Structure of Tripeptidyl-peptidase I Provides Insight into the Molecular Basis of Late Infantile Neuronal Ceroid Lipofuscinosis*

Late infantile neuronal ceroid lipofuscinosis, a fatal neurodegenerative disease of childhood, is caused by mutations in the TPP1 gene that encodes tripeptidyl-peptidase I. We show that purified TPP1 requires at least partial glycosylation for in vitro autoprocessing and proteolytic activity. We crystallized the fully glycosylated TPP1 precursor under conditions that implied partial autocatalytic cleavage between the prosegment and the catalytic domain. X-ray crystallographic analysis at 2.35 Å resolution reveals a globular structure with a subtilisin-like fold, a Ser475-Glu272-Asp360 catalytic triad, and an octahedrally coordinated Ca2+-binding site that are characteristic features of the S53 sedolisin family of peptidases. In contrast to other S53 peptidases, the TPP1 structure revealed steric constraints on the P4 substrate pocket explaining its preferential cleavage of tripeptides from the unsubstituted N terminus of proteins. Two alternative conformations of the catalytic Asp276 are associated with the activation status of TPP1. 28 disease-causing missense mutations are analyzed in the light of the TPP1 structure providing insight into the molecular basis of late infantile neuronal ceroid lipofuscinosis.

Late infantile neuronal ceroid lipofuscinosis, a fatal neurodegenerative disease of childhood, is caused by mutations in the TPP1 gene that encodes tripeptidyl-peptidase I. We show that purified TPP1 requires at least partial glycosylation for in vitro autoprocessing and proteolytic activity. We crystallized the fully glycosylated TPP1 precursor under conditions that implied partial autocatalytic cleavage between the prosegment and the catalytic domain. X-ray crystallographic analysis at 2.35 Å resolution reveals a globular structure with a subtilisin-like fold, a Ser 475 -Glu 272 -Asp 360 catalytic triad, and an octahedrally coordinated Ca 2؉ -binding site that are characteristic features of the S53 sedolisin family of peptidases. In contrast to other S53 peptidases, the TPP1 structure revealed steric constraints on the P4 substrate pocket explaining its preferential cleavage of tripeptides from the unsubstituted N terminus of proteins. Two alternative conformations of the catalytic Asp 276 are associated with the activation status of TPP1. 28 disease-causing missense mutations are analyzed in the light of the TPP1 structure providing insight into the molecular basis of late infantile neuronal ceroid lipofuscinosis.
Mutations in the TPP1 gene (previously named CLN2 gene) encoding tripeptidyl-peptidase I (TPP1, 3 EC 3.4.14.9) result in an autosomal recessive neurodegenerative disease of childhood called "classical late infantile neuronal ceroid lipofuscinosis" (late infantile neuronal ceroid lipofuscinosis, OMIM 204500). This enzyme was first purified from bovine pituitary glands and characterized as an acidic pepstatin-insensitive aminopeptidase (1). By exploiting its affinity to mannose 6-phosphate receptors, TPP1 was subsequently found to be missing in material from late infantile neuronal ceroid lipofuscinosis patients (2). Almost at the same time, high affinity substrates and inhibitors were characterized for TPP1 originating from rat spleen (3). Further studies demonstrated a lysosomal localization and an optimal pH of 4.0 -5.0 for the tripeptidyl exopeptidase activity of TPP1. In vitro, TPP1 cleavage preferentially occurs between hydrophobic amino acids in the P1 and P1Ј positions (according to the nomenclature of Schechter and Berger (4)) and is hampered by the presence of Pro or Lys in the P1 position or Pro in P1Ј position (5). Further studies on the substrate specificity revealed that TPP1 prefers positively charged, small amino acids in positions P2 and P3 (6,7). Additionally, TPP1 shows endopeptidase activity at a pH optimum of 3.0 (8) that mediates the autocatalytic processing of the 68-kDa TPP1 precursor to the mature 46-kDa active enzyme in vitro (9,10). During processing, the primary eukaryotic 562-amino acid translation product is trimmed to the 368 C-terminal amino acids that contain the catalytic domain and all five potential N-glycosylation sites. The elimination of one particular glycosylation site at Asn 286 results in complete loss of TPP1 activity, but the other glycosylation sites might also contribute to the stability of TPP1 (11,12). At least 56 distinct disease-causing mutations in the TPP1 gene have been reported up to now, many of which disrupt folding, processing, and trafficking of TPP1 (11,13). The sequence homology of TPP1 to bacterial pepstatin-insensitive carboxyl peptidases suggests a similar catalytic mechanism involving a Glu, Asp, and Ser triad, with the conserved serine acting as catalytic nucleophile (14 -19). We have crystallized the fully glycosylated TPP1 precursor expressed in mammalian cells at conditions permissive for autocatalytic processing, and we determined the structure by multiwavelength anomalous diffraction (MAD) analysis of the selenomethionine derivative and refined it against the 2.35 Å native data. We discuss here the structural features of TPP1 and the putative effect of mutations in the TPP1 gene.

