Identification of a novel anoikis signalling pathway using the fungal virulence factor gliotoxin

Anoikis is a form of apoptosis induced by cell detachment. Integrin inactivation plays a major role in the process but the exact signalling pathway is ill-defined. Here we identify an anoikis pathway using gliotoxin (GT), a virulence factor of the fungus Aspergillus fumigatus, which causes invasive aspergillosis in humans. GT prevents integrin binding to RGD-containing extracellular matrix components by covalently modifying cysteines in the binding pocket. As a consequence, focal adhesion kinase (FAK) is inhibited resulting in dephosphorylation of p190RhoGAP, allowing activation of RhoA. Sequential activation of ROCK, MKK4/MKK7 and JNK then triggers pro-apoptotic phosphorylation of Bim. Cells in suspension or lacking integrin surface expression are insensitive to GT but are sensitised to ROCK-MKK4/MKK7-JNK-dependent anoikis upon attachment to fibronectin or integrin upregulation. The same signalling pathway is triggered by FAK inhibition or inhibiting integrin αV/β3 with Cilengitide. Thus, GT can target integrins to induce anoikis on lung epithelial cells.

D etachment-induced apoptosis or anoikis is a crucial process to eliminate aberrant cells in the outer layer of epithelia 1 . Lack of anoikis is a hallmark of cancer progression as cells that continue to survive in suspension are prone to metastasise 1 . Originally described by Frisch et al. 2 , anoikis is typically induced by detaching cells with trypsin and preventing their re-attachment to polyHEMA-coated plates. However, this system is artificial as trypsin inappropriately modifies adhesion molecules leading to the activation of signalling pathways that may not reflect physiological ways of anoikis.
Cells attach to the extracellular matrix via integrins. Integrins consist of transmembrane α and β chains, which form at their extracellular N termini an interaction interface with the Arg-Gly-Asp (RGD) motif of matrix components such as fibronectin or vitronectin 3 . On the intracellular side integrins recruit components that link adhesion signals to cell survival, cell cycle control and cytoskeletal rearrangement 1,4,5 . Key players are focal adhesion kinase (FAK) 6 , integrin-linked kinase (ILK) 7 and Src tyrosine kinase 8 . FAK is autophosphorylated at Y397 upon integrin activation 9 and subsequently phosphorylates adapter molecules such as paxillin, vinculin and Rho-, Rac-and Cdc42 GTPasemodulating proteins to regulate the actin cytoskeleton 1,6,10 . FAK and ILK further activate PI3K/AKT, ERK/MAPK and JNK signalling pathways 6,7,11-13 . Apoptosis can be induced by extrinsic death receptor signalling where activated receptors of the tumour necrosis facto superfamily recruit and activate caspase-8 via FADD leading to the cleavage and activation of effector caspase-3 14 . Alternatively, so called BH3-only proteins of the Bcl-2 family sense apoptotic signals and convey them to Bax/Bak-mediated mitochondrial outer membrane permeabilization 15 . The subsequent release of cytochrome c induces the formation of the Apaf-1/caspase-9 apoptosome, which results in caspase-3 activation. Although evidence was initially presented that death receptor signalling via FasL 16 and/or FADD 17 plays a role in anoikis induction, this could not be confirmed in other studies [18][19][20] . By contrast, several cellular processes seem to block mitochondrial apoptosis signalling in adherent cells 19 . Activation of AKT, elicited by integrin engagement, leads to the phosphorylation of the BH3only proteins Bad 21 and Bim 22 , which are sequestered and inactivated by 14-3-3 proteins. AKT also inhibits Forkhead transcription factors (FOXOs) 23 , which are responsible for the transcriptional upregulation of the BH3-only proteins Bim, Puma and Bmf 24 . Moreover, both ERK- 25 and PI3K/AKT-mediated phosphorylation 24 of Bim lead to its proteasomal degradation. However, it has been unclear if activation of these BH3-only proteins during anoikis is indeed linked to AKT and/or ERK inhibition.
