the α-secretase-initiated non-amyloidogenic APP processing pathway, since inclusion of exon 7 promotes production of Aβ and a decreased production of the p3 peptide (110). However, it should be noted that this domain is a serine-protease inhibitor domain, so would be unlikely to block the metalloproteinase α-secretase directly. Exon 8 encodes a region of the protein between the KPI domain and a heparin-binding domain. All three APP variants are subject to proteolytic processing along the amyloidogenic and non-amyloidogenic pathways described above. There is much inconsistency in the literature regarding differential expression of these isoforms in non-AlzD versus AlzD brains. Because it is difficult to compare relative values obtained from relative quantification-PCR among different studies, definitive in vivo data showing that differential AS of APP predisposes a patient to AlzD is still lacking.

BACE1 is a single span MP with its catalytic subunit located in the ectoplasm. BACE1 is a prominent target in the as-of-yet unsuccessful search for a drug to lower amyloid-β production. A recent series of publications (111–114) has described the various isoforms of BACE1 produced through AS. The longest isoform, I-501 (named for the number of amino acids), is both the most active and most commonly produced. However, AS events can also generate a series of shorter isoforms, including I-476, I-455, I-432, and I-127. The I-127 mRNA transcript contains a premature stop codon and is degraded by NMD; in addition, the small amount of translated I-127 protein isoform is subject to degradation by the proteasome (113). The BACE1 I-476, I-457, I-455, and I-432 all exhibit dramatically reduced activity relative to I-501, such that promotion of AS to generate the shorter isoforms results in decreased Aβ secretion (114). Accordingly, promotion of AS of the BACE1 pre-mRNA may be a viable therapeutic strategy for reducing Aβ production. Recent experiments in the laboratory of Michael Wolfe have shown that I-501 production is activated through the binding of heterogeneous nuclear ribonucleoprotein H (hnRNP H) to a G-rich element in the third exon of BACE1 pre-mRNA (115). Because the 5' splice site responsible for I-501 production is thought to be weaker than the upstream 5' splice site (115) leading to the I-457 and I-453 isoforms, targeting the activation of I-501 production by hnRNP H is a reasonable therapeutic strategy in the search for a way to reduce BACE1 activity, in this case by reducing production of the most active isoform of the enzyme.

PS is the aspartyl protease in the heterotetrameric γ-secretase complex responsible for intramembrane cleavage of a large number of different single span MP substrates (116–118), including the 99-residue TM C-terminus (C99) of the APP that is released by BACE1 cleavage. Cleavage of C99 by γ-secretase releases the Aβ peptides (see Figure 6A). After assembly into the γ-secretase, cleavage of auto-inhibited presenilin within the intracellular loop located between TM segments 6 and 7 generates active γ-secretase. The now active presenilin is comprised of a heterodimer of its N-terminal fragment (NTF) and a loop-containing C-terminal fragment (CTF). The NTF and CTF each contribute an aspartic acid residue necessary for catalysis (102, 119–121). Prior to its cleavage, the loop that is clipped to generate active presenilin is thought to extend into the active site to block catalysis.

Inherited mutations of presenilin-1 or presenilin-2 (PS-1 and PS-2) are one of the known causes of familial (early onset) AlzD. Some of these mutations appear to impact pre-mRNA splicing. Additionally, aberrant splicing of the presenilin transcript has been shown to occur in the more common sporadic AlzD.

It has been shown that presenilin-1 (PS1) is affected by multiple splicing events, and that some heritable PS-1 mutations modulate splicing. Here, we will discuss three instances where splicing may play a role in disease progression.

First, in human tissues, two isoforms of PS1 are expressed that differ only by the inclusion or exclusion of four amino acids, VRSQ, in the N-terminal domain (122). The VRSQ motif, which is conserved among humans, rats, and mice, provides a putative phosphorylation consensus site for protein kinase C (PKC) at a downstream threonine (122). This phosphorylation event may play a role in regulating the trafficking of PS1. There is a relative decrease of the longer isoform in some cases of sporadic (123) and familial AlzD (124), suggesting a link between altered trafficking of PS1 and AlzD etiology.

Aberrant splicing of PS1 can also be induced by mutation. For instance, a mutation within intron 4 was identified in several patients with early-onset AlzD (125, 126), which results in production of three aberrant species of PS1. Two of these contain full or partial deletions of exon 4 and result in truncation of the protein product; the third encodes insertion of a single threonine residue between the canonical residues 113 and 114 (126). Only the threonine-insertion species was detected in brain homogenates, suggesting the AlzD phenotype most likely results from the insertion species. This is supported by observation that transfection of the Thr-insertion form of the enzyme into HEK-293 cells resulted in increased Aβ₄₂ secretion (126).

Another interesting aberrant splicing event occurs when a mutation destroys a splice acceptor site at exon 9 (127). This causes the deletion of exon 9, which encodes part of the autoinhibitory loop including the site of activating cleavage of presenilin within the auto-inhibitory loop (128). Because disruption of this loop is required to activate presenilin activity (128, 129) and the elimination of the autocleavage site results in constitutively active presenilin (116), it was originally thought that its elimination via AS is solely responsible for the associated AlzD phenotype. However, it was later demonstrated that the single point mutation (S290C) that also results from this splicing event may contribute to AlzD pathogenesis by promoting increased Aβ₄₂ production (Figure 6B) (130).

