How SARS-CoV-2 Spike Protein Affects Immune Response

Canonical TGF-β cellular signaling is driven by SMAD transcription factors. R-SMADs (SMAD2 and SMAD3) are phosphorylated at their C-termini by the TGF-β-receptor complex (TβRI and TβRII), leading to the formation of trimeric complexes with SMAD4 (Co-SMAD) that enter the nucleus and modulate gene expression (44, 45). To examine TGF-β signaling in SARS-CoV-2 replication, we assayed infected human lung adenocarcinoma/epithelial cells (Calu-3). Calu-3 cells were infected with SARS-CoV-2 (Ancestral Strain) at an MOI of 0.1 and lysed at 72 hpi. SARS-CoV-2 infection led to an increase in SMAD3 phosphorylation, which was inhibited by the TβRI inhibitor SB-431542 (Fig. 1A). Previous work has shown elevated PAI-1 transcripts (SERPINE1) in response to SARS-CoV-2 (33, 39, 41, 46). Consistent with activation of TGF-β signaling by SARS-CoV-2, PAI-1 was induced during infection, and this induction was inhibited by SB-431542 (Fig. 1B).

We next sought to determine the role for S protein in the activation of TGF-β signaling and PAI-1 induction as S protein is sufficient to trigger TGF-β induction (33, 52). We treated cells with recombinantly expressed SARS-CoV-2 S protein ectodomain (S protein; residues 1–1,208 derived from Wuhan strain D614G, HexaPro variant) or the SARS-CoV-2 S protein receptor-binding domain (RBD; residues 319–541) and examined TGF-β expression by ELISA (53, 54). HaCaT cells were selected as they represent a well-established model for assaying TGF-β signaling, and these cells natively express ACE2 (55–58). ACE2-expressing HEK293T (HEK293T^+ACE2) cells and non-ACE2-expressing HEK293T (HEK293T^WT) cells were included in these experiments to determine the requirement for ACE2 (59). TGF-β was induced following S protein or RBD treatment and was dependent on ACE2 expression, indicating that the activity of S protein maps to the RBD (Fig. 2A).

In support of this, SARS-CoV-2 S protein delivered to cells by pseudotyped replication-defective lentivirus (Lenti^S-Anc) led to TGF-β expression (Fig. 2A). S protein led to activation of the SMAD3 (CAGA₁₂) luciferase reporter assay, indicating an optimal concentration of 105 ng/mL consistent with activation of TGF-β signaling (Fig. 2B, P < 0.01). These results seemed somewhat surprising since the concentration of S protein required for pathway activity was 10-fold lower than what had previously been shown to activate TGF-β (33). Additionally, S protein was sufficient to induce SMAD3 phosphorylation, which was ACE2-dependent and inhibited by SB-431542 (Fig. 2C). The same novel concentration of S protein (105 ng/mL) also provided a significant increase in SMAD3 phosphorylation, where increasing the dosage increased this activity (Fig. 2D, P < 0.05). SMAD3 phosphorylation also occurred during Lenti^S-Anc transduction. This activity induced by Lenti^S-Anc peaked at 24 hpi (Fig. S1A and B).

Having established that S protein promotes R-SMAD3 phosphorylation and activity, we expanded our analysis to include additional SMAD luciferase reporters. Both full-length S protein and the RBD stimulated SMAD2/4 (ARE-Luc), SMAD3 (SBR₆-Luc), and SMAD3/4 (CAGA₁₂-Luc) reporter assays and were dependent on ACE2 (Fig. 3A and B; Fig. S2). While S protein modestly increased SBR₆-Luc activity in ACE2-expressing cells, the difference compared to cells lacking ACE2 was not statistically significant (Fig. S2). Activation of ARE-Luc and CAGA₁₂-Luc reporter assays requires the activity of TβRI, which was inhibited by SB-431542 (Fig. 3C and D).

TGF-β is sufficient to induce PAI-1 expression assayed by immunofluorescence and immunoblot assay in HaCaT^WT cells, and this activity is replicated by S protein and RBD (Fig. 4A and B). PAI-1 induction by S protein and RBD was ACE2-dependent and inhibited by SB-431542, indicating the role of the TGF-β pathway (Fig. 4C). We observed that ACE2-expressing cells treated with Lenti^S-Anc, but not the VSV.G control (Lenti^VSV.G), significantly induced PAI-1 expression. Consistent with our previous findings, induction of PAI-1 was inhibited by the TβR1 inhibitor SB-431542 (Fig. 4D and E).

We next addressed whether the activity of recombinantly expressed S protein was replicated in our S pseudotyped virus model. We observed that ACE2-expressing cells treated with Lenti^S-Anc induced PAI-1 expression (Fig. 4D and E). Consistent with our previous findings, induction of PAI-1 was inhibited by the TβR1 inhibitor SB-431542. Additionally, the control VSV.G pseudotyped lentivirus (Lenti^VSV.G) was unable to induce PAI-1 potently (Fig. 4D and E).