The mould Aspergillus fumigatus causes a severe pulmonary disease termed invasive aspergillosis 26 . Under healthy conditions, airborne conidia released by A. fumigatus are successfully eliminated from the pulmonary cavities by alveolar macrophages, neutrophils and leucocytes 27 . In immunosuppressed patients, however, A. fumigatus germinates, invades the lung and causes severe and often lethal systemic infections 26,27 . The breakage of the epithelial barrier is the most likely cause for the invasive property of A. fumigatus. Accumulating evidence suggests a crucial role of the major virulence factor gliotoxin (GT) in this process because fungi lacking GT production are much less virulent than wild-type (WT) strains 28 . We previously showed that GT induces a rapid detachment of human lung epithelial cells and mouse fibroblasts before they undergo caspasedependent apoptosis 29,30 . GT-induced apoptosis requires a JNK-mediated triple phosphorylation of Bim at S100/T112/S114, which increases the pro-apoptotic activity of Bim 30 . Moreover, both in vitro and in vivo, GT-induced cytotoxicity depended on Bak 29 indicating that epithelial barrier breakage and lung invasion after A. fumigatus infection may be due to GT-mediated anoikis.
Here we use GT to delineate for the first time an entire anoikis signalling pathway in human lung epithelial cells that leads to the direct activation of the pro-apoptotic family member Bim. GT covalently modifies the RGD-binding domain of integrin α and β chains, leading to rapid cell detachment followed by FAK inactivation and subsequent activation of a RhoA-ROCK-MKK4/ MKK7-dependent signalling pathway, which activates JNK-and Bim-mediated apoptosis.

Results
GT employs MKK4 and MKK7 to activate JNK-dependent apoptosis. We previously reported that JNK is required for GTinduced apoptosis 30 . We therefore sought to identify the kinase(s) responsible for JNK activation. Possible candidates were the mitogen-activated protein kinases MKK4 and MKK7. Indeed, after 4-6 h of GT treatment of human bronchial epithelial cells (BEAS-2B) both MKK4 and MKK7 were phosphorylated in their activation loops (S257/T261 and S271/T275, respectively) as detected by phosphospecific antibodies (Fig. 1a). This coincided with the cleavage of the caspase-3 substrate PARP.
GT triggers a Rho-dependent phosphorylation cascade. Since GT causes rapid cell detachment associated with cytoskeletal changes ( Supplementary Fig. 1), we looked for an upstream MKK4/MKK7 activator, which is linked to these events. Recent evidence indicated that Rho-related small GTPases such as RhoA, Rac1 and Cdc42 do not only control actin remodelling but also the activity of the JNK cascade 31 . This prompted us to investigate if the Rho-associated protein kinase (ROCK) was involved in GT-induced MKK4/MKK7 activation and detachment-induced cell death.
For that purpose, we treated BEAS-2B cells with two pharmacological ROCK inhibitors, H-1152 and Y-27632, before applying GT for 6 h. Both inhibitors completely abolished GT-induced JNK phosphorylation and caspase-3 and PARP processing (Fig. 2a) as well as Bim phosphorylation at T112/S114 (Fig. 2b). An in vitro JNK activity assay showed that GT-induced c-Jun phosphorylation was ablated after H-1152 treatment (Supplementary Fig. 2E and 2F). Importantly, the general caspase inhibitor QVD did not affect GT-induced JNK phosphorylation but expectedly blocked caspase-3 activation (Fig. 2a).
Similar results were obtained in MEFs. H-1152 diminished MKK4 and JNK activation, Bim phosphorylation and caspase-3 processing after GT treatment (Fig. 2c). Consequently, both ROCK inhibitors abrogated caspase-3/7 activity (Fig. 2d) and apoptosis (Fig. 2e) to the same extent as the caspase inhibitor QVD indicating that GT-induced caspase-3 activation was caused by increased ROCK activity.
We next examined the role of ROCK-activating Rho GTPases in GT-induced anoikis signalling. For that purpose, we took advantage of Rhotekin, a known binding partner and substrate of RhoA and RhoC 32 . We performed GST-Rhotekin pulldowns of untreated and GT-treated BEAS-2B cell extracts and examined them for the abundance of active RhoA. As a positive control, we used the bacterial toxin CNFy, which inhibits the GTPase activity of Rho proteins, thereby keeping them in a permanent GTP-bound active state (Fig. 3a) 33 . As shown in Fig. 3a, the amount of active RhoA-GTP in the GST-Rhotekin pulldowns started to increase after 40 min of GT treatment, a time that conincided with GT-induced cell detachment (Supplementary Fig. 1). To confirm that increased Rho activity was crucial for GT-induced anoikis signalling, we treated BEAS-2B cells with the bacterial toxin C3, which inhibits Rho activity by ADP ribosylation 34 . Rho inhibition by C3 reduced GT-induced phosphorylation of MKK4 and JNK (Fig. 3b) as well as the phosphorylation of Bim at its T112/S114 JNK phosphorylation sites (Fig. 3c).