EXPERIMENTAL PROCEDURES
Materials-Cell culture medium was purchased from Invitrogen. Oligonucleotides for the introduction of the C-terminal tag were synthesized by MWG-Biotech (Munich, Germany). Molecular weight protein markers were obtained from Amersham Biosciences. Endo H (cleaves within the core of high mannose oligosaccharides from N-linked glycoproteins leaving one N-acetyl-D-glucosamine residue bound after cleavage) and PNGase F (completely cleaves off all types of oligosaccharide chains from N-linked glycoproteins) were supplied by New England Biolabs (Frankfurt, Germany). All other reagents were purchased from Sigma.
Expression of TPP1-The amino acid sequence "RSHHH-HHH" was introduced at the C terminus of the TPP1 cDNA to facilitate the purification of TPP1. Transfection of the modified TPP1 cDNA and selection of HEK 293 cells were performed as described previously (11). Both the native and selenomethionine derivative recombinant protein samples of TPP1 were purified from the cell culture supernatant. The medium was cleared by centrifugation at 3000 ϫ g and 4°C for 60 min and then filtered with a 0.2-m pore membrane. After having added 20 mM K 2 HPO 4 , pH 7.5, 0.5 M NaCl, and 40 mM imidazole, the crude solution was loaded on a HisTrap HP column (GE Healthcare). Bound TPP1 was eluted with an imidazole gradient and was typically released at 70 -90 mM imidazole. For crystallization the purified TPP1 fractions were pooled and concentrated using a centrifugal filter (Millipore, Schwalbach, Germany). The final solution contained 10 mg/ml TPP1 and was adapted to 20 mM KH 2 PO 4 , pH 4.6, and 50 mM NaCl.
Deglycosylation of TPP1-Endo H and PNGase F digestions were performed at nondenaturing conditions according to the supplier's protocol with one exception; the PNGase F treatment of autoprocessed TPP1 was done at pH 7.0 to minimize TPP1 degradation. TPP1 activity was measured according to Vines and Warburton (3). Total activity of autoprocessed TPP1 was defined as 100% and the activities of other TPP1 variants were related to it.
Mass Spectrometry-Glycosylated and deglycosylated TPP1 were incubated with trypsin (Promega GmbH, Mannheim, Germany) overnight at 37°C and analyzed by matrix-assisted laser desorption ionization (MALDI) mass spectrometry (MS) and electrospray ionization liquid chromatography tandemmass spectrometry (LC-MS/MS). The MALDI-MS analysis was performed on a 4800 MALDI TOF/TOF instrument (Applied Biosystems, Frankfurt, Germany), and the LC-MS/MS analysis was done on an LTQ Orbitrap XL (Thermo Scientific, Hamburg, Germany).
Crystallization-All solutions used for crystallization purposes were sterile-filtered and contained 0.03% sodium azide. Crystals were grown by mixing 2 l of 10 mg/ml TPP1 solution with 2 l of reservoir solution containing 7% PEG4000, 0.02 M zinc sulfate, 0.1 M sodium acetate, pH 5.0, and 0.1 M ammonium sulfate. Diffraction quality crystals appeared after 7 days. Both the native and selenomethionine derivative crystals crystallized under the same conditions. SDS-PAGE Analysis of Crystals-Selected crystals were washed in reservoir solution before dissolving them in 5 l of water and loading them in SDS buffer. SDS-PAGE showed the coexistence of both the cleaved mature and uncleaved immature TPP1 molecules in the crystals.  flash cryo-cooled using 25% glycerol and 5% MPD (in steps of 5% increasing gradient) using MiTeGen loops. A native dataset to 2.35 Å was recorded on the beamline PX2 at the Swiss Light Source on a MAR225 mosaic CCD detector. The selenomethionine derivative datasets (peak and inflection point at selenium edge) were collected with a MAR165 CCD detector on the beamline 14.2 at BESSY, Berlin, Germany. Data were processed, scaled, and analyzed using the programs HKL2000 (20), SAD-ABS, and XPREP (Bruker AXS, Madison WI).
