Structural and functional evidence for a substrate exclusion mechanism in mammalian tolloid like‐1 (TLL‐1) proteinase

MINT‐7386098: BMP‐1 (uniprotkb:P13497) cleaves (MI:0194) Chordin (uniprotkb:Q9H2X0) by protease assay (MI:0435)

BMP-1/TLD proteinases have an N-terminal protease domain related to the digestive enzyme astacin [11] followed by non-catalytic CUB (Complement/Uegf/BMP-1) and epidermal growth factorlike (EGF) domains and a short C-terminal sequence (Fig. 1A). mTLD, TLL-1 and TLL-2 have 5 CUB and 2 EGF-like domains whereas BMP-1 lacks two CUB and one EGF C-terminal domains. There is evidence that the non-catalytic domains have a negative regulatory role [12,13] and we recently provided an explanation as to why mTLD is a poor proteinase relative to BMP-1 [14]. Our findings illustrated that BMP-1 is a monomer whereas mTLD forms a dimer in the presence of calcium-ions, which restricts its chordinase activity by a substrate exclusion mechanism. It is unknown whether this mechanism extends to other BMP-1/TLD family members, and in particular TLL-1 which is a more efficient proteinase than mTLD.
In the present report, we provide evidence that TLL-1 forms a dimer and show that truncated forms of TLL-1 having the same domain structure as BMP-1 are monomers, irrespective of the presence/absence of the unique C-terminal sequence. Surprisingly, the truncated TLL-1 molecules have higher chordinase activity not only than full-length TLL-1, but also BMP-1. Together these data provide evidence that the protease activity of TLL-1 is restricted by substrate exclusion.

Expression and purification of recombinant proteins
TLL-1 was generated from an IMAGE clone comprising the first 1870 base pairs and the remaining 1169 base pairs synthesized by Genescript with a 3 0 V5-His (6) tag and ligated into the pCEP4 vector (Invitrogen) using Kpn1 and XhoI. TLL-1 molecules terminating after CUB3 with (TLL1TC3-T) and without (TLL1TC3) the C-terminal sequence, were amplified by PCR using TLL-1 and ligated into the pCEP-Pu vector using Not1 and Xho1. 293-EBNA cells were cultured and transfected as described in [14]. TLL-1 and variants were purified using a combination of nickel affinity and size exclusion chromatography as were BMP-1, mTLD and chordin as described in [14]. Enzyme and chordin concentrations were quantified based on comparison to known amounts of bovine serum albumin (BSA) using SDS-PAGE and GeneTools software (SynGene). A range of defined BSA quantities are visualised and a calibration curve plotted. The amount of enzyme in a known volume can then be determined.

Multi-angle laser light scattering (MALLS)
Samples (0.5 ml) were separated in 10 mM Tris-HCl (pH 7.4) containing 0.5 M NaCl in the presence of either 1 mM CaCl 2 (Superdex-200 10/300 GL column) or 2 mM EDTA (Superdex-75 10/300 GL column) at 0.71 ml/min and passed through a Wyatt EOS 18-angle laser photometer with the 13th detector replaced with a Wyatt QELS detector. This was coupled to a Wyatt Optilab rEX refractive index detector and the hydrodynamic radius, molecular weight moments and concentration of peaks analysed using Astra5.3.2.

TEM
TLL-1 (10 lg/ml) in the presence of 1 mM CaCl 2 was absorbed onto EM grids and stained with 4% (w/v) uranyl acetate and observed in an FEI Tecnai12 TEM at 120 keV. Images were recorded on a CCD camera at 69 000Â magnification between 0.5 and 1.5 lm defocus and processed using Imagic5 software [18]. Four hundred and eighty-five particles were band-pass filtered with high and low frequency cut-offs of 20 Å and 180 Å.

Activity assays
Purified chordin (2 lg) was incubated in the presence or absence of 30 ng enzyme in 50 mM Tris-HCl (pH 7.4) containing 150 mM NaCl and 5 mM CaCl 2 at 37°C for 4 h [19]. Reactions were stopped by adding LDS sample buffer (Invitrogen) and heating to 95°C for 5 min. Reaction products were separated by SDS-PAGE and visualised by silver staining. Chordinase assay products were quantified by densiometry using SynGene software.