Finally, we wanted to know if the C3 toxin had any inhibitory effect on GT-induced ROCK activation. We therefore monitored the phosphorylation of a major substrate of ROCK, myosinbinding subunit of myosin phosphatase (MYPT1) 31 by western blot analysis. As shown in Fig. 3d, e, while the phosphorylation of MYPT1 gradually increased after GT treatment, this was not the case when BEAS-2B cells were pretreated with the C3 toxin. Thus, GT triggers Rho activation (particularly RhoA) to stimulate

Mkk4
-/-/Mkk7 GT disrupts the focal adhesion complex. Adherent cells treated with GT rapidly detach before they die (Supplementary Fig. 1) 30 . This suggests that disruption of focal adhesions might be an early event of GT action. Paxillin is a scaffold protein at the cytoplasmic side of focal adhesions responsible for recruiting FAK, a crucial mediator of integrin signalling 35 . We used green fluorescent protein (GFP)-labelled paxillin to monitor its subcellular localisation before and after GT treatment by confocal time-lapse microscopy. While in adherent healthy BEAS-2B cells, GFP-paxillin was present in focal adhesions as well as in the cytosol (Fig. 4a), it mainly localised to vesicular structures after GT treatment before the cell rounded up and died ( Fig. 4a and Supplementary Movie 1). Co-transfection with the endosomal marker mRuby-Endo-14 revealed that GFP-paxillin resided on endosomal membranes after GT treatment (Fig. 4b).
Since GT changed the structure/composition of focal adhesions, we sought to study the role of FAK in GT-induced apoptosis. FAK is an interesting downstream target of GT because it is known to regulate Rho GTPases during stress fibre formation and focal adhesion turnover 36 . Moreover, it is known to phosphorylate and activate p190RhoGAP, which negatively regulates Rho activity in adherent cells 36 . We therefore investigated (i) if FAK activity was regulated by GT and (ii) if this affected the activity state of RhoA and therefore the RhoA-ROCK-MKK4/MKK7-JNK-mediated anoikis signalling.
Within 30 min of GT treatment of BEAS-2B cells, the activating phosphorylation of FAK at Y397 was lost (Fig. 5a). Simultaneously, the phosphorylations of its substrates p190Rho-GAP (Fig. 5c) and paxillin (Fig. 5b) diminished and paxillin was degraded (Fig. 5b). Dephosphorylation of p190RhoGAP results in a lower GAP activity towards Rho proteins, therefore favouring their active GTP-bound state 36 . Indeed, we observed a higher level of active RhoA (Fig. 3a) and phosphorylation of its Inhibition of FAK mimics GT-induced anoikis signalling. If inactivation of FAK by GT triggered apoptosis via Rho-ROCK-MKK4/7-JNK activation, pharmacological inhibition of FAK should activate the same signalling pathway. We therefore used the FAK inhibitor FAK14, which is highly specific for FAK because it blocks recruitment of its downstream target Src at Y397 37 . As shown for GT, FAK14 induced cell detachment (Fig. 5d) and apoptotic cell death of BEAS-2B cells within 6 h at low (3-10 μM) and high (50 μM) doses (Fig. 5e, f). In addition, both doses of FAK14 caused the rapid dephosphorylation of FAK at Y397, dephosphorylation and hence inactivation of p190Rho-GAP, phosphorylation of MKK4 and JNK (Fig. 5g, h) and phosphorylation of Bim at T112/S114 (Fig. 5g). Hence FAK inhibition stimulated the same anoikis signalling pathway as GT (Fig. 5i).
To confirm that inactivation of FAK was crucial for GTinduced JNK activation and caspase-mediated apoptosis, we overexpressed WT FAK or activated mutant forms of FAK (myrFAK, superFAK (FAK K578E/K581E) and myr-superFAK) 38,39 in BEAS-2B cells by lentiviral transduction. All variants of FAK were overexpressed to similar levels, were phosphorylated at Y397 and triggered enhanced phosphorylation of their substrate p130Cas 38 confirming their high kinase activities (Supplementary Fig. 3A and 3C). As a consequence, GT-induced anoikis signalling, i.e., JNK and Bim phosphorylation and caspase-3 processing and activation were delayed (Supplementary Fig. 3A, 3B, 3D and 3E). However, anoikis signalling was not fully blocked most likely because GT could still dephosphorylate and inactivate myrFAK, superFAK and myr-superFAK (Supplementary Fig. 3A and 3C). This finding is consistent with previous observations that the downstream signalling of super-FAK still depends on integrin-mediated adhesion 39 .
