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

1. Introduction 

Bone morphogenetic protein-1 (BMP-1)/tolloid (TLD) family proteinases serve two main purposes; firstly the processing of extracellular matrix (ECM) precursors which range from structural components such as fibrillar collagens [1], enzymes [2], cellular anchoring proteins [3], [4] and small leucine-rich proteoglycans [5], [6]. Secondly, the release of transforming growth factor (TGF)-β superfamily members, including BMP-2 and -4, growth and differentiation factors (GDF) 8/11 and TGF-β1 from their latent complexes. Through these actions, BMP-1/TLD proteinases affect processes ranging from dorsal ventral patterning, growth of skeletal muscle, neurogenesis, to cellular behaviour [7], [8], [9], [10].

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 factor-like (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.

View Large Image Download to PowerPoint

Fig. 1. (A) Domain structures are displayed schematically. MP, metalloproteinase domain; C, CUB domain; E, EGF-like domain; T, C-terminal sequence. Protein derived from the Bmpl gene is shown in white and from the Tll1 gene in grey. (B) Reducing SDS–PAGE of TLL-1, molecular weight markers in kDa are indicated. (C) MALLS analysis of TLL-1, in the presence of 1mM CaCl2 the molecular mass was 219500Da (i) but in 2mM EDTA it was 105600Da (ii). The theoretical molecular mass of TLL-1 (including purification tag) is 100465 Da.

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.

2. Materials and methods 

3. Results 

Purified TLL-1 (Fig. 1B) eluted from SEC-MALLS in buffer containing calcium as a single peak with a mass of 219500Da (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 105600Da, 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 self-association (Kd=2nM at 0.5mM [CaCl2]).

Table 1. Data was collected by (a) MALLS, (b) AUC: molecular weight (MW) in Daltons; hydrodynamic radius (Rh) in nm; sedimentation coefficient (S20,w) and frictional coefficient (f/f0). Theoretical MWs (including purification tag) are 100465Da (TLL-1), 68341Da (TLL1TC3-T) and 67113Da (TLL1TC3). ND=not determined.

		MWa	MWb	Rh,b	S20,W	f/fo
	TLL-1 Ca2+	219500±659	196000	5.52	8.88±0.15	1.44
			73900	5.33	4.59±0.21	1.75
	TLL-1 EDTA	105600±12672	ND	ND	ND	ND
	TLL1TC3 Ca2+	ND	67100	3.36	4.91±0.33	1.26
	TLL1TC3-T Ca2+	ND	65990	3.55	4.73±0.38	1.32
	TLL1TC3-T EDTA	ND	67200	3.67	4.59±0.32	1.36

To investigate the dimeric arrangement of TLL-1, we used sedimentation velocity AUC. In the presence of 1mM calcium-ions, the predominant species has an S20,W of 8.88 (Fig. 2A). This species has a mass estimate of 196000Da, 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 (Rh) of 5.33nm (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).

View Large Image Download to PowerPoint

Fig. 2. (A) Sedimentation velocity AUC of TLL-1in the presence of 1mM CaCl2. The majority of protein is present as a dimer at 8.3 Sapp. (B) Bead models generated for the TLL-1 monomer. A horseshoe-like conformation gives the closest fit to experimental Rh of 5.33nm. (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=21.6nm). Potential dimer packing models are shown in more detail in [14].

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 Sapp (Fig. 3A) suggesting almost all protein is monomeric (Table 1). Since there are no species at higher (>4) Sapp in samples containing calcium, we conclude that TLL1TC3-T does not form calcium-ion dependent dimers.

View Large Image Download to PowerPoint

Fig. 3. (A) Sedimentation velocity AUC of TLL1TC3-T in the presence of 1mM CaCl2 (solid line) and 2mM EDTA (dotted line). The peak at 3.5 Sapp corresponds to the monomer. (B) Non-reducing SDS–PAGE of truncated TLL-1 variants. Molecular weight markers in kDa are indicated. (C) TLL1TC3 appears almost identical to TLL1TC3-T when analysed by sedimentation velocity AUC. (D) Silver stained SDS–PAGE of chordin following incubation with enzymes, the chordin domain structure is shown (CR, cysteine-rich repeat; myc, c-myc epitope; (H)6, poly-histidine tag; arrows indicate cleavage sites). (E) Densitometric analysis of chordin cleavage is shown for each enzyme represented as the amount of product (intermediate fragment) relative to substrate (full-length chordin). Data are the mean of three independent experiments±S.E. (F) Table showing the values from the densitometric analysis used to plot (E).

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.

4. 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 probiglycan. 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 C-terminal 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 K724. Our domain boundary was determined by analysis of the intron-exon boundary and contained an additional six amino acids (up to D730). 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 self-association 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 self-association 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.