Results
Purified TLL-1 (Fig. 1B) eluted from SEC-MALLS in buffer containing calcium as a single peak with a mass of 219 500 Da (Fig. 1C(i)) suggesting TLL-1 is a dimer in solution. To determine whether the TLL-1 dimer requires the presence of calcium, we analysed samples in the presence of EDTA (Fig. 1C(ii)). Under these conditions, the mass was reduced to 105 600 Da, in good agreement with a monomer ( Table 1). The strength of TLL-1 self-association was assessed using sedimentation equilibrium AUC (data not shown). TLL-1 was predominantly dimeric with very strong selfassociation (K d = 2 nM at 0.5 mM [CaCl 2 ]).
To investigate the dimeric arrangement of TLL-1, we used sedimentation velocity AUC. In the presence of 1 mM calcium-ions, the predominant species has an S 20,W of 8.88 ( Fig. 2A). This species has a mass estimate of 196 000 Da, similar to that expected for a dimer.  To gain insight into the structure of the TLL-1 monomer we generated bead models and a horseshoe-like conformation (Fig. 2B) gave the best fit to the experimentally-derived hydrodynamic radius (R h ) of 5.33 nm ( Table 1). We used single particle TEM to investigate the arrangement of the dimer, following classification it was evident that TLL-1 particles were present in a single orientation showing side-by-side stacking of monomers (Fig. 2C).
To investigate if TLL-1 is restricted by a substrate exclusion mechanism we made a TLL-1 variant with the same domain structure as BMP-1 (TLL1TC3-T with unique C-terminal sequence). Due to aggregation we were unable to obtain a mass for TLL1TC3-T using MALLS, therefore we analysed protein samples by sedimentation velocity AUC. In the presence of EDTA or calcium, the c(s) profile are the same with a predominant species at 3.5 S app (Fig. 3A) suggesting almost all protein is monomeric (Table 1). Since there are no species at higher (>4) S app in samples containing calcium, we conclude that TLL1TC3-T does not form calcium-ion dependent dimers.
In a previous report, a construct analogous to TLL1TC3-T lacking the C-terminal sequence formed oligomeric aggregates observed on non-reducing SDS-PAGE thought to be due to misfolding [12]. To test this possibility, we generated TLL1TC3 which lacks the C-terminal sequence. Neither variant (with or without the C-terminal sequence) showed evidence of aggregation (Fig. 3B). To confirm the oligomeric status of TLL1TC3 we used sedimentation velocity AUC. In the presence of calcium-ions the c(s) profile is almost identical to that of TLL1TC3-T (Fig. 3C), suggesting that TLL1TC3 is also monomeric.
To determine the proteolytic activity of the TLL-1 variants, we compared their chordinase activity to that of mTLD, BMP-1 and TLL-1 (Fig. 3D and E). Chordinase activity was detected for mTLD but this was considerably lower than that of BMP-1 and TLL-1, which appear to have similar levels of chordinase activity. However, the truncated forms of TLL-1 are approximately twice as efficient as TLL-1 at processing chordin.

Discussion
In this study we find that TLL-1 forms a calcium-dependent dimer via a side-by-side stacking of monomers and that the dimer has a much tighter association than mTLD. We also show that the region responsible for self-association resides in the C-terminal three domains. These findings suggest dimerisation is most likely a feature of all mammalian BMP-1/TLD family members that have EGF2, CUB4 and CUB5. Based on these data, we believe that the proteolytic activity of the TLL-1 dimer is restricted by a substrate exclusion mechanism that requires the presence of the C-terminal domains. This conclusion is based on the observations that monomeric forms of TLL-1 lacking the C-terminal domains are more efficient chordinases than the full-length molecule and these ''BMP-1 forms" of TLL-1 are also more efficient chordinases than BMP-1. This latter observation suggests that differences in activity between TLL-1 and mTLD are due to variations in amino acid residues rather than gross mechanistic differences. This is further supported by an independent study in which the proteinase activity of a construct analogous to TLL1TC3-T was investigated [12]. This molecule was described as being more efficient than full-length TLL-1 at cleaving procollagen I and probigly-  Fig. 2. (A) Sedimentation velocity AUC of TLL-1 in the presence of 1 mM CaCl 2 . The majority of protein is present as a dimer at 8.3 S app . (B) Bead models generated for the TLL-1 monomer. A horseshoe-like conformation gives the closest fit to experimental R h of 5.33 nm. (C) The TLL-1 dimer was investigated using single particle TEM. (i) Three representative classes are shown; the number of particles in each class is indicated (box size = 21.6 nm). Potential dimer packing models are shown in more detail in [14].
can. However, what was also apparent in their results, is that this molecule was also more efficient than BMP-1. These data support our findings and suggest the substrate exclusion mechanism that restricts the activity of TLL-1 extends to other substrates.
We also find that the presence or absence of the C-terminal sequence has little effect on the chordinase activity of truncated TLL-1 molecules. In a previous report, it was observed that a molecule analogous to TLL1TC3 was a poor C-proteinase whereas with the Cterminal sequence it was highly efficient [12]. Since the form lacking the C-terminal sequence was observed to form aggregates, these differences were thought to be due to misfolding, and it was hypothesised that the C-terminal sequence was required for correct folding. In contrast, our studies reveal the presence or absence of the C-terminal sequence has no effect on the oligomeric status or chordinase activity of these molecules. These discrepancies can be explained by small differences in the CUB3 domain. In the previous report, the molecule terminated after K 724 . Our domain boundary was determined by analysis of the intron-exon boundary and contained an additional six amino acids (up to D 730 ). Presumably these aided in the correct folding of our molecules.
The difference in activity between mTLD and TLL-1 could be due to subtle differences in the intrinsic affinity of the BMP-1/mTLD and TLL-1 protease domains for substrates. In support of this, chordin cleavage by the isolated protease domains from BMP-1/mTLD and TLL-1 showed that the TLL-1 domain is significantly more efficient [12]. Alternatively, differences in the affinity of the non-catalytic domains for chordin may provide an explanation. Despite   the similarities in overall architecture, the strength of TLL-1 selfassociation is greater than mTLD. It is conceivable that this may result in different patterns of activity in vivo. The expression of Bmpl and Tll1 overlap in a number of tissues [20], differences in selfassociation could favour the effect of one proteinase depending on the calcium and/or protein concentrations present. Our findings also raise the possibility of mTLD/TLL-1 heterodimers, which could add another layer of complexity.