ROCK has been reported to regulate focal adhesions via an inside-out cytoskeletal signalling 40 . We therefore examined if the ROCK inhibitor H-1152 or the RhoA inhibitor toxin C3 had any effect on FAK phosphorylation and cell detachment. The inhibitors themselves did not detach BEAS-2B cells (Supplementary Fig. 1) and H-1152 induced a minor FAK dephosphorylation/inactivation without affecting JNK phosphorylation ( Supplementary Fig. 2G) as reported before 40 . In the presence of GT, both H-1152 and C3 slightly delayed cell detachment ( Supplementary Fig. 1). This did however not affect focal adhesions since FAK was similarly dephosphorylated by GT irrespective of the presence of absence of H-1152 ( Supplementary Fig. 2G). Only JNK and caspase-3 failed to be activated in the presence of the ROCK inhibitor ( Supplementary  Fig. 2G) confirming that ROCK is a downstream mediator of GTinduced apoptosis and not a major inside-out signalling regulator of cell detachment in this system.
GT can directly target integrins at the RGD binding site. Since GT caused cell detachment and inactivation of the focal adhesion complex, it may directly target integrins. GT contains an intramolecular disulphide bond essential for its cytotoxic activity 28 . It may therefore covalently modify cysteine residues in integrin α and β chains, which are critical for integrin activation and/or their binding to extracellular matrix components 41,42 . For that purpose, we incubated recombinant human integrin αVβ3 with GT and determined peptides with possible GT-cysteine adducts by mass spectrometry (MS) analysis. Two cysteines were found to be modified by GT, Cys158 in the seven blade β-propeller domain of αV and Cys258 in the ligand-binding β-I domain of β3 integrins ( Supplementary Fig. 4A). Both cysteines are highly conserved among the α and β chains of various integrins (Supplementary Fig. 4B) and form intracellular disulphide bridges (Cys158 with Cys138 in αV and Cys258 with Cys299 in β3) 43,44 that determine efficient binding of integrins to the RGD motif in fibronectin and vitronectin 41,42 . To provide further evidence that GT modified integrins at cysteines in the RGD binding site, we treated BEAS-2B cells with GT for 30 min and subjected a total cellular extract to anti-GT immunoprecitations (IPs) followed by anti-integrin αV or β1 western blot analysis. As shown in Supplementary Fig. 5A and 5B, both integrin chains were specifically detected in anti-GT as compared to control IgG1 IPs. This was however not the case when the cells were pretreated with an RGD peptide before GT addition, or the extract was incubated with dithiothreitol (DTT) and iodoacetamide before anti-GT IP (Supplementary Fig. 5A and 5B). Interestingly, GT seemed to also interact with E-cadherin ( Supplementary Fig. 5C) but not with the epidermal growth factor receptor (EGFR) (Supplementary Fig. 5D).
To confirm that integrin binding to extracellular matrix components is indeed perturbed by GT, we determined integrin-binding capacity of untreated and GT-treated BEAS-2B cells by fluorescence-activated cell sorting (FACS) analysis using a  Fig. 1). By confocal immunofluorescence analysis using antibodies against pFAK and the endosomal protein EEA1 as well as GFP-paxillin overexpression we further showed that phosphorylated FAK (pFAK) colocalized with paxillin in focal adhesions in untreated cells ( Supplementary  Fig. 6B). In response to GT paxillin is endocytosed (Supplementary Fig. 6B and Fig. 4a) but rarely any pFAK is found on endosomes ( Supplementary Fig. 6A). This confirms that after GT treatment FAK gets rapidly dephosphorylated in the focal adhesion plaques at the plasma membrane before it is taken up into the cell together with paxillin.
Integrins are also endocytosed after cell detachment and recycle back to the plasma membrane 45 . We therefore monitored the fate of integrin β3 surface expression on BEAS-2B cells (Fig. 6b) and integrin β1 surface expression on MEFs (Fig. 6c) after GT treatment by FACS analysis using β chain-specific antibodies. Both integrin chains were removed from the cell surface (Fig. 6b, c). This however occurred only after 2 h of GT treatment suggesting that the primary action of GT was to inactivate RGD binding (Fig. 6a) before integrins were endocytosed.
