Main

Viruses hijack the biochemical activity of specific cell surface proteins to initially recognize the target cell and subsequently overcome the energetic barrier necessary for virus–host membrane fusion. SARS-CoV-2 uses its Spike (S) protein to recognize its primary cell surface receptor, ACE2 (refs. 1,2), which is highly expressed in the aerodigestive tract, liver, kidneys and sex organs3. However, to bring virus and host membranes sufficiently close for fusion, the S protein requires proteolytic processing at the S1/S2 cleavage site to enable large conformational changes and expose subunit 2 (S2) harboring the matured membrane fusion machinery (Fig. 1a)4,5. Transmembrane protease, serine 2 (TMPRSS2), which colocalizes with ACE2 at the cell membrane6, has been identified as the dominant proteolytic driver of S protein activation and SARS-CoV-2 infection of the aerodigestive tract1,7,8.

Fig. 1: Engineered activation and structural characterization of stabilized TMPRSS2 ectodomain.
figure 1

a, Membrane-bound TMPRSS2 zymogen undergoes autocleavage activation at the Arg255-Ile256 peptide bond and the matured enzyme proteolytically processes SARS-CoV-2 Spike protein (magnified) docked to the ACE2 receptor (yellow) to drive membrane fusion. b, Engineered recombinant TMPRSS2 ectodomain (dasTMPRSS2) containing the LDLR-A domain, a Class A SRCR domain and a C-terminal trypsin-like S1 peptidase (SP) domain, features a DDDDK255 substitution to facilitate controlled zymogen activation. Purified, concentrated dasTMPRSS2 can self-activate. The noncatalytic (LDLR-A + SRCR) and catalytic (SP) chains are tethered by a disulfide bond and the activation status can be interrogated by SDS–PAGE under nonreducing and reducing (5% β-mercaptoethanol) conditions. Gel results are consistent with n ≥ 3 independent biological experiments. c, X-ray crystal structure of dasTMPRSS2 pretreated with nafamostat, resulting in phenylguanidino acylation (gray sticks). d, Close-up view of the SP catalytic triad residues (His296, Asp345 and Ser441) and the postactivation Asp440:Ile256 salt bridge showing complete maturation of the protease. Polar contacts are shown as yellow dashed lines. e, The interdomain disulfide (Cys244-Cys365) maintains covalent attachment of the SRCR and SP domains.

Source data

Full size image

TMPRSS2 belongs to the type 2 transmembrane serine protease (TTSP) family that comprises 19 surface-expressed trypsin-like serine proteases that normally participate in pericellular proteolytic cascades for degradative remodeling of the extracellular matrix9 and proteolytic activation of membrane proteins10, among other key epithelial homeostasis roles11. TTSP family members experience complex posttranslational regulation. All TTSPs are glycoproteins expressed as single-chain proenzymes (or zymogens) that require proteolytic cleavage activation at a conserved (Arg/Lys)-(Ile/Val) peptide bond directly preceding their catalytic domains to mature their active site through interaction with the newly generated N terminus (Fig. 1a)12. Some TTSPs can undergo autocleavage activation13,14 while others appear to require external activation by another protease15,16. On activation, TTSPs can recruit endogenous, Kunitz-domain containing proteinaceous inhibitors that block substrate access and are themselves regulated to modulate active TTSP activity17,18. Dysregulated TTSP activity is a common feature that arises in cancers and results in increased tumor cell proliferation, invasiveness and metastasis19,20, yet even basal activity levels of TMPRSS2, −4 (refs. 6,21,22), −11d (refs. 22,23) and −13 (refs. 22,23) in normal tissues can be exploited by coronaviruses24, influenza A25,26, and influenza B26 viruses for efficient activation and infection at the cell membrane. This family of proteases, and particularly TMPRSS2, therefore represent prime targets for therapeutic intervention to disable aggressive cancers and block initial viral infection. Antivirals that can effectively disable essential TTSP family members responsible for complete activation of the S protein8,27 may be drugs insensitive to new variant mutations owing to the conserved virus–host tropism with these proteases5,24,26.

Despite this overwhelming motivation, TMPRSS2 and many other TTSPs have resisted therapeutic targeting campaigns in large part due to the difficulty associated with procuring active sources of recombinant enzyme conducive to rapid inhibitor screening and X-ray structural characterization of protein–inhibitor complexes. TTSPs are disulfide-rich, require site-specific proteolytic activation and pose inherent host cell toxicity challenges during recombinant production if proteolytic activity is uncontrolled during overexpression. Furthermore, TTSP inhibitors themselves pose unique pharmacological challenges. Since they are extracellular targets, the drugs targeting TTSPs may not need to cross the plasma membrane (if administered intravenously), but instead they must be exceedingly selective so as to avoid the concentrated pools of trypsin-like proteases in plasma responsible for coagulation cascades. TTSP drugs must also be biologically stable and have favorable biodistribution profiles to engage the target tissue with sufficient residence time to exert meaningful protective (or anti-tumor) effects. As such, key structural insights are required to design TTSP-specific or pan-TTSP therapeutics.

Here we report a reliable method to produce highly active TMPRSS2, enabling the investigation of TMPRSS2-mediated SARS-CoV-2 S protein activation and associated inhibition by a panel of known clinical serine protease inhibitors. We determined the TMPRSS2 X-ray crystal structure, in a complex with nafamostat, a potent but nonselective trypsin-like serine protease inhibitor being investigated as a COVID-19 therapeutic. Our TMPRSS2 protein engineering and production strategy may be generally applicable for other TTSPs to enable their selective and rational targeting and uncover the molecular basis of their biological and pathobiological functions.

Results

Production and structure of activatable TMPRSS2 ectodomain

TMPRSS2 is composed of an intracellular domain, a single-pass transmembrane domain and a biologically active ectodomain with three subdomains: a low-density lipoprotein receptor type-A (LDLR-A) domain, a Class A Scavenger Receptor Cysteine-Rich (SRCR) domain and a C-terminal trypsin-like serine peptidase (SP) domain with a canonical Ser441-His296-Asp345 catalytic triad (Fig. 1a)28. As TMPRSS2 is synthesized as a zymogen, it requires cleavage at a conserved Arg255-Ile256 peptide bond within its SRQSR255↓IVGGE activation motif14,28. We achieved this in high yield by replacing SRQSR255↓ with an enteropeptidase (TMPRSS15)-cleavable DDDDK255↓ sequence to prohibit auto-activation during overexpression, allowing purification of a secreted form of the full TMPRSS2 ectodomain zymogen from insect cells, analogous to a strategy used for the TTSP, matriptase (Methods)29. Subsequent proteolytic activation with the addition of recombinant enteropeptidase afforded highly active, homogenous TMPRSS2 to milligram yields and was accordingly named the directed activation strategy TMPRSS2 (dasTMPRSS2, Fig. 1b and Extended Data Fig. 1a). Serendipitous efforts revealed that concentrated dasTMPRSS2 could self-activate without enteropeptidase addition across 6 hours despite this nonnatural sequence replacement, enabling simple size-exclusion purification of the activated species (Extended Data Fig. 1b,c). We determined the X-ray crystal structure of dasTMPRSS2 refined to 1.95 Å resolution after acylation of the catalytic Ser441 residue with nafamostat, a broad-spectrum synthetic serine protease inhibitor (released under Protein Data Bank (PDB) ID 7MEQ). We obtained clear electron density for residues 149–491 spanning the SRCR and SP domains but not residues 109–148 containing the flexible LDLR-A domain responsible for linking the protease to the plasma membrane, likely due to autoproteolytic cleavage of this domain before or during the crystallization process (Extended Data Fig. 1d). The engineered DDDDK255 activation motif was not resolved in the structure but rather terminated in an unstructured loop, consistent with matured TMPRSS13 (PDB 6KD5) and hepsin (PDB 1Z8G) structures containing their native activation motifs30,31. The newly exposed N-terminal Ile256 of the SP domain formed a salt bridge with the side chain of Asp440 (Fig. 1d), confirming full maturation of the activation pocket and taken together with the Cys244-Cys365 interdomain disulfide, confirms that this structure represents a bioactive, stabilized form of the protease (Fig. 1e).