The peak and inflection point datasets collected at the absorption edge of selenomethionine were used for the initial phasing and model building of TPP1. The structure solution using SHELXD (21) identified eight selenium sites, consistent with the presence of a dimer in the asymmetric unit. With MAD phases calculated using these sites, the high resolution native dataset was used for phase extension with a new ␣ test autotracing version of SHELXE (22) that was able to trace about half of the main chain. The crystal used for the MAD experiment diffracted poorly, accounting for the relatively high merging R values, but the R p.i.m. values of under 2% indicate that this was more than compensated for by the high redundancy of the data. This was confirmed by the very clear SHELXD solution for the selenium substructure with a correlation coefficient of 42.0% and a peak height ratio of 0.768:0.189 between the lowest correct site and the highest noise peak. Using the selenomethionine positions, bulky side chains, and biochemical properties as a guideline, the model was completed manually using COOT (23). Refmac_5.2.0019 (24) was used for structure refinement. The model was refined first isotropically to convergence and then treating each molecule in the asymmetric using as a translation, libration, and skew rigid group. The final model consists of 1050 amino acid residues, 8 N-acetyl-D-glucosamine substituents, 8 zinc ions, 2 calcium ions, and 2 sulfates. For the prosegment, amino acids Ser 20 through Ser 180 were observed, and for the catalytic domain residues, His 197 through Pro 563 and the first amino acid "Arg" of the purification tag was modeled.

Glycosylation of TPP1 Is Essential for in Vitro Autoprocessing
and Protease Activity-Previous studies have pointed to the importance of N-glycosylation for intracellular folding and trafficking of TPP1 (12,25). In particular, replacement of Asn 286 by Ser or Gln prevents intracellular processing of TPP1. To elucidate the importance of N-glycosylation for in vitro protease activity, we studied autoprocessing of purified TPP1 after complete deglycosylation and measured in vitro activity of autoprocessed TPP1 after deglycosylation. PNGase F-treated TPP1 precursor was unable to autoprocess at pH 3.5 and remained inactive, even after 24 h of incubation. Autoprocessed TPP1 almost completely lost its peptidase activity after deglycosylation with PNGase F (Fig. 1A). In contrast, partial deglycosylation by Endo H digestion did not prevent TPP1 autoprocessing nor did it significantly affect TPP1 protease activity (Fig. 1B). Control experiments with heat-inactivated PNGase F or Endo H ruled out that factors other than glycosidase activities were responsible for these effects (data not shown). These results indicate that TPP1 requires some N-linked glycosylation for cell-free peptidase activity. Tryptic digests of glycosylated and deglycosylated TPP1 were analyzed by MALDI-MS and electrospray ionization-LC-MS/MS for additional modifications that could explain the loss of TPP1 activity. No abnormalities were detected.
Partial Autoprocessing of TPP1 Precursor during Crystallization-Human TPP1 was engineered with a C-terminal His 6 tag, overexpressed in HEK 293 cells, and purified by metal ion affinity chromatography. Fully glycosylated TPP1 precursor was incubated at crystallization conditions that allowed slow autoprocessing. Crystals appeared after 7 days of incubation at pH 4.9 and room temperature. SDS-PAGE analysis of representative crystals showed the presence of TPP1 precursor as well as autoprocessed TPP1 along with the TPP1 prosegment (Fig. 2). This result points to the coexistence of unprocessed and processed TPP1 within the crystals and is consistent with experiments showing that TPP1 autoactivation is relatively slow at pH 4.5-5.5 and does not involve degradation of the TPP1 prosegment (10).