To define the pathway by which S protein induces PAI-1, we generated loss-of-function cell lines for SMAD2 and SMAD3 through CRISPR/Cas9 genome editing (Fig. 5A and B). These cells were specifically deficient in SMAD2 (HaCaT^SMAD2KO) and SMAD3 (HaCaT^SMAD3KO) expression and signaling (Fig. S4A through D). ARE-Luc and CAGA₁₂-Luc are specific reporters and binding motifs for SMAD2/4 complexes and SMAD3/4 complexes, respectively (60–62). Loss of SMAD2 and SMAD3 specifically disrupted induction by TGF-β and S protein of the corresponding reporter assay (Fig. 5A through C). Induction of PAI-1 was strongly attenuated in HaCaT^SMAD3KO cells exposed to TGF-β, S, or RBD protein, but was unperturbed in HaCaT^SMAD2KO cells (Fig. 5D; Fig. S4F through H). These results are supported by SMAD3-specific inhibitor SIS3 blocking both pSMAD3 activity and PAI-1 induction by S protein (Fig. 5D; Fig. S5). Our data are consistent with previous findings of PAI-1 being downstream of SMAD3 following TGF-β stimulation.

Consistent with the findings of Biering et al., we have shown that S protein is sufficient to activate TGF-β signaling (33). Biering et al. demonstrated that an RGD peptide phenocopied the activity of S protein in triggering endothelial barrier dysfunction and downregulation of sialic acid expression (33). RGD motifs are ligands for a subset of integrin complexes and can initiate subsequent cellular signaling and remodeling (63). Signaling by RGD to integrins is antagonized by the ATN-161 small-molecule inhibitor. S protein contains an RGD motif in the S1 fragment (residues 403–405) and is highly conserved in early SARS-CoV-2 lineages until the Omicron BA.2 variant. Although Biering et al. implicate the S protein RGD motif in barrier dysfunction, the effects of mutating the S protein RGD motif on TGF-β signaling have not been examined.

We observed that ATN-161 potently suppressed induction of SMAD2 and SMAD3 luciferase reporter assays and SMAD3 phosphorylation in response to recombinant S protein and RBD and S protein pseudotyped lentivirus (Fig. 6A, B and D). To determine if the effects of ATN-161 were mediated by the S protein RGD motif and not RGD motifs present in ACE2 and extracellular matrix proteins, we generated an S protein pseudotyped lentivirus carrying the D405N substitution (Lenti^S-D405N), mutating this motif from RGD to RGN. This mutation was chosen to replicate the residue changes found in the Omicron BA.2 VOC where this RBD site remained structurally functional with a similar binding energy (S-Anc.: −1,780 kcal/mol and S-D405N: −1,794 kcal/mol) but saw a critical acidic residue alteration, which may alter integrin binding capacity (S-Anc.: pH: 6.69 and S-D405N: pH: 7.69) (21). Lenti^S-D405N was unable to induce SMAD3 phosphorylation mapping the activity of S protein to this motif (Fig. 6C through E). We generated an S protein pseudotyped virus in which the RGD motif was replaced by three alanine residues (Lenti^S-AAA), which has been previously shown not to disrupt ACE2 binding or S protein folding (21). Similarly to Lenti^S-D405N, Lenti^S-AAA was unable to induce SMAD3 phosphorylation (Fig. 6D through F). RT-qPCR analysis of SERPINE1 (PAI-1 transcripts) confirmed PAI-1 induction was dependent on an intact RGD motif and was inhibited by ATN-161, SB-431542 (Fig. 6F and G), and SIS3 (Fig. S5E).

TGF-β is intricately associated with both innate and adaptive immune responses through the regulation of type I interferons by STAT1 and IRF3 signaling cascades (64, 65). To determine if TGF-β signaling induced by the RGD motif exerted an effect on type I interferons, we evaluated IFN-β mRNA expression by RT-qPCR following stimulation with the synthetic dsRNA analog Poly(I:C) (66, 67). Poly(I:C) potently induces IFN-β gene expression, which is suppressed by TGF-β (Fig. 7A). Lenti^S-Anc is able to suppress IFN-β induction in a similar manner, and this activity was sensitive to SB-431542, implicating the TGF-β signaling cascade and TβR1 activity. Moreover, sensitivity to ATN-161 and the absence of suppression with lenti^S-D405N highlight the importance of the integrin-binding motif RGD (Fig. 7A). We subsequently confirmed this observed IFN-β suppression by SARS-CoV-2 S protein reliant on SMAD3, where in the presence of the SIS3 inhibitor, SMAD3 expression would decrease (Fig. 7B). These results raise the possibility that S protein presenting to cells during SARS-CoV-2 infection might act to suppress local antiviral immunity prior to entry.