TMPRSS2 has an accommodating substrate binding cleft

The TMPRSS2 SP domain is highly conserved with all TTSPs and conforms to the canonical chymotrypsin/trypsin fold with two six-stranded beta barrels converging to a central active site cleft harboring the catalytic triad (Fig. 2a)32. Divergent protein substrate specificity of these closely related proteases is conferred through highly variable, surface-exposed loops, denoted Loops A–E and Loops 1–3 (Fig. 2a)32. Unique subsites formed on the face of the SP domain, S4-S3-S2-S1-S1’-S2’-S3’-S4’ recognize substrate P4-P3-P2-P1↓P1’-P2’-P3’-P4’ amino acid positions spanning the scissile bond (Fig. 2a,b). To rationally assign these subsites for TMPRSS2, we superposed the peptide-bound hepsin and TMPRSS13 SP domains (40.1 and 41.4% sequence identity of their SP domains, respectively) belonging to the same hepsin/TMPRSS subfamily as TMPRSS2. The S1 position of TMPRSS2 is occupied by the phenylguanidino moiety of nafamostat, forming salt bridges with the highly conserved Asp435, Ser436 and Gly464 residues in the same binding mode as the guanidino of P1 Arg residues observed in hepsin and TMPRSS13 (Figs. 1d and 2b,c and Supplementary Fig. 1a,b). The TMPRSS2 S2 subsite has a distinguishing Lys342 residue that likely confers a preference for small and/or electronegative P2 substrates, similar to the S2 Lys in enteropeptidase that prefers P2 Asp residues33. The S3 and S4 subsites appear open to accommodate various P3 and P4 amino acids and may make favorable receptor contacts with the respective Gln438 and Thr341 positions (Fig. 2b,c). N-terminal to the scissile bond, the buried S1’ site appears to accept small, hydrophobic P1’ residues. Overall, the TMPRSS2 active site appears capable of binding various substrate sequences with the strictest preference for the P1 and P2 positions.

Fig. 2: Molecular recognition of TMPRSS2 substrates is mediated through surface loops and potentially through an unpaired cysteine residue.
figure 2

a, The substrate binding face of TMPRSS2 makes use of three disulfides (yellow sticks) and eight loops, Loops 1–3 (L1–L3) and LA–LE, to confer protein substrate specificity. Nafamostat (gray sticks) covalently bound to the catalytic Ser441 engages the S1 subsite of TMPRSS2 with residues from L1 and L2. b, Multiple sequence alignment of the TTSP family with protease subsites S1’–S4 residues highlighted. The catalytic serine residue acylated by nafamostat is denoted with an asterisk. The S2–S4 protease subsites show greater variability than S1’ and S1 and confer divergent substrate specificity within the TTSP family. c, The subsites of TMPRSS2 (blue) superposed on the corresponding residues of hepsin (magenta, PDB 1Z8G) and TMPRSS13 (orange, PDB 6KD5) with their respective KQLR and decanoyl-RVKR covalent peptide ligands. d,e, A multiple sequence alignment of the human TTSP family members (d) shows complete conservation of the interdomain Cys244-Cys365 disulfide bond but identifies TMPRSS2 as uniquely possessing the only unpaired cysteine residue, Cys379, in the SP domain (e), which is conserved among mammalian TMPRSS2 orthologs. f, The highly conserved SRCR–SP interdomain disulfide is at the backside of the SP domain and is in close proximity to the unpaired Cys379 of TMPRSS2. g, MOE protein patch analysis of the TMPRSS2 ectodomain surface reveals a contiguous 360 Å2 hydrophobic residue patch (green) highlighted with white arrows that is adjacent to the unpaired Cys379 at the backside of the SP domain.

Full size image

Among TTSPs, the SP domain of TMPRSS2 uniquely possesses three disulfides and a single unpaired cysteine residue, Cys379 (Fig. 2d). In all other human TTSPs, this position forms a disulfide bond with an additional Cys at the residue equivalent of Thr447, or both cysteines are absent, resulting in three or four intradomain disulfides within their serine protease domains, respectively. This unpaired cysteine is conserved in mouse, rat, feline, bovine, porcine, equine and chimpanzee TMPRSS2 orthologs (Fig. 2e). Furthermore, the unpaired Cys379 is bordered by an expansive 360 Å2 patch of exposed hydrophobic surface area in our structure that may serve as an interaction hub for TMPRSS2 binding partners (Fig. 2f,g).

The TMPRSS2 stem domain likely orients the SP domain

The SRCR domain is found enriched in proteins expressed at the surface of immune cells as well as in secreted proteins, and are thought to participate in protein–protein interactions and substrate recruitment by TTSPs34. Of the TTSPs, only hepsin and TMPRSS13 structures have been solved containing their complete ectodomains, motivating us to use these as templates for comparison to our structure. The Class A SRCR domain of TMPRSS2 is located on the backside of the SP domain away from the active site and is structurally similar to that of TMPRSS13 despite sharing only 19% sequence identity and both have two intradomain disulfides (Fig. 3a,b). These two SRCR domains adopt a compact, globular fold with similar orientations relative to their SP domains. The SRCR domain of hepsin (7.5% sequence identity) diverges substantially from TMPRSS2/13 with three intradomain disulfides and a tighter SRCR–SP association dominated by complementary electrostatic patches and buried surface area (Fig. 3a,b)31. Hepsin lacks an LDLR-A domain, evidently using the SRCR domain to connect the SP domain to the plasma membrane, potentially explaining its large 78° rotation relative to TMPRSS2/13. When hepsin/TMPRSS2/TMPRSS13 SRCR domains are superposed, they show high degrees of structural similarity owing to the conserved placement of disulfide bonds (Fig. 3b). These overall interdomain conformational differences may therefore play a role in orienting the SP domain relative to the plasma membrane as well as modulating activity through recognition or recruitment of partner proteins.