Overall Structure and Fold of TPP1-TPP1 is globular and has an overall fold similar to that of subtilisin and to members of the S53 family of sedolisin peptidases such as sedolisin and kumamolisin. The asymmetric unit contains two TPP1 monomers. Each monomer is a complex of two chains, the prosegment and the catalytic domain (Fig. 3A). The data collection and refinement statistics are displayed in Tables 1 and 2. The pro-segment spans residues Ser 20 through Ser 180 and has a calculated molecular mass of 17 kDa. It is composed of five ␣-helices formed by ␣1, ␣2, ␣3, ␣4, and ␣5 and two sets of antiparallel ␤-strands, formed by ␤1-␤5-␤6 and ␤7-␤2-␤4-␤3, and stabilized by a disulfide bridge between Cys 111 and Cys 122 on ␤3 and ␤4, respectively (Fig. 3B).
The catalytic domain contains residues His 197 through Pro 563 and has a calculated molecular mass of 39.7 kDa. It contains 10 ␣-helices and 14 ␤-strands. The core of the TPP1 molecule includes seven all-parallel ␤-strands ␤8-␤9-␤10-␤11-␤12-␤15-␤17 and is sandwiched and prominently flanked by four helices ␣8, ␣9, ␣11, and ␣13 near the active site. There are two disulfide linkages, the first between Cys 365 and Cys 526 and the second between Cys 522 and Cys 537 , stabilizing the antiparallel ␣-strands ␤20 and ␤21 (Fig. 3, C and D). Two proline residues, Pro 378 and Pro 449 , are found in cis conformation. The schematic topology of the TPP1 structure is given in Fig. 4.
Each molecule has an octahedrally coordinated Ca 2ϩbinding site, coordinated by carboxylate oxygens of Asp 517   Ramachandran plot (%) Residues in preferred regions 91.6 Residues in allowed regions 6.9 Residues in disallowed regions 1.5 and Asp 543 , the main chain carbonyls of Val 518 , Gly 539 , and Gly 541 , and a water molecule (Fig. 5A). Also, there are four Zn 2ϩ -binding sites in each of the TPP1 units that are quite likely to be nonphysiological and a result of the crystallization procedure (Fig. 5B). As expected, glycosylations were observed at residues Asn 210 , Asn 286 , Asn 313 , and Asn 443 and were modeled as N-acetyl-D-glucosamine. No electron density is observed for the glycosylation site at residue Asn 222 in both monomers. The residues Ser 181 to Leu 196 could not be modeled as no density is observed for these residues (Fig. 3A). This might be explained by the variation of the cleavage sites at pH values 4.5-5.5 that result in a slow TPP1 processing and variable cleavage after residues Ser 181 , Glu 189 , and Gly 195 (10). However, no other structural ambiguity was found in any of the regions near the active site, indicating that autoprocessing is not associated with any major change in the secondary structure of TPP1. Attempts to refine group occupancies of the prosegment and the catalytic domain indicate that both are probably fully occupied, suggesting that the prosegment is homogeneously located within the crystals no matter whether the TPP1 precursor is cleaved or not. These data are in agreement with the lack of differences in the CD spectra measured after TPP1 autoprocessing (10) and consistent with recent studies suggesting that the TPP1 prosegment remains tightly bound to the catalytic domain at pH 4.5 even after cleavage (26). Multiple molecular contacts between the prosegment and the catalytic domain explain this strong binding that covers ϳ21 and 15% of the total solvent accessible surface areas of the prosegment and the catalytic domain (if glycosylation is not taken into account), respectively (supplemental Fig. S1). The catalytic domain and the prosegment contribute 56 and 65 residues, respectively, to form the interface between the two domains. Salt bridges are formed predominantly between the guanidinium group of Arg 339 and the carboxylates of Asp 118 , the ⑀-amino group of Lys 346 and carboxylates of Asp 118 , carboxylates of Glu 302 , and the guanidinium group of Arg 175 . In addition, there are several hydrogen bonds stabilizing the complex between the prosegment and the catalytic domain (supplemental Table S1). The strongest interface interactions involve polar amino acid side chains and hence are sensitive to pH changes. Hence, lowering the pH will release the catalytic domain from the trap formed by the prosegment and will allow autoproteolytic cleavage of the linker.