As entry of SARS-CoV-2 downregulates the expression of ACE2, we tested antiviral signaling by the S protein RGD motif with a virus model that does not undergo ACE2-dependent entry (3). We selected Vaccinia virus (Western Reserve strain, WR-VACV) as it has previously been shown to be sensitive to type I interferons in vivo and in viral plaque formation (68–70) (Fig. 7C). When treating the cells with TGF-β prior to WR-VACV infection, we observed a significant 1.5-fold increase in the size of the plaques when compared to the untreated control (Fig. 7D, Fig. S6). Treatment with recombinant RBD led to a similar 1.3-fold increase in the size of WR-VACV plaques at either 4 dpi (Fig. 7C) or 2 dpi (Fig. 7D) from the untreated control. If S and RBD protein were acting to promote WR-VACV spread by TGF-β suppression of interferons, we would expect the proviral activity to be sensitive to inhibitors of the TGF-β pathway and RGD antagonists. Cotreatment with either ATN-161 or SB-431542 abolished the effect of RBD, with plaque sizes not significantly different from those of untreated controls at 2 dpi (Fig. 7D). These data are consistent with the S protein RGD motif possessing proviral activity by TGF-β induction suppressing local cellular immunity.

We next examined the requirement for TGF-β signaling in SARS-CoV-2-induced cytopathology in lung epithelial cells. Treatment of Calu-3 cells with SB-431542 rescued cell viability in SARS-CoV-2 (A.2.2)-infected cells (Fig. 7E, P < 0.001) and shifted the EC₅₀ by 4.8-fold. Inhibition of SMAD3 using SIS3 trended in a similar direction, increasing EC₅₀ by 4.5-fold (Fig. S7A). Collectively, these data identify TGF-β signaling as proviral during SARS-CoV-2 infection.

The infection of the air-liquid interface human primary bronchial epithelial cell (ALI-HPEC) model offers physiological relevance as it encompasses the differentiation of multiple lung epithelial cell types. By analyzing the previously discussed ALI-HPEC data set by Wang et al., we investigated whether TGF-β signaling is induced in a cell-autonomous manner during infection (71). Bulk RNA-seq data from multiple ALI-PBEC cultures and across three time points (24 h, 48 h, and 72 h) revealed that basal cells, which represent a low proportion of SARS-CoV-2-infected cells (Fig. S7C and D), exhibited strong induction of TGF-β1 expression and a TGF-β signature (72) (Fig. 7F, P < 0.0001,Fig. S7B). We identified two subsets of integrins: (i) RGD-binding integrins (broad) and (ii) RGD-binding and ATN-161-sensitive integrins (narrow), given our results demonstrating the sensitivity of S protein induction of TGF-β to ATN-161 (Fig. S7B) (↗21, 22, 73, 74). We also included αvβ1 in the narrow subset despite the lack of experimental evidence for sensitivity to ATN-161 as it has been shown to bind S protein in an RGD-dependent manner (Fig. S7B) (21, 22, 73, 74). In a SARS-CoV-2-infected ALI-PBEC culture, TGF-β1 expression correlated significantly with narrow RGD-binding integrins (Fig. 7G, P = 0.01), while a TGF-β1 signature correlated strongly with both broad and narrow RGD-binding integrins (Fig. S7F; P =0.001, Fig. 7H, P = 0.003).

Human keratinocytes (HaCaT^WT; RRID: CVCL_0038) and Human Embryonic Kidney cells (HEK293T^WT, ATCC CRL-3216) were maintained in Dulbecco’s modified Eagle’s medium (DMEM; Life Technologies). All media were supplemented with 10% fetal bovine serum (FBS), 292 µm/mL L-glutamine, 100 units/mL penicillin, and 100 unit/mL streptomycin (DMEM-FPSG) and incubated at 37°C and 5% CO₂. Human epithelial lung adenocarcinoma cells (Calu-3; ATCC HTB-55) were used for authentic SARS-CoV-2 (A.2.2 variant) live viral infection. HEK293T^+ACE2 were sourced from Stuart Turville.

gRNAs for SMAD2 and SMAD3 were designed using the CRISPR design tool (101) along with online gRNA design tools, Benchling and eCRISPR, for paired gRNA nickase knockouts. gRNAs were cloned into the nickase mutant pSpCas9n(BB)−2A-GFP vector (PX461, Addgene #48140), and integration of the insert was verified via Sanger sequencing (AGRF, Westmead). HaCaT^WT cells were transfected with the SMAD2 and SMAD3 gRNA integrated plasmids and sorted for GFP expression by FACS (BD FACSAria IIu).