Fig. 3: The stem domain of TMPRSS2 is structurally similar to TMPRSS13.
figure 3

a, Superposed serine protease domains (blue), and SRCR (green) of TMPRSS13, TMPRSS2 and hepsin. The LDLR domain of TMPRSS13 is shown in magenta with bound calcium in orange and the absence of the LDLR in TMPRSS2 is highlighted in red. b, Ribbon representation of the superimposed SRCR domains of TMPRSS2 (blue), Hepsin (magenta, PDB 1Z8G) and TMPRSS13 (orange, PDB 6KD5). c, Superposed models of TMPRSS13 in cyan and TMPRSS2 in salmon and its symmetry-related molecule in yellow. The LDLR domain that is present in TMPRSS13 structure would clash with the symmetry-related molecule in the TMPRSS2 crystal lattice, suggesting that the LDLR domain is not present in the crystal lattice.

Full size image

We observed no electron density for the LDLR-A domain of TMPRSS2, despite a similar ectodomain construct to that which afforded the TMPRSS13 crystal structure31. However, inspection of our X-ray crystal structure superposed on TMPRSS13 indicates that the LDLR domain that is present in the TMPRSS13 structure would clash with the symmetry-related molecule in the TMPRSS2 crystal lattice, suggesting that our LDLR domain is likely cleaved before or during crystallization (Fig. 3c and Extended Data Fig. 1d).

TMPRSS2 displays robust in vitro peptidase activity

To evaluate TMPRSS2 inhibitors and provide groundwork for future structure–activity relationship studies, we established in vitro proteolytic activity and inhibition assays. The generic TTSP fluorogenic peptide substrate Boc-Gln-Ala-Arg-7-aminomethylcoumarine (AMC) was rapidly cleaved by dasTMPRSS2, C terminal to Arg, thereby releasing AMC product and enabling initial reaction velocities (Vo) measurement within 60 seconds of enzyme addition (Fig. 4a and Methods). In Assay Buffer, dasTMPRSS2 had a Km of (200 ± 80) µM, Vmax of (0.7 ± 0.2) nmol min−1, kcat of (18 ± 4) s−1, kcat/Km of (5.4 ± 0.2) µM−1 min−1 and specific activity at (0.22 ± 0.03) µmol min−1 mg−1 enzyme purified to apparent homogeneity (Fig. 4b). This level of activity is high compared to other described recombinant TMPRSS2 enzyme16,22,35,36, and enzyme activity was unaffected by the presence of Ca2+, NaCl concentrations ranging 75–250 mM, EDTA and tolerant of 2% (v/v) dimethylsulfoxide (DMSO) that is encouraging for use in high-throughput inhibitor screening campaigns for new antiviral discovery (Fig. 4c).

Fig. 4: dasTMPRSS2 displays robust in vitro peptidase activity.
figure 4

a, The generic Boc-Gln-Ala-Arg-AMC fluorogenic peptide substrate is efficiently cleaved by 3.4 nM dasTMPRSS2 in Assay Buffer pH 8.0. Representative progress curves are shown in technical duplicate (n = 2) with datapoints shown as mean values ± s.d. b, Michaelis–Menten plot of initial reaction velocities for kinetic parameter estimation after curve fitting in GraphPad. Each reaction velocity was tabulated in technical quadruplet (n = 4) and datapoints are shown as mean values ± s.d. c, Relative peptidase activity (normalized to Assay Buffer; 25 mM Tris pH 8.0, 75 mM NaCl, 2 mM CaCl2) calculated from initial reaction velocities under Tris pH 8.0 buffering conditions with varying concentrations of NaCl, CaCl2, EDTA and DMSO. Activity was measured in technical triplicate (n = 3) and data are shown as mean values ± s.d. No statistically significant differences from mean peptidase activity in Assay buffer were found (denoted as NS) in the indicated buffering conditions as determined by one-way analysis of variance for n = 3 biologically independent samples examined over one independent experiment.

Source data

Full size image

Nafamostat rapidly acylates TMPRSS2 and slowly hydrolyzes

Nafamostat and camostat are serine protease inhibitors under investigation as anti-TMPRSS2 COVID-19 therapeutics (Clinical Trial.gov identifiers NCT04583592 and NCT04625114). Both are reactive esters that form the same slowly reversible phenylguanidino covalent complex with the catalytic serine residue of trypsin-like serine proteases (Fig. 5a,b), as evidenced in the enteropeptidase–camostat costructure (Fig. 5c). Nafamostat and camostat dramatically increased the apparent melting temperature (TM,a) of dasTMPRSS2 by (25.5 ± 0.1) and (24.8 ± 0.3) °C, respectively, as measured by differential scanning fluorimetry (DSF)37 (Extended Data Fig. 2a) and was a key stabilizing feature to enable protein crystallization (Methods). Nafamostat demonstrated enhanced potency over camostat with half-maximum inhibitory concentration (IC50) values of (1.4 ± 0.2) and (9 ± 4) nM, respectively, with 5 min of preincubation before the assay (Fig. 5d). However, IC50 values were time dependent and required further kinetic interrogation to assess their divergent potencies. Near simultaneous coaddition of these ester inhibitors with substrate allowed us to monitor the real-time acylation and plateau of enzymatic activity, analogous to kinetic investigations of camostat’s acylation of enteropeptidase (Fig. 5e,f)38. Nafamostat was 40-fold more potent than camostat with respective kinact/Ki values of (1.44 ± 0.4) and (0.035 ± 0.002) nM−1 min−1 (Extended Data Fig. 2b). These results emphasize that single timepoint IC50 values are insufficient for ranking mechanism-based, covalent inhibitors of this highly active protease in structure–activity relationship studies. As previously identified for matriptase, the nafamostat leaving group, 6-amidino-2-napthol, fluoresces and can be used as a sensitive burst titrant to calculate the concentration of active protease by quantifying its production (Extended Data Fig. 2c,d and Methods)36. The half-life of the phenylguanidino acyl–enzyme complex was (14.7 ± 0.4) hours as measured by the gradual rescue of dasTMPRSS2 peptidase activity after stoichiometric acylation with nafamostat (Extended Data Fig. 2e,f).