Homology of TPP1 to Members of the Sedolisin Family of Serine
Proteases-Although the amino acid sequence identity of TPP1 relative to other members of the S53 family of peptidases is rather low, there is a considerable structural homology. The TPP1 precursor shares only 18.7% of its amino acid sequence with pro-kumamolisin that is currently the only reported structure of a full-length S53 peptidase. However, superposition of the reported structure of pro-kuma-  Residues that are involved in the catalytic mechanism are marked by red letters; the 6 cysteine residues that form disulfide bridges are colored yellow on black background, and the asparagine residues that carry N-linked oligosaccharides are highlighted by a turquoise background. No density could be observed for a possible glycosylation at Asn 222 (highlighted with gray background). The linker region (Ser 181 -Leu 196 ) was poorly defined (gray letters). The first amino acid R of the purification tag RSHHHHHH is displayed in gray. molisin (17) with this full-length TPP1 structure reveals a 1.64 Å root mean square deviation of the 376 matching C-␣ atoms (supplemental Fig. S2). The TPP1 and kumamolisin prosegments (TPP1 residues 20 -180) have a 1.67Å root mean square deviation of the 125 matching C-␣ atoms, and the two catalytic domains (TPP1 residues 197-563) have a 1.47 Å root mean square deviation of the 270 matching C-␣ atoms. This indicates that the structural homology between the TPP1 precursor and pro-kumamolisin is high and that the catalytic domains of the two proteases match even better than their prosegments. The latter finding is underlined by the fact that the catalytic domains are longer and yet show a lower deviation. The crystal structure of TPP1 reveals a subtilisin like fold and contains a Ser 475 -Glu 272 -Asp 360 catalytic triad similar to kumamolisin and sedolisin (15) (Fig. 6A). Ser 475 , the nucleophile, resides on helix ␣13, and Glu 272 , the general base catalyst along with Asp 276 , lies on helix ␣8. Asp 360 lies on a loop connecting the strands ␤12 and ␤13, whereas Asp 327 is observed on the loop connecting strand ␤11 to helix ␣10 (Fig. 3D). Superposition of the catalytic center of TPP1 with that of kumamolisin, sedolisin, and subtilisin discloses the high degree of structural homology among them (Fig. 6, A and B). The overall structure of the catalytic domains of TPP1, kumamolisin, and sedolisin resemble with each other, and even the subtilisin fold contains similar regions (supplemental Fig. S3, A-C). Although the TPP1 prosegment bears good similarity to the kumamolisin prosegment, both differ significantly from the corresponding regions of subtilisin (supplemental Fig. 3D). The calcium ion is observed at a position homologous to that of other reported sedolisin structures (Fig. 3, C and D, and Fig. 5A).

DISCUSSION
We have demonstrated the importance of N-glycosylation for in vitro autoprocessing and protease activity of TPP1. These results extend previous studies on the intracellular processing of TPP1 that showed the effect of mutagenesis of one or several N-glycosylation sites on intracellular folding and trafficking of TPP1 (12). Our experimental data reveal that in vitro unprocessed or processed TPP1 loses its protease activity after complete deglycosylation by PNGase F treatment. This loss of activity cannot be explained by instability, because deglycosylated TPP1 precursor showed no evidence for accelerated degradation (Fig. 1A). In addition, previously autoactivated and thus presumably properly folded TPP1 protease becomes inactive after complete deglycosylation with PNGase F. Interestingly, partial deglycosylation by Endo H treatment is not associated with a disturbed autoprocessing and loss of protease activity. This suggests that either the complex N-linked oligosaccharides that are not cleaved by Endo H or the terminal N-acetyl-D-glucosamine residues that remain after Endo H cleavage are critical for autoprocessing and protease activity of TPP1. This finding is supported by the fact that Endo H-deglycosylated TPP1 precursor is indistinguishable from fully glycosylated TPP1 in terms of autocatalytic processing of the precursor and enzymatic properties of the mature protease (accompanying paper by Guhaniyogi et al. (28)). It can be speculated that complete deglycosylation might lead to a significant conformational alteration or that N-linked oligosaccharides are involved in the catalytic process that is mediated by TPP1. Further studies are required to determine the precise oligosaccharide components and molecular interactions that are essential for TPP1 activity.