Lenti^S-Anc and Lenti^VSV.G used in Fig. 2, 4 and 6 were constructed as previously described in (102, 103). Lenti^S-Anc, Lenti,^S-D405N and Lenti^S-AAA were constructed separately with the same protocol used in Fig. 6 and 7. HEK293T cells were grown to 75% confluency (1.75 × 10⁷ cells in a T75 flask) prior to transfection. Pseudotyped virus particles were generated by co-transfecting a S protein expression construct (0.36 µg/mL; with either: pCG1-SARS-2 ∆18, pLVX-S.Anc, pLVX-S.D405N, and pLVX-S.AAA) or a VSV.G control plasmid (pHCMV-G), a GFP-luciferase vector (0.72 µg/mL; with either pCDH-GFP or pBCKS-GFP) and with either a single lentivirus packaging plasmid (0.54 µg/mL; ps.PAX2, Fig. 6 and 7) or a three-plasmid tat, gagpol, and rev system (pcDNA.31tat101mL, pHCMVrevmlwhvpre, and pHCMgagpolmlstwhv, Fig. 2, 4 and 6) into HEK293T^WT cells, using OptiMEM (Gibco) low serum media supplemented with either Fugene HD (Promega) or Lipofectamine 2000 (Invitrogen, Thermo Fisher Scientific). Following overnight incubation, low serum media was replaced with DMEM supplemented with 10% FBS and 1% PSG. Forty-eight hours post-transfection, the supernatant was removed and centrifuged at 200 x rcf for 5 min to pellet cell debris and additionally passed through a 0.45 µm syringe filter. The viral filtrate was concentrated via the addition of one volume of Lenti-X (Takara Bio USA, Inc) to three volumes of the filtrate and stored at 4°C overnight. The final mixtures were spun down at 3,200 x rcf for 30 min. Viral pellets were resuspended in 1.5 mL SFM.

HEK293T^+ACE2 cells were plated at 70% confluency in 96-well plates. A 1:2 dilution series of lentivirus samples was incubated with the cells for 1 hour in a final volume of 100 µL with SFM. Following incubation, the infection media containing the pseudovirus was replaced with DMEM supplemented with 10% FBS and 1% PSG. Cells were maintained in culture for 2 days to allow for complete pseudo-infection, transduction, and expression of the GFP transgene. Wells were imaged on the Nikon Eclipse Ti-E Microscope, and the number of fluorescent cells was segmented via Ilastik (1.4.1b7-arm64) and subsequently selected and quantified Fiji/ImageJ (2.16.0/1.54 p). The fluorescent titer of the pseudotyped lentivirus was estimated by the expression of GFP in particulate infectious units/mL (PFU).

Additionally, the viral titer was quantified using the Lenti-X p24 Rapid Titre Kit (Takara Bio USA, Inc.; obtained from Scientifix) as per the manufacturer’s instructions. For p24 measurements of lenti^S Pseudotyped viruses, viral supernatant samples were diluted 1:100,000 after initially performing a titration. Samples were analyzed using a CLARIOstar Plus Plate Reader (BMG LABTECH) at 450 nm and the CLARIOstar Plus (v6.20, 15.09.2022) Reader software and MARS (v.20, 15.09.2022) data acquisition software. From the absorbance values and the concentrations of p24 determined via the constructed standard curve, the following equation was used to determine the infectious units / mL (IFU/mL) of lentivirus, based on 1 LP containing 8 × 10⁻⁵ pg of p24.

HaCaT^WT, HEK293T,^+ACE2 or HEK293T^WT cells were seeded at a density of 1.25 × 10⁶ cells per well in a 6-well plate and inoculated with pseudotyped virus at an MOI of 5. Pseudotyped virus infection was performed by “spinoculation” at 800 x rcf at 35°C for 1 h. Cells were left at 37°C with 5% CO₂ for either 2 days for SDS sample buffer cell lysis for immunoblot or 3 days before fixation in 4% paraformaldehyde for immunofluorescence. TGF-β treatment at 2 ng/mL (Peprotech Pty. Ltd. and Sigma-Aldrich Pty. Ltd.) was included as a positive control for pathway induction. Where treatment with TβR1 inhibitor SB-431542 (10 µM; Sigma-Alrich Pty. Ltd.) was required, cells were pretreated for 1 h.

Quantitative PCR was run on HaCaT^WT cell culture samples infected with multiple SARS-CoV-2 pseudotyped viruses. Cells were washed 24 hpi with PBS to remove remaining serum, cellular debris, and suspended cells. Adherent cells were detached using trypsin and pelleted at 200 x rcf for 5 min. The supernatant was aspirated, and RNA from these cell culture pellets was extracted using the RNA Isolate II mini kit (Meridian Biosciences Inc.) following the manufacturer’s instructions. The resulting yield was estimated by the NanoDrop1000 (Thermo Fisher Scientific, Australia) and ~500 ng of RNA was reverse transcribed into cDNA using iScript SuperMix (Bio-Rad Laboratories Inc.). Quantitative PCR was carried out using the SYBR Select Mastermix (Applied Biosystems) following the primers provided below:

5′-CCGGAACAGCCTGAAGAAGTG-3′ (PAI-1 FW)

5′-GTGTTTCAGCAGGTGGCGC-3′ (PAI-1 REV)

5′-ACAACTTTGGTATCGTGGAAGG-3′ (GAPDH FW)

5′-GCCATCACGCCACAGTTTC-3′ (GAPDH REV)

5′-AAACTCATGAGCAGTCTGCA-3′ (IFN-β FWD)

5′-AGGAGATCTTCAGTTTCGGAGG-3′ (IFN-β REV)

PCR was performed on the QuantStudio6 Pro Instrument (Thermo Fisher Scientific, Australia). Technical replicates were averaged and normalized against GAPDH. Data were obtained from four biological replicates (12 technical replicates) and expressed as fold change ∆∆CT from the respective control groups.