Fig. 5: dasTMPRSS2 peptidase activity is blocked by clinical protease inhibitors.
figure 5

a, Nafamostat and camostat are attacked by the catalytic Ser441 of TMPRSS2, with respective leaving groups 6-amidino-2-napthol and 4-Hydroxy benzeneacetic acid 2-(dimethylamino)-2-oxoethyl ester. b, A common phenylguanidino acyl–enzyme complex is formed after nafamostat or camostat treatment that specifically engages the S1 protease subsite. c, TMPRSS2 (blue) and TMPRSS15 (enteropeptidase, salmon; PDB ID 6ZOV) after nafamostat and camostat treatment, respectively. Catalytic triad residues are highlighted in bold and the conserved S1 subsite residue Asp435 shown. d, Clinical protease inhibitors preincubated for 5 min with dasTMPRSS2 block peptidase activity against Boc-QAR-AMC fluorogenic substrate with varying inhibitory potencies. Data are shown as mean ± s.d. and were performed in technical and biological duplicate (total n = 4) and curve fit for absolute IC50 in GraphPad. e, Reaction progress curves of residual dasTMPRSS2 peptidase activity after 10 s preincubation with the indicated concentrations of nafamostat and camostat inhibitors in the presence of 100 µM Boc-QAR-AMC substrate. Plateaus within progress curves demonstrate time-dependent acylation resulting in complete inhibition of dasTMPRSS2 peptidase activity. Data are shown as mean ± s.d. in technical duplicate (n = 2) and results were consistently obtained across n = 3 biological experiments.

Source data

Full size image

Noncovalent trypsin-like serine protease inhibitors benzamidine and sunflower trypsin inhibitor-1 (SFTI-I) were less potent with respective IC50 values of (120 ± 20) and (0.4 ± 0.2) µM (Fig. 4d), and Ki values of (80 ± 10) and (0.4 ± 0.2) µM (Extended Data Fig. 3a,b). The nafamostat leaving group 6-amidino-2-napthol also competitively disabled dasTMPRSS2 activity with an IC50 of (1.6 ± 0.5) and Ki of (1.1 ± 0.3) µM (Extended Data Fig. 3c).

TMPRSS2 cleaves SARS-CoV-2 S protein at multiple sites

Cells expressing TMPRSS2 have been shown to efficiently cleave the S protein of SARS-CoV-1 at the S1/S2 (SLLR667↓) cleavage site and additional peripheral K/R residues to induce the necessary conformational changes leading to virus–host fusion at the plasma membrane5,39 (Figs. 1a and 6a,b). TMPRSS2 can activate S protein in virus-producing cells (termed cis-cleavage) as well as naïve host target cells activating S protein at the cell surface (trans-cleavage)39. However, the unavailability of validated recombinant TMPRSS2 enzyme has led to challenges in unambiguously mapping TMPRSS2 cleavage sites in SARS-CoV-1 and −2, as the typical workflow with cell-based systems and S protein western blotting produces complex fragment patterns and do not apply unbiased, discovery-based mass spectrometry approaches to localize residue cleavages. Notably, the ‘S2’ site’ (PTKR797↓for SARS-CoV-1 and predicted PSKR815↓for SARS-CoV-2), which was identified as a trypsin cleavage site for SARS-CoV-1 (ref. 5), has been proposed as an important TMPRSS2 cleavage site for SARS-CoV-2 by analogy, but no biochemical evidence is currently available to support cleavage at this residue.

Fig. 6: dasTMPRSS2 efficiently cleaves recombinant SARS-CoV-2 Spike protein constructs at multiple sites.
figure 6

a, The trimeric Spike protein (depicted here in its monomer form) is presented at the surface of the viral membrane and binds the transmembrane human ACE2 receptor through its RBD. Three distinct TMPRSS2 protease cleavage sites are predicted (indicated by scissors). b, Schematic map of known (S1 and S2) and inferred protein fragments after TMPRSS2 treatment, as derived from protein bands (c–g) with the indicated approximate molecular weights (in parentheses) produced as a result of 0–3 protease cleavage events. Recombinant S protein is cleaved at the S1/S2 site, at an unknown site within S2 termed the X/Y cleavage site and within RBD termed the RBD–TMPRSS2 (RT) site that is spanned by a disulfide bond. Fragments containing RBD are highlighted in red, and cut patterns denoted with an asterisk are only detected using S1/S2 KO S protein, HexaPro. c, HexaFurin S protein is converted to S1 and S2 band fragments after 16 h incubation with 10 U of recombinant furin protease (NEB). Mw, molecular weight. d, dasTMPRSS2 fully converts untreated HexaFurin to S1 and S2 fragments within 5 min. e, HexaPro S protein treated with 300 nM TMPRSS2 for 30 min is cleaved at both the X/Y and RBD–TMPRSS2 cleavage sites, with the latter cleavage event apparent only under reducing (R) SDS–PAGE conditions rather than nonreducing (NR). f, Recombinant S protein constructs containing only the RBD and an RBD-Fc fusion are cleaved at the RBD–TMPRSS2 cleavage site after 120 min of TMPRSS2 incubation. g, Western blot analysis of HexaFurin S protein digestions with an RBD-specific primary antibody shows the expected S1 band fragment, the N + RBD fragment and several intermediate sized fragments also containing RBD. All gels and blots are consistent with n = 3 independent biological experiments.

Source data

Full size image

The most significant cleavage site for SARS-CoV-2 infection is the multibasic S1/S2 sequence (RRAR685↓, Fig. 6a,b), which was initially hypothesized to confer preferential processing by intracellular furin protease27, possibly explaining the severe and systemic COVID-19 disease symptoms. This hypothesis was supported by cell infection studies showing that multibasic, peptide-based furin inhibitors prevented S1/S2 cleavage and attenuated viral entry7. However, subsequent studies showed that these inhibitors are promiscuous and disable multiple cell surface-expressed proteases that process multibasic substrates in addition to furin, and more selective furin inhibitors cannot fully abrogate S activation4. Furthermore, furin-deficient cells can still generate S1/S2 cleaved virus8. Propagation of live SARS-CoV-2 virus in TMPRSS2-deficient but not furin-deficient cell lines results in a loss of the multibasic S1/S2 site8, attenuating viral infectivity toward TMPRSS2+ cells. We sought to characterize TMPRSS2’s proteolytic activity toward S1/S2 by incubating recombinant dasTMPRSS2 and/or furin protease with stabilized SARS-CoV-2 S protein with S1/S2 knocked out (RRAR685 → GSAS685; HexaPro construct) or with S1/S2 intact (denoted HexaFurin, Fig. 6b). As expected from previous studies using recombinant, S1/S2 intact S protein, HexaFurin sustained partial S1/S2 cleavage during production in human embryonic kidney 293 cells presumably due to endogenously expressed furin40 (Fig. 6c). Incubation with recombinant furin for 16 hours converted the remaining intact HexaFurin to a product yielding S1 and S2 fragments by SDS–PAGE (Fig. 6c) and was unable to cleave HexaPro due to the S1/S2 KO (Extended Data Fig. 4a).