We have crystallized fully glycosylated TPP1 at conditions that allow slow autoprocessing and obtained an ϳ50:50% mixture of unprocessed and processed TPP1 within the crys-FIGURE 6. A, superposition of the catalytic residues of TPP1, kumamolisin (PDB ID 1KDV), and sedolisin (PDB ID 1SI0). The stick diagram displays the corresponding side chains of the catalytic residues of TPP1 (blue), kumamolisin (red), and sedolisin (orange) positioned on the cartoon model of the ␣-carbon backbone of TPP1 (gray). Note that Glu 80 is displayed in two alternative conformations reported for sedolisin. The high structural similarity of the catalytic center of TPP1 with that of kumamolisin and sedolisin includes the type of amino acids as well as their relative positions to each other. B, superposition of the catalytic residues of TPP1 and subtilisin (PDB ID 1CSE). The stick diagram displays the corresponding side chains of the catalytic residues of TPP1 (blue) and subtilisin (green) positioned on the cartoon model of the ␣-carbon backbone of TPP1. The overall structure of TPP1 and subtilisin are similar. However, a number of structural details vary, and both proteases possess different catalytic triads. TPP1, Ser 475 -Glu 272 -Asp 360 versus subtilisin Ser 221 -His 64 -Asp 32 . Additionally, the positions of the catalytic residues to each other are not conserved.
tals. Nevertheless, the resulting crystal structure did not suffer from any major structural ambiguities, and only the linker region carrying the potential cleavage sites for autoprocessing was poorly defined. As pointed out by Guhaniyogi et al. (28), there is a good agreement between the structures of the glycosylated TPP1 precursor and the Endo H-deglycosylated one (PDB ID 3EDY). Furthermore, the crystal structure of pro-kumamolisin, the most homologous bacterial member of the sedolisin family, showed that the connecting linker runs through the entire active site cleft (17). Superposition of pro-kumamolisin with the active kumamolisin revealed that the catalytic domain of the proenzyme exhibits an almost identical structure and is already properly prefolded within the proenzyme (18). Considering the high degree of structural homology between kumamolisin and TPP1, it can be assumed that also the TPP1 linker runs through the active site cleft of TPP1 and hence is partially cleaved and therefore disordered within our crystals. However, the cleaved off prosegment remains bound to the catalytic domain at pH 4.5 (or higher pH), as has been demonstrated recently (26). In conclusion, our data suggest that uncleaved and cleaved off TPP1 prosegments show virtually identical structures and positioning and that the catalytic domains of unprocessed and processed TPP1 possess very similar structures as well. The lack of significant structural differences between unprocessed and processed TPP1 is supported by the absence of alterations in CD spectroscopy observed after TPP1 processing (10). However, we find densities for two conformations of Asp 276 , one position superposing with the equivalent residues in kumamolisin (Asp 82 ) and sedolisin (Asp 84 ) and the other superposing with the Endo H-deglycosylated TPP1 precursor (PDB 3EDY) that is not autoactivated (Fig. 8). Although we did not model alternative conformations in view of the modest resolution, it is likely that these two conformations reflect the active and inactive state of the catalytic center. As described below, Asp 276 is involved in the catalytic process that requires the proper positioning in the catalytic center.
The high degree of structural homology between TPP1 and other members of the sedolisin family also implies that the catalytic mechanism that mediates the proteolytic cleavage is very similar. The active site of TPP1 includes the three amino acids Ser 475 , Glu 272 , and Asp 360 from which the name of the SEDolisin family of peptidases is derived. During the catalytic process, the Ser 475 hydroxyl group loses the O-␥ proton and so becomes a stronger nucleophile. The only plausible proton acceptors are the carboxylic oxygens of Glu 272 . In the approaching polypeptide substrate, the scissile peptide bond is oriented toward Ser 475 O-␥ with its carbonyl group inserting into the oxyanion hole, fixed by subsite interactions of both flanking peptide moieties (Fig. 7). This would allow the nucleophilic Ser 475 O-␥ to attack the polarized carbonyl group of the scissile peptide bond to form a tetrahedral intermediate and to pass through the transition state with simultaneous proton transfer from the acidic Ser 475 oxonium cation via the Glu 272 carboxylic group to Asp 276 . The Asp 360 side chain helps to create the oxyanion hole, stabilizing the tetrahedral intermediate of the reaction. Apart from the nucleophilic Ser 475 , all other residues participating in the catalytic mechanism are acidic amino acids. To mediate the catalytic electron transfer in both directions, these residues have to exist at equilibrium between the protonated and deprotonated states. Thus the surrounding pH should be close to their pK a values, which explains the requirement for an acidic environment for catalysis.