WR-VACV LifeAct(LA)-GFP was chosen as an unrelated dsDNA viral model to investigate the impact of SARS-CoV-2 S protein on HaCaT cells. Lifeact is a 17-amino acid peptide that is used to visualize actin, and the GFP tag allows for visualization of the F-actin cytoskeleton in live cells (104). Lifeact-GFP was previously generated and cloned in VACV genes C10L and C11R by homologous recombination, as previously described in (70). Infections with WR-VACV were conducted using a standard methodology. Initially, the growth medium was aspirated from the cell cultures, followed by a single wash with warm PBS. Subsequently, each virus was diluted to the required MOI in DMEM without FBS (serum-free medium, SFM). This diluted virus was then applied to the cells. The cells were incubated with the virus mixture for 1 h at 37°C, 5% CO₂. Post-incubation, the virus inoculum was removed, and cells were washed once with warm PBS and then overlaid with fresh growth medium consisting of a 1:1 mixture of 2 x MEM (supplemented with 10% FBS and 1% PSG, Gibco) and 1.5% carboxy-methyl cellulose (CMC, Sigma-Alrich Pty. Ltd.). Plaques were imaged at either 2 or 4 dpi under 488 nm light via a ChemiDoc plate imager (BioRad) and quantified using Fiji/ImageJ (2.14.0/1.54 f).

Several protein treatments were used as either controls or assessed during this research. Each of the proteins used was maintained in DMSO and frozen for extended storage according to indicated instructions either from the supplier or the manufacturer. For TGF-β treatment, cells were treated similarly to viral infections, where TGF-β (2 ng/mL; Peprotech) was added to both the SFM and recovery medium. Inhibition of TGF-β signaling or SMAD3 phosphorylation during infection was achieved by including SB-431542 (10 µM; Tocris) or SIS3 (10 µM; Selleck) in the infection and recovery medium. For signaling-based experiments, S protein and RBD proteins were treated at 105 ng/mL and 200 ng/mL, respectively, within 10% DMEM, and cells were left for 24 h before transcriptional or protein-based experiments. In plaque assays, S protein and RBD were treated at various concentrations from 0.1 to 20 µg/mL (0.1 µg/mL, 1 µg/mL, 5 µg/mL, 10 µg/mL, and 20 µg/mL) after serum starvation (0.1% FBS) for 24 h.

Recombinantly produced SARS-CoV-2 S protein HexaPro and SARS-CoV-2 receptor binding domain (RBD) proteins were provided by the Drug Discovery Initiative (University of Sydney). The SARS-CoV-2 S protein HexaPro expression plasmid was a gift from Jason McLellan (Addgene plasmid, https://www.addgene.org/154754/↗). The HexaPro version of the SARS-CoV-2 S protein was selected as it possesses six beneficial proline mutations that confer improved expression and stability of the S protein trimer relative to the parental protein. The gene also contains a C-terminal HRV3C-His₆-Strep tag to enable the purification of the protein. SARS-CoV-2 S protein HexaPro was expressed in EXPI293F cells via transient transfection of the expression plasmid using polyethyleneimine (PEI, 25 kDa). Cultures were incubated at 37°C in 5% CO₂ with orbital shaking at 130 rpm for 96 h following transfection. At 96 h post-transfection, cells were harvested via centrifugation at 4,000 × rcf for 20 min, and the supernatant was separated and filtered using a 0.22 µm filter. SARS-CoV-2 S protein HexaPro in the supernatant was purified by immobilized metal affinity chromatography (IMAC) using Ni-NTA agarose. Following IMAC, the protein was dialyzed into a buffer comprising 2 mM Tris pH 8.0 and 200 mM NaCl and was further purified via size-exclusion chromatography (SEC) using a HiLoad 16/600 Superdex 200 column (Cytiva). The protein was eluted from the SEC column using a buffer containing 2 mM Tris, 200 mM NaCl, pH 8.0, concentrated, and stored at −80°C. The protocol used was adapted from references 53, 54.

SARS-CoV-2 RBD (residues 319–541 of SARS-CoV-2 S protein flanked with the N-terminal signal peptide (residues 1–14) from the native SARS-CoV-2 S protein to enable secretion of the expressed protein and a C-terminal His₆-tag to enable purification) was produced using an expression vector kindly provided by Dr. Florian Krammer (Icahn School of Medicine, Mt Sinai) (105, 106). This plasmid was used to transiently transfect EXPI293F cells using PEI (25 kDa). Cultures were incubated at 37°C in 5% CO₂ with orbital shaking at 130 rpm for 72 h following transfection. Supernatants were separated via centrifugation at 4,000 × g for 20 min 72 h post-transfection. SARS-CoV-2 RBD in the supernatant was purified using Ni-NTA agarose pre-equilibrated in 20 mM NaH₂PO₄ pH 8.0, 500 mM NaCl, and 20 mM imidazole. The protein was eluted from the Ni-NTA agarose using a buffer comprising 20 mM NaH₂PO₄ pH 7.4, 300 mM NaCl, and 500 mM imidazole. SARS-CoV-2 RBD-containing fractions from IMAC were pooled, concentrated, and purified further via size exclusion chromatography (SEC) using a Superdex 200 10/30 Gl column (Cytiva). SARS-CoV-2 RBD was eluted from SEC column using a buffer containing 20 mM HEPES pH 7.4 and 150 mM NaCl, concentrated, and stored at −80°C.