GT-induced anoikis does not occur in suspension cells. Suspension cells do usually not express active integrins on their surface and hence do not form mature focal adhesions. Therefore, these cells should not be killed by GT if the toxin specifically targets focal adhesions in adherent cells. Indeed, while BEAS-2B cells effectively underwent apoptosis in response to GT, three different human and mouse suspension cell lines, BAF3, Jurkat and FL5.12, were insensitive to GT-induced apoptosis (Fig. 6d). Consistent with this finding, neither MKK4 nor JNK was phosphorylated/activated (Fig. 6e) and caspase-3 was not processed (Supplementary Fig. 8D). All cell lines expressed integrins β1 and β3 on their surface (Fig. 6a-c, Supplementary Fig. 7A and 7B). On BEAS-2B cells, they were in an active state as evidenced by RGD-FAM (Fig. 6a) and anti-active integrin β1 staining (Supplementary Fig. 7C). This was however not the case on Jurkat cells where integrin β1 was not detected with an anti-active integrin antibody ( Supplementary Fig. 7C). Moreover, while GT diminished surface staining of integrins β3 (Fig. 6b) and β1 (Fig. 6c) on BEAS-2B and MEF, respectively, none of the suspension cell lines showed such an effect. To further substantiate the integrin-dependence of GT for its pro-apoptotic action, we made use of K562 cells, which also grow in suspension but express very little integrin β1 and no integrin αV and β3 at all (Fig. 7e). Again, GT could not trigger caspase-3 activity (Fig. 7b) or processing (Fig. 7e) in these cells. However, when the cells were treated with 10 ng/ml phorbol 12-myristate 13-acetate (PMA), they largely upregulated active integrins αV, β1 and β3 on their surface (Fig. 7c-e) and became sensitive to GT-induced cell detachment (Fig. 7a), MKK4, JNK and Bim phosphorylation and caspase-3 processing (Fig. 7e) and activation (Fig. 7b). Similarly, BAF3 suspension cells made adherent by plating them on fibronectin-coated plates overnight displayed activated surface integrins as evidenced by RGD staining (Supplementary Fig. 8B) and acquired sensitivity to GTinduced detachment ( Supplementary Fig. 8A), caspase-3 processing ( Supplementary Fig. 8D), apoptosis ( Supplementary Fig. 8C) and activation of the same MKK4/JNK signalling pathway ( Supplementary Fig. 8D) as previously seen in BEAS-2B cells.
Hence GT exerts its cytotoxic activity primarily on adherent cells expressing active integrins, which qualifies it as a bona fide anoikis inducer.
Blocking αV/β3 integrins mimics GT-induced anoikis signalling. If activation of the JNK/Bim-dependent anoikis pathway by GT is a consequence of integrin inactivation, the same pathway should be triggered by integrin inhibitory antibodies or drugs. Cilengitide is an RGD-based compound that primarily blocks integrins αVβ3 and αVβ5 at lower and β1 at higher concentrations 46 . We first confirmed that BEAS-2B cells express αV, β3 and β1 integrins (Supplementary Fig. 7D). When these cells were exposed to 25 μg/ml Cilengitide, they exhibited cell detachment ( Supplementary Fig. 7E), FAK dephosphorylation and activation of JNK1/2 with subsequent Bim phosphorylation at T112/S114 ( Supplementary Fig 7F and 7G), activation and processing of caspase-3 processing (Supplementary Fig. 7G and 7H) and cell death ( Supplementary Fig. 7I) within 6-24 h. However, in contrast to GT, the cells treated with Cilengitide detached in clusters (instead of single cells) ( Supplementary Fig. 7E), and FAK dephoshorylation ( Supplementary Fig. 7F) and subsequent anoikis signalling were less pronounced and delayed (Supplementary Fig 7G-I). This was also true when Cilengitide was combined with an anti-integrin α5/β1 inhibitory antibody, which was ineffective alone and did not further enhance Cilengitideinduced anoikis (Supplementary Fig. 7I). This might be due to the  Fig. 4C, 4D and 4E). While GT induces maximal cell detachment and anoikis because it targets cysteines in the RGD binding groove, which are conserved among all integrins, we would need a combination of several inhibitory integrin antibodies to achieve the same effect.