In contrast, using both the HexaFurin and HexaPro constructs, we observed that dasTMPRSS2 could cleave the S protein at three distinct sites with variable efficiency (Fig. 6d–g). HexaFurin was cleaved to only the S1 and S2 fragments within 5 min of dasTMPRSS2 addition (Fig. 6d), demonstrating the S1/S2 site was best recognized by TMPRSS2 across a minimal incubation. This is the first biochemical evidence that TMPRSS2 recognizes the SARS-CoV-2 S1/S2 site with high efficiency. HexaPro, lacking the S1/S2 cleavage site, was cleaved across 30 min to produce a larger 150 kDa fragment X and 70 kDa fragment Y when analyzed under nonreducing conditions, motivating us to name this putative site as an ‘X/Y’ cleavage site (Fig. 6e). Running the same 30 min digest sample under reducing SDS–PAGE revealed an additional putative cleavage site that is spanned by two cysteine residues participating in a disulfide bond, splitting the observed fragment X into daughter fragments Q at 120 kDa and N + receptor binding domain (RBD) at 35 kDa (Fig. 6e). Notably, S protein band fragments ran at a slightly higher apparent molecular weight under reducing conditions, likely reflecting increased flexibility and reduced gel mobility. We localized this cleavage site hidden by a disulfide bond to near the C-terminal end of the RBD using two S protein constructs containing only RBD. TMPRSS2 treatment produced a common cleavage product (highlighted with an arrow, Fig. 6f), motivating us to name this putative cleavage site the RBD–TMPRSS2 cleavage site.

To visualize all three cleavage sites simultaneously, we treated HexaFurin for 30 min with dasTMPRSS2 and analyzed digests with a western blot using an antibody directed toward the S protein RBD (Fig. 6g). At least seven bands were observed on reducing SDS–PAGE that are consistent with the three proposed cleavage sites, dominated by rapid, initial S1/S2 cleavage (Extended Data Fig. 4b). The western banding patterns observed (S1/S2, X/Y and RBD–TMPRSS2 cleavages) are consistent with cellular studies monitoring SARS-CoV-1 S protein processing by TMPRSS2 that produced many fragments and enabled shedding of the S1 fragment39, which acts as an immune decoy to compromise the ability of neutralizing S protein antibodies.

Stoichiometric amounts of nafamostat completely blocked dasTMPRSS2-mediated HexaPro activation over 2 h (Extended Data Fig. 4c) and are consistent with this drug’s ability to potently block SARS-CoV-2 pseudovirus entry to TMPRSS2 + Calu-3 (ref. 41) and Caco-2 (refs. 1,22) lung and colorectal cancer cell lines. Bromhexine hydrochloride, another agent previously under clinical investigation for anti-TMPRSS2 COVID-19 therapy (NCT04273763)42, showed no inhibition in either the peptidase or HexaPro cleavage assay formats (Extended Data Fig. 4d,e), corroborating reports of its ineffectiveness in blocking SARS-CoV-2 pseudovirus entry43 and further underscores the need for new, selective TMPRSS2 inhibitors.

Discussion

We have produced and characterized a source of TMPRSS2 enzyme that will enable future inhibitor development as antivirals and thorough molecular interrogation of coronavirus and influenza virus activation, owing to TMPRSS2’s critical and widespread role in viral tropism. Although nafamostat potently neutralizes TMPRSS2 activity, it is nonselective and disables trypsin-like serine proteases involved in coagulation such as plasmin, FXa and FXIIa, as well as other TTSPs through its generic arginine-like engagement with the S1 subsite44,45. In line with this promiscuity and nonspecific acylation, nafamostat is biologically unstable with a half-life of 8 minutes (NCT04418128 and NCT04473053). Camostat operates similarly to nafamostat but with poorer TMPRSS2 inhibitory potency and is hydrolyzed to two additional products in vivo, each with further compromised potencies22. These selectivity and consequential stability issues challenge the abilities of these drugs to adequately engage active TMPRSS2 at pharmacological concentrations in vivo, despite their favorable in vitro antiviral potency41,43. Clearly, selective and biologically stable drugs for TMPRSS2 must be further explored and may be achieved through inhibitors engaging the more TMPRSS2-specific S2, S3 and S4 protease subsites identified in our crystal structure.

Our demonstration that TMPRSS2 can efficiently cleave the multibasic S1/S2 site of the S protein supports the notion that instead of conferring furin dependence, the virulent properties of this site on the S protein may derive from recognition and cleavage by airway-expressed TTSPs, which is supported by the demonstrated roles that TMPRSS4 (refs. 6,21), TMPRSS11d (refs. 13,22,23) and TMPRSS13 (refs. 22,23), which colocalize with ACE2 (ref. 6), play in enabling SARS-CoV-2 infection across various tissues.

The X/Y and RBD–TMPRSS2 cleavage sites proposed here may have significant roles in TMPRSS2-mediated SARS-CoV-2 activation and infection of cells. In future studies, we will identify the precise residue at which cleavage occurs, enabling enzymatic and cell-based KO experiments to interrogate their biological function.

Unfortunately, many cellular studies of TTSPs1,22,23,46,47, with some exceptions13,48,49, have not interrogated the zymogen activation status to ensure their activity. This limits what conclusions can be drawn on the biological significance of each protease target in models of viral infectivity, as the (typically transfected) cell lines in use may simply not provide the requisite activators, protein partners or physiological conditions conducive to a matured and active TTSP. This variable can be removed by quantifying the relative abundance of the catalytic chain (under reducing conditions) through SDS–PAGE or western blotting and/or developing activity-based, TTSP-selective chemical probes to accurately dissect the biological activity of each protease.

Our characterization and X-ray crystal structure of dasTMPRSS2 motivates interrogation of a mechanism by which the native, membrane-bound enzyme could be (auto)proteolytically processed peripheral to the activation motif and thereby shed as a soluble enzyme into the extracellular space, analogous to matriptase50. A study using TMPRSS2-specific antibodies has reported detection of a secreted enzyme product in the airways that was proposed to play a functional role in pericellular proteolytic activation14. Due to the disulfide-linked nature of activated TMPRSS2 outlined here (Fig. 1b), former studies may have mischaracterized the catalytic (or noncatalytic depending on the epitope recognized by the antibody) subunit as a shed SP domain when it would instead resolve to the intact species under nonreducing conditions (Fig. 1b). Taken together, a biochemical characterization of these secreted species is required to clarify their activation status and subunit organization, as a shed form of TMPRSS2 localized to the extracellular milieu would have profound pathobiological and therapeutic targeting implications.

Methods

Construct design and cloning

A construct encoding residues 109–492 comprising soluble TMPRSS2 ectodomain was amplified by two PCR fragments (Addgene plasmid no. 53887) and subcloned into the pFHMSP-LIC C donor plasmid by the ligation independent cloning (LIC) method. The final construct contained a N-terminal honeybee melittin signal sequence peptide and C-terminal His8-tag (Fig. 1b). Mutations targeting the activation sequence SSQSR255↓IVGGE (arrow indicates the cleavage site) were implemented to replace the SRQSR255 residues with an enteropeptidase-cleavable DDDDK255 graft with two sets of primer pairs (Supplementary Table 1) generating mutations for S251D/R252D/Q253D/S254D/R255K. Our engineered dasTMPRSS2 protein expression construct is available on Addgene (plasmid no. 176412). Plasmid transfer vector containing the TMPRSS2 gene was transformed into Escherichia coli DH10Bac cells (Thermo Fisher, catalog no. 10361012) to generate recombinant viral Bacmid DNA. Sf9 cells were transfected with Bacmid DNA using JetPrime transfection reagents (PolyPlus Transfection Inc., catalog no. 114-01) according to the manufacturer’s instructions, and recombinant baculovirus particles were obtained and amplified from P1 to P2 viral stocks. Recombinant P2 viruses were used to generate suspension culture of baculovirus infected insect cells for scaled-up production of TMPRSS2.