HaCaT, HEK293T^+ACE2, or HEK293T^WT cells were seeded onto round or square glass coverslips; poly-L-lysine solution (0.1% wt/vol, Sigma, Australia) was added to coverslips before HEK293T^+ACE2 or HEK293T^WT cells were seeded. When cells were 80% confluent, infection or treatment was carried out as previously described. Upon reaching the designated timepoint, coverslips were fixed with paraformaldehyde (PFA) in cytoskeletal buffer (CB) [10 mM 2-(N-morpholino) ethanesulfonic acid (MES) buffer, 0.15 M NaCl, 5 mM ethylene glycerol tetraacetic acid (EGTA), 5 mM MgCl2, 50 mM glucose, pH 6.1].

Post-fixation, cells were permeabilized in 0.1% Triton X-100 in CB, followed by 20 min in blocking buffer (1% (vol/vol) bovine serum albumin (BSA), 2% (vol/vol) fetal bovine serum in 1X CB). Cellular protein PAI-1 was visualized using anti-PAI-1 (Santa Cruz Antibodies, C-9) or fluorescently tagged lenti^S-Anc. Primary antibodies were left for 30 min at 1:200 dilution. The corresponding secondary antibody was added at 1:1,000 dilution for 30 min. Finally, slides were stained in DAPI diluted in CB (1 µg/mL) for 2 min and then mounted on glass slides with Mowiol mounting media (10% (wt/vol) polyvinyl alcohol 4–88, 25% (wt/vol) glycerol).

Slides were visualized using an Olympus BX51 Microscope with Reflection Fluorescence System, Mercury Burner (U-RFL-T), F-view monochrome fluorescence camera and DAPI (347 nm/442 nm [#31013 v2]), FITC (495 nm/515 nm [#31001]), and TxRed (594 nm/610 nm [#31004]) Chroma filters. Micrographs were captured using AnalySIS LS Starter (Olympus Soft Imaging Systems, v2.8) and processed using Fiji/ImageJ (2.14.0/1.54 f).

PAI-1 activity was enumerated by measuring the mean fluorescent intensity per individual cell in three replicate images via ImageJ. Figure corrections for color blindness were completed in ImageJ, where blue was changed to magenta, green was changed to yellow, and red was changed to green. For each of the measurements, regions of interest (ROIs) were chosen based on the cell nuclei, where Alexa Fluor 568 was indicative of the detection of PAI-1. Results were then normalized in comparison to the negative (untreated/uninfected) and TGF-β control.

HaCaT or HEK293T cells were transfected with firefly-luciferase reporter and Renilla control to perform a luciferase assay, as described previously by the literature (107). The firefly-luciferase reporter plasmids used within this study included SMAD2-dependent ARE-Luc, SMAD3-dependent SBR₆-Luc, and SMAD3/4-dependent CAGA₁₂-Luc.

To obtain the highest transfection efficacy in cells and prevent cytotoxicity, the transfection conditions were optimized with 0.5 µg of Luciferase reporter used and 0.005 µg of Renilla control plasmid diluted in SFM. Lipofectamine 2000 (Invitrogen, Thermo Fisher Scientific) was used as a transfection reagent where 3 µL was used for every 1 µg of DNA and diluted in SFM. This reagent was left initially for 5 minutes and then combined with DNA (including reporter assays) for a subsequent 30 min. DNA and Lipofectamine 2000 solution were then applied to cells in SFM and left for 4–6 h at 37°C. Media was subsequently aspirated and replaced with 10% DMEM and left for 24 h allowing for effective transfection to occur.

Luciferase reporter assays were performed in cells transfected with each of the reporter plasmids (107–109). Cells were infected or treated after reaching 80% confluence, and infection/treatment time differed: 8 h (TGF-β) or 24 h (S protein, RBD). After infection/treatment time, cells were lysed using Passive Lysis Buffer (Promega, USA) to extract firefly-luciferase and Renilla in each sample. Using a TD-20/20 luminometer (Turner Designs), the light emission of firefly-luciferase was divided by Renilla to determine the relative amount of luciferase present in each sample.