Discussion
Here we used GT, the major virulence factor of A. fumigatus, to identify a novel anoikis signalling pathway. GT qualifies for an anoikis inducer for the following reasons: (i) it induces rapid cell detachment prior to apoptosis induction; (ii) it can directly modify N-terminal cysteines of α and β chains of integrins thereby interfering with their binding to extracellular matrix components; and (iii) it cannot kill suspension cells of the hematopoietic system and does not activate the anoikis signalling pathway in these cells.
Only when suspension cells are made adherent, i.e., either plated on fibronectin or induced to express integrins on their surface, they become sensitive to GT-induced anoikis via the same signalling pathway as epithelial cells. Although we provide compelling evidence that GT can directly target integrins by modifying cysteines at the RGD binding interface, we cannot exclude that it also modifies and disrupts binding domains in other adhesion molecules, including inflammatory receptors or cadherins. Indeed, we found that E-cadherin could be coimmunoprecipitated with anti-GT antibodies. This may explain why cells treated with GT detach as single cells while those treated with integrin inhibitory antibodies or compounds detach as cell sheets. Further studies are needed to identify the impact of GT on signalling pathways regulated by cadherins. Protection from anoikis was suggested to involve activation of ERK/MAPK and PI3K/AKT signalling [11][12][13] . These pathways should be turned off upon GT action, resulting in dephosphorylation of Bad and Bim and their release from 14-3-3 proteins 21,22 , the transcriptional upregulation of Bim, Puma and Bmf by FOXO activation 23,24 and the stabilisation of Bim due to lack of proteasomal degradation 25 . However, as we previously reported, neither Bad, Bmf nor Puma was required for GTinduced apoptosis, and the ERK/MAPK and AKT signalling pathways were still transiently activated after GT treatment 30 . Moreover, although Bim was essential for GT-induced cytotoxicity it required JNK-mediated phosphorylation at S100/T112/ S114 rather than increased protein stability for effective Bax/Bak activation 30 . This indicated that GT uses a JNK-dependent, but ERK/AKT-independent pathway for anoikis signalling 30 . Previously, Stupack et al. 47 reported that unligated integrins or β-integrin tails recruit caspase-8 to the membrane and induce apoptosis distinct from anoikis in death receptor-independent manner. We therefore tested if caspase-8 was activated and required for GT-induced apoptosis. However, while recombinant FasL rapidly activated caspase-8 in BEAS-2B cells, we did not detect any caspase-8 activation after GT treatment for up to 6 h ( Supplementary Fig. 9A). Moreover, knocking down caspase-8 expression did not affect the kinetics of GT-induced anoikis but it blocked FasL-induced apoptosis (Supplementary Fig. 9B). Thus, GT uses a JNK-dependent, but ERK/AKT-and caspase-8independent pathway for anoikis signalling.
JNK/c-Jun-induced transcription in response to lysophosphatidic acid was previously shown to be mediated by RhoA-ROCK-MKK4 signalling 31 . The same pathway was involved in arsenic trioxide-induced apoptosis of chronic myelogenous leukaemia cells 48 . However, in both cases it remained elusive how RhoA was activated. Here we show that RhoA is activated by dephosphorylation and inactivation of p190RhoGAP as a result of GT-induced FAK inhibition. This is consistent with a report showing FAK-induced downmodulation of RhoA activity via p190RhoGAP 36 . The role of Rac or Cdc42 GTPases in mediating GT anoikis could be excluded because they do not activate ROCK 31,40 and have recently been shown to be inactivated by GT 49 .
ROCK1 and 2 play essential roles in regulating cell morphology, motility and cell fate 31,40,50 . Whether all these effects are mediated through changes in the actin cytoskeleton is still debated 50 . Ethanol, doxorubicin and serum starvation were found to induce caspase-dependent apoptosis via RhoA-ROCK1mediated myosin light chain phosphorylation and subsequent cytoskeletal rearrangements rather than JNK activation although the link to caspase-3 activation was not provided 50,51 . In other studies ROCK activated JNK but the result was stimulation of cell migration rather than apoptosis 52 . Furthermore, ROCK was shown to affect the membrane blebbing of apoptotic cells 53 . However, in this case ROCK was activated by caspase-3-mediated cleavage, which occurs downstream of cytochrome c release. Although we cannot exclude a cytoskeletal involvement for GTinduced, ROCK-mediated anoikis, our data show that ROCK does not signal back to focal adhesions (FAK) but stimulates the downstream phosphorylation/activation of MKK4 and MKK7 essential for anoikis. The precise mechanism of this activation is not yet understood. It either occurs through intermediate kinases such as MLK or ASK 54,55 and/or the scaffold proteins hCNK 56 or JIP-3 57 , which were shown to link ROCK to JNK activation. Once activated MKK4 and MKK7 then phosphorylate different sites in the activation loop of JNK 54 , and we indeed find that both kinases are required for GT-induced JNK activation.