The fundamental difference between TPP1 and other members of the sedolisin family of peptidases is the orientation of cleavage. Whereas all known prokaryotic members of the sedolisin family are putative carboxyl peptidases, TPP1 preferentially cleaves tripeptides from the N-terminal end of proteins. This preference can be explained by a unique hydrogen bond between Tyr 325 and Asp 327 (distance 3.24 Å) that can sterically constrain the P4 residue of the substrate (Fig. 8). Tyr 325 interacts with Phe 304 by -stacking and is localized within a hydrophobic pocket with several stacked aromatic systems. All these interactions occlude the putative S4 subsite and thus explain the preference for substrates with P1-P3 residues and a free unsubstituted N terminus at P3. Crystallization of complexes between TPP1 and substrates/inhibitors might prove this hypothesis.
More than 50% of the known TPP1 mutations represent single amino acid substitutions that result in the biosynthesis of TPP1 mutants. None of these missense mutants shows any measurable residual TPP1 activity, and all are associated with severe symptoms such as developmental regression, epileptic seizures, ataxia, and progressive loss of motor abilities, speech and vision. On the basis of the resolved TPP1 structure, we have described the molecular effects of all currently known missense mutations (Table 3 and Fig. 9). The majority of these mutations would disrupt folding leading to instability and rapid degradation. The amino acid substitutions V277M, Q248P, G284V, G473R, and S475L probably compromise the active center and result in loss of proteolytic activity. Only two mutations, G77R and S153P, are located within the prosegment and are likely to disturb processing of TPP1. The N286S substitution results in loss of one glycosylation site, which leads to almost complete loss of protease activity without obvious alterations. Also, no apparent conformational destabilization is observed for the missense mutations V216M, R266Q, and K428N. Further biochemical and structural analyses are underway to elucidate the mech- The position of the pentapeptide is derived from the eglin c backbone in complex with subtilisin Carlsberg (27). The hydrogen bonding between Asp 327 and Tyr 325 (distance 3.24 Å) constrains the space occupied by the P4 substrate residue resulting in a clash. For the Asp 276 residue an additional secondary conformation is indicated by Asp 276 Ј. Interestingly, this conformation is the only one present in the structure of Endo H-deglycosylated TPP1 (PDB ID 3EDY), which does not autoprocess itself. In the Asp 276 Ј position, this residue cannot participate in the proton transfer of the catalytic mechanism and hence most likely reflects the inactive conformation. Ser 475 ϭ S475, Glu 272 ϭ E272, Asp 276 ϭ D276, Asp 360 ϭ D360.

TABLE 3 Molecular effect of single amino acid substitutions
Reported amino acid substitutions are listed from the N terminus to the C terminus of full-length TPP1. The positions of the affected residues are described in respect to the TPP1 structure; ␣ corresponds to ␣-helix and ␤ to ␤-strand. Most of the amino acid substitutions lead to destabilization of folding and compromise the proteolytic activity. Two alleles, R127Q and H175R, which are associated with an almost normal TPP1 activity, have not been included in the table.

Mutation
Structural position Putative functional consequence anisms that explain the functional loss of TPP1. The thorough biochemical and structural analyses of these mechanisms are fundamental for the design of small molecule compounds that are meant to rescue TPP1 activity and hence to prevent or protract disease progression. The resolution of the TPP1 structure has not only extended our understanding of the molecular mechanisms underlying late infantile neuronal ceroid lipofuscinosis but could also lead to novel therapeutic options for this devastating disorder.  Fig. 3D and rotated. The currently known pathogenic TPP1 missense mutations are mapped onto the TPP1 structure as labeled spheres and are colored according to their impact on the TPP1 structure. Red colored spheres point to mutations compromising the catalytic activity; blue colored spheres indicate conformational destabilization, and gray colored spheres designate unclear structural consequences. The Asn 286 residue (turquoise) corresponds to one of the five N-glycosylation sites. B, view of the catalytic cleft of the cartoon model in Fig. 3D. TPP1 missense mutations are shown as stick models, coloring scheme corresponds to A, except for mutations with unclear effect that are colored in green.