Cells were washed in PBS and harvested on ice in sodium dodecyl sulfate (SDS)-reducing sample buffer (62.5 mM Tris-HCl, 0.25 M glycerol, 2% SDS, 0.01% bromophenol blue, 12.5% β-mercaptoethanol) and boiled three times at 95°C for 3 min. Proteins were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE) in running buffer (25 mM Tris Base, 190 mM glycine, 0.1% SDS, pH 8.3) using 15 well-precast mini-PROTEAN gels (BioRad, USA). Separated proteins were transferred to a nitrocellulose membrane (Hybond-C Extra; Amersham Biosciences) by electroblotting using Transfer Buffer (48 mM Tris, 39 mM Glycine, 20% methanol, 0.04% SDS).

Membranes were probed with primary antibodies in TBST-Milk (5% milk in tris-buffered saline, TBS with 0.1% Tween 20) or TBST-BSA (4% BSA in TBS with 0.1% Tween 20) for the detection of proteins. A primary monoclonal antibody specific to SMAD3 (C67H9), phosphorylated SMAD3 (D12E11), or SMAD3-dependent protein PAI-1 (SC-5297) was assessed. Anti-β-actin (Sigma-Aldrich, AC-74) was used as a cellular control. Membranes were washed three times in TBST-Milk or TBST-BSA and then probed with appropriate secondary antibodies conjugated to either IRDye 650LT or IRDye 800CW. These stain-free secondary antibodies were then exposed to UV light for 45 minutes to excite the proteins, providing a fluorescent signal and image from an LI-COR Odyssey CLx image scanner (Millennium Science). Images were captured using ImageLab Software and processed using Fiji/ImageJ (2.14.0/1.54 f).

To assess TGF-β signaling induced by the SARS-CoV-2 Ancestral Variant (A.2.2), live virus experiments were conducted under biosafety level 3 (BSL-3) conditions at the Kirby Institute, UNSW, Australia. Live SARS-CoV-2, initially isolated from diagnostic respiratory specimens that tested RT-qPCR positive, was propagated in VeroE6TMPRSS2 cells at an MOI of 0.025 and incubated for 24 h before the supernatant was collected, cleared, and frozen. Calu3 cells were seeded in a six-well plate and grown to 80% confluency. These cells were subsequently infected with SARS-CoV-2 at an MOI of 0.01 and left for 72 h. Prior to infection, half of the cells were treated with 10 µM SB-431542 for 1 h, and after the media was replaced during infection, the inhibitor was maintained at the same concentration. The same lysis procedure was performed as detailed in the immunoblot section using (SDS)-reducing sample buffer. Subsequent immunoblot analysis was conducted in BSL-2 conditions.

To assess the role of TGF-β signaling in SARS-CoV-2 Ancestral Variant (A.2.2)-induced cell death, the same live viral infection reagents and conditions as the immunoblot analysis were utilized. The methodology for these experiments was adapted from prior work by the Turvlle Lab and Anupriya Aggarwal (110, 111). Calu-3 cells were seeded in a 384-well plate and grown to confluency. An hour before infection, eight independent biological cell replicates were treated with either SB-431542 (10 µM), DMSO (0.1% vol/vol), or left untreated. These cells were infected with eight subsequent 5-fold serial dilutions (starting with 1:200) of virus and left for 72 h. Following incubation, Hoechst-33342 live nuclear dye (5% vol/vol; Invitrogen, R37605↗) was added to the wells 2 h prior to imaging on the InCell Analyzer HS2500 high content microscope (Cytiva). The entire well surface was imaged, and nuclei were enumerated using fluorescence microscopy. Fluorescence images were processed with IN Carta analysis software (Cytiva) to obtain total nuclei counts per well. For survival curves and EC₅₀ calculations, nuclei counts were normalized so that 100% represents the average cell number for mock-infected controls and 0% represents the average cell number for the highest viral concentration test, using Min-Max standardization.

A dose-response curve was generated by plotting % survival against viral concentration (fivefold log dilution). A nonlinear regression model using a 4PL sigmoidal curve (dose-response curve with variable slope and four parameters) was implemented to plot curves for each of the three treatment conditions (SB, DMSO, and untreated). Select dilutions were chosen for a direct comparison between SB-treated and untreated conditions to assess statistical significance.

Primary human bronchial epithelial cells (PBECs) were isolated as described in the previous study by Wang et al. (71) To examine cell-type specific responses and intercellular signaling during SARS-CoV-2 infection in an air-liquid interface model, we re-analyzed single-cell RNA-seq data from ALI-cultured primary human airway epithelial cells (ALI-PBEC) derived from five adult donors infected with SARS-CoV-2 (A.2.2. Strain, 3 × 10⁵ PFU) and harvested at 48 hpi. Cell populations were annotated based on published markers, focusing on the basal and ciliated cell populations. For each donor, we quantified five variables at the single-cell level: basal cell proportion, TGF-β1 and type I IFN transcript abundance, TGF-β pathway signature, and expression of RGD-binding integrins (described in Fig. S7B). A scatterplot for three independent time points from each of the patients' ALI-PBEC models was constructed for bivariate data and correlation analysis.