For presenilin-2 (PS2), an aberrant splicing event in sporadic AlzD, resulting in exon 5 skipping (PS2V) and protein truncation, has been extensively characterized (131–135). The removal of exon 5 causes a frameshift in exon 6 and introduces a premature stop codon. The resulting PS2V mRNA encodes only the amino terminus containing one predicted TM segment and an additional five C-terminal residues (131, 132). The PS2V protein is found in the hippocampus and cerebral cortex of patients with sporadic AlzD, accumulates in inclusion bodies that impair the unfolded protein response, and results in increased Aβ production through an undefined mechanism (132). It is possible that it may exert its effect through associations with other γ-secretase proteins or through disruption of the unfolded protein response, preventing disposal of Aβ.

Recently, a novel splice variant of PS2, termed PS2β, was described to be a γ-secretase inhibitor (136). Its expression is therefore predicted to be beneficial in terms of reducing the risk for AlzD. This splice variant encodes the entire NTF plus the autoinhibitory hydrophilic loop domain that typically resides between TM segments 6 and 7. However, the exons encoding the entire CTF are not included in this isoform, such that this splice variant is catalytically inactive. PS2β was found to reduce the interaction between APH-1 and nicastrin and to inhibit Aβ secretion (136). It seems then, that PS2β may inhibit γ-secretase activity by preventing assembly of the holoenzyme. Perhaps this isoform may represent an endogenous means of repressing γ-secretase activity (Figure 6B).

Another γ-secretase subunit, anterior pharynx defective-1 (APH-1) is a seven-TM spanning protein encoded by two homologous human genes, either APH-1a or APH-1b (137). APH-1 is known to be responsible for assembly of an initial subcomplex with nicastrin, and together these two proteins recruit the presenilin and PEN2 (138–142). The APH-1a transcript can be alternatively spliced to produce one of two isoforms, APH-1aS and APH-1aL, which differ only at their C-termini (137). These splice variants have been shown to reside in distinct, active γ-secretase complexes (143). Additionally, a novel (third) splice variant, APH-1bΔ4, has been identified that lacks the entire fourth exon, meaning it also lacks the entire fourth TM segment (144), resulting in predicted inversion of topology for the three C-terminal TM segments. This omitted TM segment also contains a GXXXG motif that is thought to be critical for assembly of the γ-secretase multisubunit complex (145) (Figure 6C). Endogenous expression of this variant form of the protein is very low, and the protein is destabilized (144). Whether alternative or aberrant splicing of APH-1 plays any role in promoting or preventing AlzD has yet to be determined.

Nicastrin is a highly glycosylated type I MP that participates, along with APH-1, in the assembly of the γ-secretase complex. Two independent genome surveys (146, 147) have found evidence for association of sporadic AlzD with the region of chromosome 1 on which the nicastrin gene resides.

In 2005, a novel skipped splice variant of nicastrin was reported to be expressed in rat tissues and a human neuroblastoma cell line (148). This variant, which encodes a truncated 62-residue protein due to exon 3 removal resulting in a premature stop-codon, was shown to be preferentially made in the central nervous system and is subject to degradation by NMD. It is possible that NMD plays a specific role in the healthy CNS to down-regulate expression of active γ-secretase, thereby reducing Aβ production.

Soon thereafter, a second novel nicastrin splice variant was identified, this time lacking the entirety of exon 16 (149). As a result, this splice variant is missing 71 residues preceding the N-terminal region of the TM domain, which was shown to be required for interaction with the γ-secretase complex (150).

The serotonin (5-hydroxytryptamine) receptor 5-HT₄, a GPCR, has been implicated in the regulation of APP processing and receptor agonists have been shown to increase the production of the soluble APP ectodomain released by α-secretase cleavage of full length APP to initiate the non-amyloidogenic pathway (151–153). Alternatively spliced isoforms of 5-HT₄ have intracellular C-terminal domains of different lengths: a short isoform, 5-HT_4(d), which contains only two amino acids after the C-terminal splice site, located after the region encoding TM segment 7 and preceding the intracellular C-terminal tail, and a larger isoform that contains 20 amino acids after the C-terminal splice site, 5-HT₄. These isoforms were compared with respect to the promotion of non-amyloidogenic APP processing (154). It was found that not only does activation of the short 5-HT_4(d) isoform by agonist promote generation of non-amyloidogenic soluble-APPα (sAPPα, the ectodomain released upon α-secretase cleavage), but also that, unlike the longer isoform, stimulation of the short isoform reduces Aβ production. It is possible then, that 5-HT₄ receptor isoforms may differentially regulate APP processing. Presumably, this may come from signal transduction differences related to the differing C-terminal tail lengths, possibly involving signaling through arrestins, which become activated when they bind to the multiply-phosphorylated C-termini of activated GPCRs. Such arrestin-based signaling would be eliminated in splice variants that removed all or key parts of the C-terminal domain of the 5-HT₄ receptor. Further insight into how 5-HT₄ receptors influence APP processing and their relationship with the secretases is needed to shed light on this data.