Wolfe MS, Selkoe DJ.
Biochemistry. Intramembrane proteases--mixing oil and water.
Science. 2002 Jun 21;296(5576):2156-7.
Please login to recommend the paper.
To make a comment you must login or register.
The Martoglio paper describes the identification human SPP. The experiments are solid. The question at hand is whether the sequence homology within the SPP and PS transmembrane domains that harbor the presumptive catalytic aspartate residues now "provide compelling evidence that presenilins are proteases" as stated by Wolfe and Selkoe. Martoglio's evidence that SPP are intramembrane proteases is supported by the experiment where the SPP gene is expressed in yeast along with the substrate. There appears to be no SPP homolog in yeast and hence, the authors suggest that the exogenous SPP is responsible for processing. This may be true, but an unambiguous demonstration would require complete reconstitution of activity in a lipid bilayer.
This is a critical issue for the following reason: while there may not be SPP homologues in yeast, SPP-like sub-family members are expressed (Fig 2). It is conceivable that ectopically expressed human SPP associates with proteins that would otherwise bind to the SPP sub-family members in a complex, and it is the complex that is required for activity. A similar proposal has been voiced for the PS-nicastrin-XX protein complex. It should be noted that the final purification step (Figure 1E) of the paper shows the presence of at least three other proteins that are at similar stoichiometry. What are the identities of these proteins?
Finally, I think it is important to say that in addition to SPP and its subfamily members, the second pore loop of mammalian tandem pore K+ channels (over two dozen known in worms and dozens in mammals) has a sequence that is essentially identical to that shown for SPP and PS. The consensus sequence is: T/I/V G L/F GD F/Y. Are there other similar sequence motifs in other unrelated polytopic membrane proteins that are (presumably) unrelated? Who knows?"—Sangram Sisodia, University of Chicago.
Reply by Sam Sisodia—Posted 2 July 2002
Mike and Todd seem quite content with the conclusion that presenilins are gamma secretases. However, I hasten to add that "the critical role of two completely conserved presenilin aspartates located precisely where one would expect catalytic residues" is a huge leap of faith. That is because there is neither structural information available about the "precise" locations of the aspartate residues, nor is the data compelling that aspartate 385 resides within a transmembrane domain. Several additional questions remain unanswered.
First, why is AβPP processed at multiple sites within the transmebrane domain, while Notch is processed at a single site? Mike’s and Dennis [Selkoe’s] argument is that there has not been an exhaustive analysis of other potential cleavage sites within the Notch TM domain. Fair enough, but I remind you all that mice with a knockin mutation at the P1 site of Notch to a large extent show the same phenotype as a Notch null mice.
Second, why are the FAD-linked L166P mutation, and artificial variants at 286 (Kulic et al., 2000) that promote enhanced production of Aβ42 while also generating robust levels of Aβ40, very inefficient (in the latter case, dead!) in Notch processing?
Third, how does one explain that of the full repertoire of FAD-linked PS1 mutations, which span the entire length of molecule, all enhance Aβ42 production? A host of these mutations occur in the "loop" between transmembrane domains and in the cytosolic domain immediately adjacent to TM6.
Regarding the "indispensibility" of asp257 variant in Aβ production, as claimed by Hui Zheng and colleagues (see related news item), I think the experiment is silent. Data from Christian Haass' lab and my own have shown unequivocally that this variant blocks Notch processing, not Aβ production. Indeed, the combined D257A/D385A PS1 variant has no effect on Aβ production, either (Kim et al., 2001).
A diaspartyl protease? Let's not forget that the aspartate variants are dominant-negative inhibitors of high molecular weight complex formation (St. George Hyslop and colleagues). Unfortunately, this still does not explain the differential processing of AβPP-CTF and Notch.
Finally, there must be some way to accomodate the finding that AβPP-CTFs AND full-length AβPP accumulate at the cell surface upon expression of PS1asp variants (Capell et al., 2000; Kim et al). One might argue that lack of cleavage relegates the substrates to accumulate at the cell surface. But why would this occur if there are endocytic signals still present on the tails of both full-length and AβPP-CTFs? Perhaps these molecules are being continually recycled to the surface? The problem is that significant levels of AβPP accumulated on the surface. These certainly are not the penultimate substrates of Aβ. Is there a general disruption in trafficking of selected proteins in late compartments, but independent of whether these molecules serve as "gamma secretase” substrates?
We are pleased that Bart de Strooper and Sangram Sisodia
commented on our Science paper in the Alzheimer Research Forum.