The TGF-β signature used was based on a bulk-RNA data set (GSE84135↗, (72)↗), in which cultures of submerged airway epithelial cells (basal cells) were treated with TGF-β and measured for gene expression over 72 hours. The DESeq2 differential expression script from this study was adapted, where significantly upregulated genes (BH-adjusted P-value < 0.05, log2foldChange > 1) were considered TGF-β-responsive. The full list of genes used in the GVSA and comparisons is listed below:

LAMC2, ITGB6, FN1, SERPINE2, TGFBI, SERPINE1, ABCA1, VIM, NEAT1, MALAT1, TALAM1, COL4A1, KRT14, LAMA3, PMEPA1, HTRA1, SLITRK6, SKIL, TPM1, TNC, THBS1, COL1A1, SERPINB5, COL7A1, LTBP2, IGFL1, ITGAV, PLAU, FSTL3, COL4A2, STON2, FLRT2, JAG1, NT5E, DIXDC1, PLEK2, FSTL1, HMGA2, ITGA5, SPARC, AMIGO2, SOX4, IVNS1ABP, TMSB4X, PTPRK, F2RL1, CHD9, FERMT1, LIMA1, TGFBR1, NLRP1, ITGB1, SPOCK1, ITGA6, MCAM, ADAM19, GJB2, GPCPD1, FAT1, JUN, VCAN, BMPR2, ATP1B1, NAV1, SNAI2, MFAP2, CALD1, BLTP1, COL17A1, ADAMTS7, TAGLN, ARHGAP21, LAMB3, VMP1, TTN, KCNJ15, DSC2, SULF2, ZFP36L1, CD59, PTPRM, TUT4, KRT16, MBOAT2, STK38L, HMGA2-AS1, TUBB3, JMJD1C, DUSP10, ITGA2, PTHLH, SCD, and GOLGB1.

To infer the relative abundance of respiratory epithelial cells within ALI-PBEC bulk RNA-seq data sets, cellular deconvolution was performed using CIBERSORTx (112). A custom single-cell reference, derived from unpublished ALI culture data sets curated by the Faiz Lab (RBMB, School of Life Sciences, UTS), was used to distinguish resting basal, suprabasal, multiciliated, club, and goblet cell populations. Raw counts were normalized to counts per million (CPM), followed by batch-corrected deconvolution using S-mode with 100 permutations as parameters. The output was returned as proportional estimates (e.g., “0 .XX”) for each cell type per sample.

These proportions were then paired with gene-level expression data from the same RNA-seq samples to generate bivariate data sets, enabling correlation analyses between cell type abundance and transcriptomic features. Spearman correlation coefficient (ρ) was calculated for each donor across the five cell types, with statistical significance assessed for each comparison, where the null hypothesis aligned with rho = 0 or no correlation. Correlation results were then aggregated across the cohort to identify consistent patterns of association.

For indirect ELISA analysis of cell supernatants for detection of TGF-β, HaCaT cells were grown to confluency in six-well dishes prior to 24 h of serum starvation (using SFM). At this point, TGF-β or S protein treatments, or Lenti^S-Anc. or Lenti^VSV.G infections were applied in SFM, as previously described. Supernatants were collected at specified time points and used alongside the Human TGF-β1 ELISA kit (Sigma-Aldrich) according to the manufacturer’s instructions. A standard curve was generated using the supplied control standard. Samples were analyzed using CLARIOstar Plus Plate Reader (BMG LABTECH) and the CLARIOstar Plus (v6.20, 15.09.2022) Reader software and MARS (v.20, 15.09.2022) data acquisition software.

Data were collected and plotted in GraphPad Prism 10 (Version 10.5.0, GraphPad Software, LLC). Error bars represent the mean ± the Standard Error of the Mean (SEM) unless reported otherwise. The significance comparison between groups differed for each set of data. Plaque assays used unpaired t-test, IFA relative fluorescence analysis, and normalized luciferase activity using a one-way ANOVA and Tukey’s multiple comparison test. Values of P < 0.05 were considered statistically significant.

The normalization of luciferase activity values was performed based on the cell line, and the data were taken from and relative to the positive (TGF-β treated) and negative (uninfected, non-treated, non-transfected) controls for each sample as transfection efficiency varied. Min-Max standardization was used to satisfy the requirements to provide the data with values between 0 and 1.

For RT-qPCR experiments, each biological replicate represents the average of six technical replicates, performed at both the standard and ⅛ dilution concentrations. C_t values were first normalized to the GAPDH housekeeping gene (∆C_t), and fold changes were calculated using the ∆∆C_t method relative to the uninfected control.

From this value, the fold change in comparison to the uninfected samples (which was denoted as 1) was calculated. The significance was assessed using a one-tailed Mann-Whitney test (a nonparametric test).

Sigmoidal dose-response curves and EC₅₀ values were obtained with GraphPad Prism 10 (Version 10.5.0, GraphPad Software, LLC) using the nonlinear regression for inhibitor versus response with variable slope and four parameters.

For bulk RNAseq data, each correlation was assessed using Spearman correlation coefficient (rho) for each donor across the five variables. Statistical significance was determined for each correlation, and results were aggregated across the cohort to identify consistent patterns.

For all tests, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.