The SARS-CoV-2 Spike ectodomain HexaPro construct was a gift from J. McLellan51, and the S1/S2 site was restored (GSAS685 → RRAR) through site-directed mutagenesis with primers in Supplementary Table 1 (HexaFurin construct).

Baculovirus mediated dasTMPRSS2 protein production in Sf9 insect cells

Sf9 cells were grown in I-Max Insect Medium (Wisent Biocenter, catalog no. 301-045-LL) to a density of 4 × 106 cells per ml and infected with 20 ml l−1 of suspension culture of baculovirus infected insect cells before incubation on an orbital shaker (145 r.p.m., 26 °C).

dasTMPRSS2 protein purification

Cell culture medium containing the final secreted protein product AA-(TMPRSS2(109–492))-EFVEHHHHHHH was collected by centrifugation (20 min, 10 °C, 6,000g) 4–5 d postinfection when cell viability dropped to 55–60%. Media was adjusted to pH 7.4 by addition of concentrated PBS stock, then supplemented with 15 ml l−1 settled Ni-NTA resin (Qiagen) at a scale of 12 l. Three batch Ni-NTA purifications were used to capture protein, with each round requiring shaking in 2 l flasks for 2 h at 16 °C (110 r.p.m.), collection by centrifugation (5 min 1,000g) and then transferring to a gravity flow column. Beads were washed with 3 column volumes of ice-cold PBS before elution with PBS supplemented with 500 mM imidazole. Elution samples were concentrated to 4.5 mg ml−1 using 30 kDa molecular weight cutoff (MWCO) Amicon filters and zymogen activation was achieved by dialyzing protein 1:1,000 against Assay Buffer (25 mM Tris pH 8.0, 75 mM NaCl, 2 mM CaCl2) at room temperature across 6 h (Extended Data Fig. 1a). Activated samples were exchanged to size-exclusion chromatography buffer (50 mM Tris pH 7.5, 250 mM NaCl), spun down at 17,000g, then loaded to a Superdex 75 gel filtration column. Fractions spanning the dominant peak eluting at 80 ml (Extended Data Fig. 1b,c) were evaluated for appropriate banding on reducing SDS–PAGE before pooling and concentrating. For dasTMPRSS2 enzyme samples, 2 µl aliquots of 10,000× enzyme assay stocks (34 µM) were prepared by concentrating protein to 1.6 mg ml−1 in Storage Buffer (50 mM Tris pH 7.5, 250 mM NaCl, 25% glycerol), then flash-frozen in liquid nitrogen and stored at −80 °C until thawed immediately before use for each enzyme assay to minimize autoproteolysis and maintain reproducible enzyme concentrations.

HexaPro and HexaFurin production and purification

Expi293F cells (Life Technologies catalog no. A1435102) were transiently transfected with expression plasmid encoding HexaPro/Furin using FectoPro transfection reagent (Polyplus transfection SA, catalog no. 116-010) with 5 mM sodium butyrate being added at the time of transfection (Sigma, catalog no. 303410). After 4–5 days posttransfection time, the cell culture was collected, supernatant cleared by centrifugation and the pH was adjusted by adding 10× Buffer (50 mM Tris pH 8.0, 150 mM NaCl). Secreted protein was captured by two rounds of batch absorption with 4 ml l−1 of pre-equilibrated Ni Sepharose beads (GE Healthcare, catalog no. 17-5318-01). The bound beads were transferred to gravity flow column and sequentially washed with 30 column volumes of Wash Buffer (50 mM HEPES 7.5, 300 mM NaCl, 5% glycerol), followed by Wash Buffer supplemented with 25 mM imidazole. Protein was eluted in Elution Buffer (Wash Buffer with 250 mM imidazole) and concentrated using Amicon Ultra Centrifugal Filter Units, 15 ml, 100 kDa (Millipore Sigma catalog no. UFC910024) before size-exclusion chromatography purification using a Superose 6 Increase 10/300 GL (GE Healthcare catalog no. 29-0915-96), in a buffer composed of 20 mM HEPES pH 7.5 and 200 mM NaCl.

Protein crystallization and data collection

After size-exclusion purification of activated dasTMPRSS2, samples were pooled and concentrated to 2 mg ml−1. Protein was treated with 3:1 nafamostat:dasTMPRSS2 for 10 min at room temperature and exchanged into Assay Buffer supplemented with 3:1 nafamostat using four spin cycles in 30 kDa Amicon MWCO filters (14,000 r.p.m., 15 min, 4 °C) to remove low molecular weight autolytic fragments from the 42 kDa enzyme. Acylated enzyme was then concentrated to 8 mg ml−1 and centrifuged (14,000 r.p.m., 10 min, 4 °C) before automated screening at 18 °C in 96-well Intelliplates (Art Robin) using the Phoenix protein crystallization dispenser (Art Robbins). Protein was dispensed as 0.3 µl of sitting drops and mixed 1:1 with precipitant. The RedWing and SGC precipitant screens were tested and amorphous, nondiffracting crystals were consistently produced when grown over 30% Jeffamine ED-2001 (Hampton Research) with 100 mM HEPES pH 7.0. To acquire a diffraction quality crystal, acylated dasTMPRSS2 was treated with 50 U of PNGase F (NEB, 37 °C for 45 min) to trim N-glycan branches, then centrifuged (14,000 r.p.m., 4 °C, 10 min) before setting 2 µl of hanging drops with 1:1 protein:precipitant and grown for 10 days. Crystals were then cryo-protected using reservoir solution supplemented with roughly 5% (v/v) ethylene glycol and cryo-cooled in liquid nitrogen. X-ray diffraction data were collected on the beamline 24-ID-E at the Advanced Photon Source.

Solving the dasTMPRSS2:nafamostat crystal structure

X-ray diffraction data were processed with XDS52. Initial phases were obtained by molecular replacement in Phaser MR53, using (PDB 1Z8G) as a starting model. Model building was performed in COOT54 and refined with Buster55. Structure validation was performed in Molprobity56. Data collection and refinement statistics are summarized in Supplementary Table 2, and a phenylguanidino ligand electron density omit map as well as the detailed TMPRSS2 residue interactions are available in Supplementary Fig. 1a,b.

Gel electrophoresis and western blotting

SDS–PAGE was carried out with 15 µl of Mini-Protean (BioRad) or 60 µl of Novex Wedgewell (Invitrogen) 4–20% Tris-Glycine gels for 30 min under constant voltage at 200 V. Protein samples were mixed with 4× Laemelli buffer (BioRad) and subjected to differential reducing (±5 mM β-mercaptoethanol, Gibco), then boiling at 95 °C for 5 min to probe the covalent nature of protein complexes and subunits. The Precision Plus Protein marker (BioRad) was used as a standard.