I would like to address two statements in their comments. Both authors say
that SPP was co-purified with other proteins, and they raise the possibility
that these proteins may be part of a complex. This, in turn, may imply to
the reader that SPP functions in a complex similar to presenilin in the
gamma secretase complex.
For purification of photo affinity-labeled SPP (see our paper), the
last step was performed under denaturing conditions in 50 percent formic
acid (described in the methods). This method is often used to separate very
hydrophobic membrane proteins. Labeled SPP and the other few proteins eluted
in the same peak on the reversed-phase column, but it is very unlikely that
they co-eluted because they form a stable complex.
Reply by Michael Wolfe
As Sam Sisodia and Bart De Strooper point out in their comments, it is true that the Weihofen Science paper does not definitively answer the question of whether presenilin is the catalytic component of γ-secretase. Nevertheless, the evidence does continue to converge asymptotically toward affirmation of this hypothesis.
In the Weihofen study, expression of human presenilin homolog 3 (PSH3) in S. cerevisiae was shown to result in SPP protease activity where there was none before. Other strains of yeast contain SPP genes, so it remains formally possible that hypothetical yeast PSH partners are expressed in S. cerevisiae and become activated in the presence of human PSH3. Sisodia emphasizes this possibility, but its likelihood seems quite remote.
First, the hypothetical yeast PSH partners would have to be expressed in S. cerevisiae even though a yeast PSH gene is absent in this strain. In no organism has a nicastrin gene been found without presenilin, so why should hypothesized PSH partners be present in S. cervisiae? Second, the human PSH3 would have to interact properly with the putative yeast partners, demanding remarkable three-dimensional conservation of the interacting surfaces from yeast to human. Third, a ketone peptidomimetic SPP inhibitor binds directly to PSH3, strongly suggesting that the active site resides in that protein. Such ketones can be hydrated in the active site of aspartyl proteases and thereby mimic a high-energy intermediate formed during catalysis. Fourth, the presence of several other proteins in the purified fraction along with the labeled PSH3 does not suggest that these other proteins might be members of a complex.
Such a complex would have to survive all of the purification steps, including a reversed phase HPLC run with 50 percent formic acid in acetonitrile and isopropanol. The other purification steps included Triton X-100, a detergent that irreversibly destroys γ-secretase activity and dissociates presenilin complexes. The purification protocol used by Weihofen et al. was optimized to enrich for the biotinylated protein target, not for an active protease complex. Keeping the presenilin complex together during purification is not straightforward, even when that is the goal. Therefore, the other bands in the last lane of Figure 1E are almost certainly irrelevant.
While the definitive proof that presenilin is the catalytic component of γ-secretase still awaits us, at this point it is nevertheless a pretty safe bet that it is true. Betting against it means believing that the aspartyl protease characteristics of γ-secretase, the critical role of two completely conserved presenilin aspartates located precisely where one would expect catalytic residues, the labeling of presenilin heterodimers by active site-directed inhibitors, and now the identification of a presenilin homolog with protease activity on its own is all somehow a remarkable collection of coincidences.
In a collaboration initiated by Chris Ponting, we have also identified the family of proteins identified by Bruno Martoglio and colleagues (Ponting et al. 2000) . In that paper we speculated that these presenilin homologs (PSH), as we called them, were likely to be proteases based on the evidence that PS were proteases. The study by Martoglio provides almost unequivocal proof that the protein they identified and named signal peptide peptidase (SPP, which we referred to as PSH3), is a protease capable of cleaving a signal peptide. This paper is simply an outstanding study that should convince many, if not all, in the field about the nature of this class of proteins. PS and PS-like proteins not only appear to be proteases but, as proposed by Wolfe and colleagues, aspartyl proteases . Future studies on other PSH should also reveal that they possess proteolytic activity, and reveal even more research opportunities for those interested in studying intramembranous proteolytic cleavage events carried out by multipass membrane proteases.
In my mind, this settles the debate over the role of PS in gamma-secretase cleavage. PS are almost certainly the catalytic components of a gamma-secretase complex. To make clear my reasoning, I will review the data on PS that support their catalytic role and also answer the criticisms put forth by others. Of course PS do not appear capable of catalyzing gamma-cleavages by themselves. Two other components necessary for this cleavage are Nicastrin and a membrane environment rich in cholesterol. Other as yet unidentified factors may also be required for cleavage.