Anoikis has been mainly studied in an artificial system, where cells are detached by trypsinization and prevented from reattaching to polyHEMA-coated plates 20 . This form of anoikis differs from that induced by GT because trypsin at least partially degrades surface integrins. Therefore, we 30 and others 58 could not confirm an earlier study 59 that JNK was involved in trypsinization-induced anoikis. By contrast, we show here for the first time that not only GT but also FAK and integrin inhibition trigger the same Rho-ROCK-MKK4/MKK7-JNK-mediated signalling pathway indicating that it represents a physiological way to induce anoikis (Fig. 8).
The novel anoikis signalling pathway will increase our understanding of diseases caused by excessive cell detachment or the resistance of detached cells to die. Overexpression of FAK has been shown to promote tumour progression and metastasis 60 , which can be counteracted by high levels of RhoA 61 . Moreover, RhoA downregulation is associated with increased breast cancer cell migration and invasion 62 . Thus, RhoA may act as a tumour suppressor not only by restricting tumour cell motility but also by inducing anoikis. Moreover, FAK is an attractive target for anticancer therapy as it has recently been proposed for pancreatic cancer 63 .
On the other hand, activation of FAK or integrins may render lung epithelial cells less sensitive to GT-induced anoikis and hence diminish increased epithelial cell permeability and subsequent lung invasion during A. fumigatus infections. An even better strategy is to develop GT inhibitors, which would prevent integrin inactivation and thereby completely block lung epithelial cell anoikis.

Methods
Reagents and inhibitors. The ROCK inhibitors Y-27632 (Rho Kinase inhibitor VI) and H-1152 were obtained from EMD Millipore (Billerica, MA, USA); Q-VD-OPh from MP Biomedicals (Eschwege, Germany); and GT from AppliChem (Darmstadt, Germany). The CNFy toxin (150 ng/ml) from Yersinia Paxillin is degraded and FAK is inactivated, resulting in inactive p190RhoGAP. RhoA is consequently active and induces anoikis via a kinase cascade from ROCK to MKK4/MKK7 to JNK, resulting in the pro-apoptotic triple phosphorylation of Bim (right). Inhibitory integrin antibodies such as Cilengitide or FAK inhibitors such as FAK14 can induce the same anoikis signalling pathway pseudotuberculosis and the C2I/C3 fusion toxin (Clostridium botulinum and Clostridium Limosum, respectively) combined with the C2II toxin from Clostridium botulinum (collectively termed C3 toxin) was produced and purified as described 33,34 . C3 toxin is a binary toxin consisting of C2II (200 ng/ml) and C2I/C3 (100 ng/ml) mixed in cell culture media. The FAK inhibitor 14 (FAK14) was purchased from Tocris (Bristol, UK), and Cilengitide, a cyclic pentapeptide RGD compound (cyclo-[RGDfN(Me)V]) 46 , the RGD peptide, DTT and iodocetamide were from Sigma (Taufkirchen, Germany). Attachment of suspension cells. Human fibronectin (Advanced BioMatrix, Carlsbad, CA, USA) was used for the attachment of BAF3 suspension cells. Twelvewell plates were coated with 50 µg/ml fibronectin in phosphate-buffered saline (PBS) overnight. On the next day, 100 000 cells were seeded per well and grown overnight. Non-attached cells were removed by washing with PBS before GT treatment or further analysis.