For SARS-CoV-2 RBD western blotting, SDS–PAGE was carried out as described, followed by wet transfer in Transfer Buffer (25 mM Tris pH 8.3, 192 mM glycine, 20% MeOH (v/v)) to a polyvinyldifluoride membrane (80 V, 53 min, 4 °C). Membranes were incubated in Blocking Buffer (5% skim milk in tris-buffered saline with Tween (TBST)) for 1 h at room temperature, washed 5× with TBST, then probed overnight with 1/3,000 mouse anti-RBD primary mAb (Abcam ab277628) solution at 4 °C. Membranes were then washed 5× with TBST and probed with 1/5,000 FITC-labeled goat anti-mouse IgG secondary pAb (Abcam ab6785) and imaged for fluorescence on the Typhoon FLA7000 biomolecular imager (GE healthcare).

Enzyme peptidase and inhibition assays

Peptidase assays with fluorogenic Boc-Gln-Ala-Arg-AMC substrate (Bachem catalog no. 4017019.0025) were performed in 96-well plates (Greiner Fluotrak) at 200 µl of reaction volumes in a FlexStation microplate reader (Molecular Devices) using the SoftMax Pro software (v.5.4.6) at 24 °C. Fluorescence was monitored with the fastest kinetic read settings across 5 min at 341:441 nm excitation:emission and converted to a product AMC concentration using standard curves at each substrate concentration to correct for the inner-filter effect. All inhibition assays contained 2% (v/v) DMSO and initial reaction velocities were tabulated over the linear portion of the first 60 s of progress curves.

To determine Michaelis–Menten kinetic parameters, 50 µl of 4× enzyme stock (12.8 nM) was added through automated addition 1:3 to microplates containing 150 µl of substrate (0.5–1000 µM) in triplicate and initial reaction velocities were plotted against substrate concentration and curve fit using GraphPad Prism (v.9.1.1).

IC50 potencies of nafamostat mesylate (MedChemExpress catalog no. HY-B0190A), camostat mesylate (MedChemExpress catalog no. HY-13512), benzamidine HCl (Sigma catalog no. 434760-25G), bromhexine HCl (SelleckChem catalog no. S2060) and SFTI-1 (Methods) were initially determined by preincubating dasTMPRSS2 with inhibitor at concentrations ranging from 0.1 nM–100 µM for 5 min, before enzyme–inhibitor mixes were added to substrate through automated addition. Then, seven inhibitor concentrations spanning three orders of magnitude across the IC50 value were used and inhibitor reaction velocities were normalized to uninhibited enzyme and plotted as one-site dose response curves in GraphPad. The apparent Ki (\(K_{{{\mathrm{i}}}}^ \ast\)) of classical competitive trypsin-like serine protease inhibitors benzamidine and SFTI-1 were determined using equation (1),

$$K_{{{\mathrm{i}}}}^ \ast \approx \frac{{{{{\mathrm{IC}}}}_{50}}}{{1 + \frac{{[{{{\mathrm{S}}}}]}}{{K_{{{\mathrm{M}}}}}}}}$$
(1)

where [S] is the concentration of substrate Boc-QAR-AMC used in the assay and KM is the Michaelis constant (200 µM).

Time-dependent IC50 measurement and k inact/K i determination

Camostat IC50 curves were generated using seven concentrations of inhibitor in the range of 0.1–1,000 nM inhibitor and nafamostat between 0.01 and 100 nM with a DMSO control as described. The time dependence of inhibitor potencies was measured by using Flexstation Flex kinetic reads that automatically transferred dasTMPRSS2-inhibitor mixes to substrate wells at the indicated preincubation timepoints (Extended Data Fig. 2b). For the 10 s timepoint, a kinetic read was performed after manual addition of enzyme, followed by substrate, using a multichannel pipette to capture the fast acylation (Fig. 5e). Kinetic parameters \(K_{{{\mathrm{i}}}}^ \ast\) and kinact were determined with the simplified equations (2) and (3), respectively, assuming a one-step kinetic inhibition mechanism

$${{{\mathrm{E}}}} + {{{\mathrm{I}}}}\mathop { \to }\limits^{k_1} {{{\mathrm{EI}}}}$$
(2)
$$K_{{{\mathrm{i}}}}^ \ast \approx \frac{{t_{50}^{(2)} - t_{50}^{(1)}}}{{\frac{{t_{50}^{(2)}}}{{I_{50}^{(1)}}} - \frac{{t_{50}^{(1)}}}{{I_{50}^{(2)}}}}}$$
(3)
$$k_{{{{\mathrm{inact}}}}} \approx \frac{1}{{t_{50}^{\left( 2 \right)}}}{{{\mathrm{exp}}}}\left[ {{{{\mathrm{ln}}}}\left( {\frac{{K_{{{\mathrm{i}}}}^ \ast }}{{I_{50}^{(2)}}} - 1} \right) + b} \right],$$
(4)

where the conversion factor b = 0.558 is applied for concentrations in µM and time in s.

Active site quantification of dasTMPRSS2

The acylation of dasTMPRSS2 by nafamostat and concomitant production of the fluorogenic 6-amidino-2-napthol leaving group was measured by incubating serial dasTMPRSS2 enzyme dilutions from 6.4 to 0.8 nM with excess (10 µM) nafamostat, similar to previous efforts with matriptase36 (Extended Data Fig. 2c,d). Microplate reading at 320 and 490 nm excitation and emission, respectively, were used to calculate the number of dasTMPRSS2 active site residues and calibrate peptidase activity and inhibition assays at 3.4 nM enzyme.

Nafamostat inhibition half-life

The half-life of the phenylguanidino acyl–enzyme complex after nafamostat treatment was measured for dasTMPRSS2 using methods established for camostat with enteropeptidase38. Briefly, dasTMPRSS2 (3.4 µM) was mixed with slight excess nafamostat (5 µM or DMSO control) and incubated at room temperature for 20 min. After incubation, unbound nafamostat was removed by passage and three washes in a 3 kDa MWCO Amicon filter centrifuged at maximum speed. Acylated or untreated dasTMPRSS2 samples were then transferred in quadruplet to a microplate containing either 125 or 250 µM Boc-QAR-AMC substrate (final concentration of 3.4 nM enzyme). Fluorescent reads were carried out immediately, analogous to IC50 assays, but across a period of 8 h to monitor for rescued dasTMPRSS2 activity after deacylation. The acylated traces were fit to a one-phase association exponential in GraphPad to derive the half-life for activity recovery, normalized to the uninhibited initial reaction velocity.