First, all of the genetic evidence is consistent with PS being gamma-secretase. FAD-linked PS mutations alter gamma-cleavage and the combined PS 1 and 2 knockout completely abolishes gamma-cleavage. We believe that reports showing otherwise are technically flawed. (However, it is possible that other minor pathways for gamma-like cleavages exist, see concluding paragraph).
Second, biochemical evidence supports this concept. PS co-purify and are enriched as one purifies gamma-secretase activity. This occurs no matter how the purification is carried out. Purifications using inhibitor affinity columns, immunoprecipitations, or just differential centrifugation or other physical separations all reveal the same thing. PS and nicastrin co-purify with gamma-secretase activity. Moreover, mutations of either of the critical aspartates in PS, deletion of the first two transmembrane domains, or the naturally occurring mutation insR352 all result in dominant-negative PS molecules. Although possible, it is hard to see how all of these alterations could alter a co-factor function in the same way.
Certainly these studies are potentially consistent with PS being a co-factor and not catalytic, but the third line of evidence is not. This evidence is pharmacologic in nature. All gamma-secretase inhibitors that have been rigorously tested appear to bind PS. Most convincingly, protease inhibitors designed to inhibit aspartyl proteases bind directly to PS. In this regard, the studies by Weihofen et al. further demonstrate the importance of the aspartates in catalysis, as mutation of the aspartate abolishes inhibitor binding. Of course, not all inhibitors of gamma-secretase that bind PS look like typical protease inhibitors, but this is not particularly relevant to the debate. Given a novel class of proteases, novel inhibitors will be identified. (If they were willing, it would be interesting to actually poll companies with active or defunct gamma-secretase programs to ask them how many of their gamma-inhibitors they have found bind to presenilin. My sense is that anyone who has looked finds this to be the case).
So what are the remaining objections?
First: PS doesn't look like typical proteases.
Well, neither does rhomboid, an intramembranous-cleaving serine protease , or site-two protease, an intramembranous-cleaving metalloprotease that cleaves the SREBP . Should we expect these proteins to closely resemble known proteases? The remarkable fact is that the catalytic mechanism may be analogous to other classes of enzymes, and that catalytic residues do in fact appear to be conserved. Indeed, the demonstration that SPP is almost certainly a protease just makes us aware that we should be prepared to expect the unexpected.
Second: We can't reconstitute gamma-secretase activity.
This certainly is a valid criticism but represents a technical shortcoming, not a conceptual one. Gamma-secretase activity is associated with a high-molecular weight complex that appears to require Nicastrin as well [6-9], and perhaps other unidentified proteins. However membrane environment is also crucial for this cleavage (at least for AβPP cleavage) with membrane cholesterol content being a prime determinant of activity . Given the nature of this complex it may very well be difficult to reconstitute activity, especially if we haven't identified all the players. Indeed, the activity of the best-known intramembranous cleaving protease, site-two protease, has yet to be reconstituted. Moreover, no one to my knowledge has purified from scratch the best-known multi-subunit protease, the proteasome. Yet there is no debate about its proteolytic activity. In this regard, I guess that it will be easier to purify the complex to homogeneity than it will be to reconstitute activity.
Third: The spatial paradox, i.e. PS are not in the right place in the cell.
This criticism has been overplayed. First, evidence in multiple systems suggests that PS and substrate co-localize and co-fractionate. Indeed, one can co-purify PS with substrate especially when gamma-inhibitors are around . Moreover, studies in the brain show that substrate and PS co-localize quite well . Second, it is possible that a small amount of PS can turn over a lot of substrate. If this is so, it may be hard to show co-localization of activity with substrate if a lot of inactive PS is around, too. Third, it is also possible that PS epitopes are buried when they are in the active complex, preventing one from seeing PS in places where substrate might exist. Finally, it is entirely possible that substrate is being actively trafficked to a site in the cell containing active PS/gamma-secretase. Thus, the interaction, when unperturbed, would be transient and difficult to localize.
Fourth: Activity paradoxes.
There are studies that claim PS aspartate mutations do not have dominant-negative effects on cleavage of certain substrates. In our hands D257, D385E, delTM-12 and insR352 inhibit Aβ production and Notch cleavage to nearly equivalent extents. Marked inhibition of Aβ production is also reported in the Zheng paper (see related news item) where the D257A PS is expressed in vivo. To infer that PS are unlikely to be gamma-secretase from reports showing differential inhibition of Notch and Aβ production to me seems to be a stretch. In all cases, substrate accumulates. We believe that, under certain circumstances, such substrate accumulation can overcome partial gamma-secretase inhibition that occurs when these dominant-negative PS are expressed, and we have unpublished data that supports this belief. Indeed, if substrate accumulates fivefold and the enzyme activity is reduced by 20 percent, you should get the same amount of product (assuming a simple first-order kinetic model). The difference in product that is seen likely has to deal with substrate turnover, trafficking of substrates, and levels of substrate to start.