Plasmids
Generation of monoclonal anti-GT antibodies. Monoclonal mouse antibodies (mAb) were generated against a fusion protein between GT and the outer surface protein C (OspC) from Borrelia burgdorferi. A unit of 100 μg GT was incubated with 3 mg p-maleimidophenyl isocyanate (Thermo Fisher) in 500 μl dimethyl sulphoxide for 1 h at room temperature. In parallel 1 mg OspC was allowed to react with 100 μg N-succinimidyl S-acetylthioacetate (Thermo Fisher) in 500 μl of buffer A (50 mM sodium phosphate, 150 mM NaCl and 1 mM EDTA, pH 7.8). The two solutions were combined, 1.5 ml of buffer A (pH 6.8) was added and the mixture was incubated for another 2 h at room temperature, then buffered to pH 7.3 in PBS. Eight-to 12-week-old Balb/c mice were intraperitoneally immunised three times  Lentiviral and retroviral transductions. In all, 2 × 10 6 HEK 293T cells were transfected with 3 µg of the plasmid of interest, 3 µg envelope vector pMD2G and 3 µg packing vector pSPAX using Attractene Transfection Reagent (Qiagen, Hilden, Germany). After 12 h protein biosynthesis was augmented using 5 mM butyrate (Sigma). Butyrate was removed after 8 h and replaced with 4 ml full medium. Lentiviruses were harvested on the next morning and supplemented with 5 µg/ml polybrene (Sigma). Target cells were spinfected with fresh virus for 10 min at 450 × g. Cells were selected with 4 µg/ml puromycin in case of shRNA-mediated knockdown or used without selection for the overexpression of WT (exo-) FAK, myrFAK, superFAK and myr-superFAK. Integrin inhibition. BEAS-2B cells were detached in Cell Dissociation Buffer, enzyme-free (Thermo Fisher Scientific) to prevent degradation of surface molecules. Cells were either suspended in medium alone, medium containing 25 µg/ml Cilengitide (Sigma) or medium containing 100 µg/ml of the inhibitory anti-integrin β1 (clone P4C10, MAB1987Z, Merck Millipore) or 25 µg/ml of the inhibitory antiintegrin α5β1 (clone JBS5, MAB1969, Merck Millipore) antibodies or both and incubated for 0-24 h. At different time points cells or cellular extracts were subjected to survival (metabolic) and caspase-3 activity assay, western blot analysis or phase contrast microscopy analysis.
Mass spectrometry. MS was performed to analyse the covalent modification of integrins by GT. Human recombinant αVβ3 integrin (5 µg) was mixed with 5 µg bovine serum albumin as an internal control in 50 mM HEPES, pH 7.5, incubated with 100 μM to 1 mM GT at room temperature for up to 6 h and then denatured in 0.1% RapiGest (Waters, Eschborn, Germany) for 45 min at 70°C. Peptides were generated by digestion with 0.2 µg trypsin overnight at 37°C and then purified using C18 STAGE solid phase extraction columns (Varian, Palo Alto, CA, USA). A Q-Exactive Plus Hybrid Quadrupole-Orbitrap system (Thermo Scientific, Darmstadt, Germany) was used for mass spectrometry and operated in the datadependent mode 67 . Peptide sequences were identified by X! TANDEM (Version 2013.09.01) in conjunction with PeptideProphet using the reviewed canonical human (20272 protein sequences) and bovine (5994 proteins) combined sequence database, downloaded from Uniprot on 26.11.2013 and 15.07.2015 respectively, together with an equal number of randomised decoy sequences, generated by DBtoolkit. The PeptideProphet minimum probability threshold was set to 0.05. The GT modification of cysteines was calculated as an adduct with a variable modification with a mass difference of 326.0395 Da. Only peptides with a probability score > 0.99 were taken into consideration. Identified peptides were mapped on the protein sequence using the software Proteator.
Integrin Affymetrix microarray and RNAseq analyses. To quantify the expression of integrin genes, raw data of the transcriptome of untreated Beas-2B cells from three high-throughput experiments, one microarray and two RNAseq assays data were downloaded from Gene Expression Omnibus and Array Express with respective series/sample IDs GSE24025/GSM591439 as well as E-MTAB-4729/ERR1406031 and ERR1406032.
Affymetrix Human Gene 1.0 ST microarray was normalised using Single-Channel Array 68 and exon expression was summarised to the gene level using the R/Bioconductor package pd.hugene.1.0.st.v1 (version 3.14.1). Transcripts were considered to be present in the cell if they have an absolute expression value > 0. RNAseq data were pseudoaligned to the cDNA transcript sequences Ensembl version 87 and quantified using kallisto (Version 0.43.1) 69 . Transcripts abundance was summarised to the gene level via the R library tximport 70 . Genes were considered to be expressed, if they have a minimal transcript per million value > 1.

Data availability
All data supporting the findings of this study are available within the article and its supplementary information files or from the corresponding author upon reasonable request Received: 30 August 2016 Accepted: 25 July 2018