SFTI-1 synthesis and purification

Reagents and solvents were purchased from commercial sources and used without further purification, unless otherwise stated. High-performance liquid chromatography (HPLC) was performed on (1) an Agilent 1260 infinity system equipped with a model no. 1200 quaternary pump and a model no. 1200 ultraviolet absorbance detector or (2) an Agilent 1260 Infinity II Preparative System equipped with a model 1260 Infinity II preparative binary pump, a model 1260 Infinity variable wavelength detector (set at 220 nm) and a 1290 Infinity II preparative open-bed fraction collector. The HPLC column used for analysis was a Phenomenex Luna C18 semipreparative column (5 μm, 45,250 × 10 mm). The HPLC column used for synthesis was a preparative column (Gemini, NX-C18, 46 5 μm, 110 Å, 50 × 30 mm) purchased from Phenomenex. Mass analyses were performed using a Waters 2695 Separation module and Waters-Micromass ZQ mass spectrometer system.

Peptides were synthesized on a Liberty Blue automated microwave peptide synthesis (CEM Corporations). Fmoc-Gly-OH was loaded on a 2-chlorotrityl resin (AdvancedChemTech) in CH2Cl2 and 2,4,6-trimethylpyridine at room temperature. The resin was capped with a solution of 10/5/85 MeOH/N,N-diisopropylethylamine (DIEA)/CH2Cl2 for 10 min. Fmoc groups were removed using 20% piperidine in dimethylformamide (DMF) at 50 °C for 10 min. Using HATU/DIEA chemistry (5 eq. each in DMF) at 50 °C for 10 min per cycle, Fmoc-Asp(OBno)-OH, Fmoc-Pro-OH, Fmoc-Phe-OH, Fmoc-Cys(ACM)-OH, Fmoc-Ile-OH, Fmoc-Pro-OH, Fmoc-Pro-OH, Fmoc-Ile-OH, Fmoc-Ser(tBu)-OH, Fmoc-Lys(Boc)-OH, Fmoc-Thr(tBu)-OH, Fmoc-Cys(ACM)-OH and Fmoc-Arg(Pbf)-OH were loaded sequentially. Fmoc-Arg(Pbf)-OH and Fmoc-amino acids coupled after Pro were coupled using two cycles. The peptide was cleaved from the resin using 20% HFIP in CH2Cl2 with no subsequent purification. The peptide was cyclized in solution using 3 eq. of PyAOP reagent and 6 eq. of DIEA in DMF using microwave heating at a concentration of 1 mg ml−1 at 90 °C for 15 min. The reaction was concentrated in vacuo. Next, the cyclized peptide was treated with 95/2.5/2.5 TFA/TIS/H2O and stirred for 3 h. The crude peptide mixture was concentrated and added to a solution of cold diethyl ether. The suspension was centrifuged at 2,500 r.p.m. for 7 min, the supernatant diethyl ether was discarded, and the solids were diluted into water, frozen and lyophilized to yield a white powder. The reaction was purified using preparative HPLC, eluted with 16–36% acetonitrile in water with 0.1% TFA over 20 min at a flow rate of 30 ml min−1. The retention time was 11.23 min and the collected fractions were frozen and lyophilized. Next, the collected white powder was dissolved in 300 μl of 50% AcOH in H2O and 30 μl of 1 M HCl. Next, 10 μl of 0.1 M I2 in AcOH was added to the reaction mixture and stirred for 5 h at room temperature, while being monitored using semipreparative HPLC. After confirmation of starting material consumption, the reaction was added to a 5 ml solution of 0.1 M NaOAc and purified using preparative HPLC, eluted with 21–41% acetonitrile in water with 0.1% TFA over 20 min at a flow rate of 30 ml min−1. The retention time was 10.13 min and the collected fractions were frozen and lyophilized to yield a white powder. Electrospray ionization–mass spectrometry calculated [M+2H]2+ for C67H106N18O18S2 757.37 and found 757.94.

DSF

Apparent melting temperature (TM,a) shifts were measured for various dasTMPRSS2-inhibitor coincubations using SYPRO Orange dye (Life Technologies, catalog no. S-6650) and monitoring fluorescence at 470 and 510 nm excitation and emission, respectively, using the Light Cycler 480 II (Roche Applied Science). Samples were prepared in quadruplet in 384-well plates (Axygen; catalog nos. PCR-384-C; UC500) at a final volume of 20 µl containing 0.05 mg ml−1 dasTMPRSS2, 1 µM compound or vehicle control and 5× SYPRO Orange. Thermal melt curves were generated between 25 and 95 °C at a gradient of 1 °C min−1 and plots prepared with the online DSFworld application57 for TM,a determination.

SARS-CoV-2 S protein activation and inhibition

Recombinant SARS-CoV-2 S protein constructs HexaFurin and HexaPro were concentrated to 0.5 mg ml−1 in Assay Buffer and incubated with the indicated concentrations of furin protease (NEB) or dasTMPRSS2. Digestions took place over 16 h for furin and from 5–120 min for dasTMPRSS2. Furin digestions were terminated by the addition of 4 mM EDTA whereas dasTMPRSS2 digestions were terminated with 5 µM nafamostat (as TMPRSS2 activity is insensitive to EDTA, Fig. 4c), then SDS–PAGE samples were immediately prepared with the addition of 4× SDS–PAGE loading buffer and boiled for 5 min at 95 °C. Next, 4 µg of S protein were loaded per well under each condition and gels were visualized by Coomassie blue staining. For anti-RBD western blotting, 2 µg of S protein were loaded per well.

For cleavage inhibition assays (Extended Data Fig. 4b,d), dasTMPRSS2 diluted to 320 nM in Assay Buffer was preincubated 15 min with inhibitor (final 1% DMSO (v/v)) or DMSO control, then assays were started by transfer of enzyme–inhibitor mixes to S protein. S protein–dasTMPRSS2 mixtures were incubated at room temperature for 2 h with nafamostat and 30 min for bromhexine.

Multiple sequence alignments

Multiple sequence alignments were prepared to compare the human TTSP family members as well as TMPRSS2 mammalian orthologs. Human TTSP FASTA sequences (isoform 1) were accessed from UniProt and mammalian TMPRSS2 orthologs identified with UniProt BLAST. Sequences were aligned with Clustal Omega58 and annotated with ESPript v.3.0 (ref. 59).

Protein visualization and property calculation

The structure of dasTMPRSS2 was inspected and compared to other TTSPs using PyMol v.2.4.1 (Schrodinger) and the Molecular Operating Environment (MOE) (Chemical Computing Group) 2019 software suite. The exposed hydrophobic patches of TMPRSS2 were calculated using the MOE Protein Patch Analyzer tool60.

Data exclusions and statistics

The two-sided Grubbs’ test was used to determine and exclude single outliers present in sample data performed in triplicate or quadruplet. Single datapoint outliers were identified in Extended Data Fig. 2b (triplicate samples, 5 and 10 nM nafamostat), Extended Data Fig. 2c (triplicate samples; 0 nM 6-amidino-2-napthol) Extended Data Fig. 2d (triplicate samples; twofold enzyme concentration) and Extended Data Fig. 2e (quadruplet samples; nafamostat 125 µM substrate + nafamostat; 250 µM substrate + DMSO). Within the supplied source data documents, these exclusions are denoted with blue text and asterisks.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.