For some time I struggled with the notion that PS were gamma-secretase because one could differentially inhibit Aβ40 and Aβ42 production , whereas a knockout of PS1 affects both equally and a combined PS knockout abolishes Aβ production [13-15]. Indeed, in a grant proposal in 1999, based on data from inhibitors such as pepstatin, I hypothesized that gamma-secretase activity was likely to be attributed to two closely related aspartyl proteases. I reconcile this today by hypothesizing that PS exist in multiple conformations/complexes that may carry out different cleavages and can be differentially inhibited. If gamma-secretase can exist in various conformations or complexes that have different activities, it is easy to see how mutations and inhibitors can have differential effects on cleavage at various sites or various substrates.
In conclusion, I think the vast majority of evidence points to a catalytic role for PS in gamma-secretase cleavage. Those who believe otherwise will need to find another protease in the gamma-complex and show that it is the real gamma-secretase in order to reinvigorate the debate. Despite my confidence that PS are the catalytic component of gamma-secretase, however, I think the question whether PS are the only gamma-secretase is worthy of further evaluation. There are some reports that do suggest that Notch and AβPP can under certain circumstances undergo gamma-like cleavages that do not appear to be PS-dependent. These do not appear to be major pathways for production of Aβor NICD like products, but may very well be real. Perhaps, PSH/SPP family members will prove to be responsible.
Reply by Bart de Strooper
I basically agree with Todd Golde’s commentary. The discussion whether presenilins are the catalytic part of the gamma-secretase can be closed unless somebody comes up with a real alternative protease candidate and good evidence that it cleaves AβPP in the membrane. The counter arguments that have been brought into the discussion until now are not strong enough to be maintained, and many of them can be played down on technical grounds. I want refer all readers to an upcoming discussion in the July issue of Nature Cell Biology concerning the detection of amyloid peptide in presenilin-deficient fibroblasts, in addition to the comments from Todd and Mike.
The more interesting questions now are those that are not yet answered by the simple “presenilin is the catalytic subunit of gamma-secretase” hypothesis. I think we can all agree that we need to work further to understand the contribution of the other (known and unknown) proteins in the gamma-secretase complex to this activity. I wonder whether all complexes will be identical, given the many different substrates, and I also wonder whether all presenilins in the cell are always catalytically active. For instance, it remains surprising that a large pool of presenilins is present in cellular compartments where apparently little gamma-secretase activity is happening. I also wonder why we have presenilin 1 and 2, and why the phenotypes of the two knockout mice are so different. Finally, I tend to think that presenilins have more than one molecular function. That is because some data are really difficult to explain by the protease activity alone (e.g. the mis-sorting of telencephalin (see related news item) and other proteins in knockout cells, and the role of presenilin 1 in the Wnt signaling pathway).
Comment by Bart de Strooper—Posted 20 June 2002
The manuscript of Weihofen et al. describes excellent work that identifies a signal peptide peptidase responsible for the intramembaneous proteolytic cleavage of signal peptides. The protease is identical to the human presenilin homolog 3 (PSH3) recently identified by Ponting et al., 2002. The family of PSH consists of hydrophobic proteins containing multiple membrane domains, with conserved amino acid sequences around two aspartic residues that are similar to the sequences in the putative catalytic site of presenilins.
The authors expressed the human signal peptide peptidase (SPP) in the yeast S. cerevisiae, which apparently does not contain an ortholog, and demonstrated that the yeast extracts exerted proteolytic activity on substrates of SPP. The authors also provide preliminary evidence for a 7-membrane topology of SPP, resulting in an inverted topology of the catalytic site compared to presenilin.
It is obvious that the data support strongly the conclusion that the identified protein is a signal peptide peptidase, and that this work constitutes a major step forward in our understanding of the importance of regulated intramembrane proteolysis. One interesting question now is what other proteins were purified together with SPP, and whether they have a role in the SPP activity. Whether this paper "should quell objections of the critics" of the presenilin-is-γ-secretase hypothesis, as Wolfe and Selkoe state in their commentary, seems to me quite optimistic. I personally believe that much more work is needed before the exact functioning of the γ-secretase complex will be understood.—Bart De Strooper, KU Leuven, Belgium.