A heterometallic ruthenium–gold complex displays antiproliferative, antimigratory, and antiangiogenic properties and inhibits metastasis and angiogenesis-associated proteases in renal cancer
Heterobimetallic compounds are designed to harness chemotherapeutic traits of distinct metal species into a single molecule. The ruthenium–gold (Ru–Au) family of compounds based on Au–N-heterocyclic carbene (NHC) fragments [Cl2(p-cymene)Ru(μ-dppm)Au(NHC)]ClO4 was conceived to combine the known antiproliferative and cytotoxic properties of Au–NHC-based compounds and the antimigratory, antimetastatic, and antiangiogenic characteristic of specific Ru-based compounds. Following recent studies of the anticancer efficacies of these Ru–Au–NHC complexes with promising potential as chemotherapeutics against colorectal, and renal cancers in vitro, we report here on the mechanism of a selected compound, [Cl2(p-cymene)Ru(μ-dppm)Au(IMes)]ClO4 (RANCE-1, 1). The studies were carried out in vitro using a human clear cell renal carcinoma cell line (Caki-1). These studies indicate that bimetallic compound RANCE-1 (1) is significantly more cytotoxic than the Ru (2) or Au (3) monometallic derivatives. RANCE-1 significantly inhibits migration, invasion, and angiogenesis, which are essential for metastasis. RANCE-1 was found to disturb pericellular proteolysis by inhibiting cathepsins, and the metalloproteases MMP and ADAM which play key roles in the etiopathogenesis of cancer. RANCE-1 also inhibits the mitochondrial protein TrxR that is often overexpressed in cancer cells and facilitates apoptosis evasion. We found that while auranofin perturbed migration and invasion to similar degrees as RANCE-1 (1) in Caki-1 renal cancer cells, RANCE-1 (1) inhibited antiangiogenic formation and VEGF expression. We found that auranofin and RANCE-1 (1) have distinct proteolytic profiles. In summary, RANCE-1 constitutes a very promising candidate for further preclinical evaluations in renal cancer.
KeywordsBimetallic Ruthenium–gold Renal cancer Mechanisms Auranofin
American Type Culture Collection
A disintegrin and metalloproteinase
Human clear cell renal cell carcinoma
- HCT 116
Human colorectal carcinoma
Human umbilical vein endothelial cells
Human foetal lung fibroblast
Vascular endothelial growth factor
Heterometallic compounds have been developed to enhance the anticancer properties of single metallodrugs. The hypothesis is that the incorporation of two different biologically active metals in the same molecule may improve their antitumor activity as a result of metal specific interactions with distinct biological targets (cooperative effect) or by the improved chemicophysical properties of the resulting heterometallic compound (synergism) . Recent reports have highlighted the success of this approach [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28], although there are very few articles documenting comparisons of the heteronuclear compounds with the monometallic fragments (alone or in combination) [1, 3, 15, 23, 24, 25, 26, 27]. Our group has developed heterometallic complexes containing Au(I) phosphane and Au(I) N-heterocyclic carbene moieties as potential cancer chemotherapeutics [23, 24, 25, 26, 27, 28]. We have established a number of titanocene-Au derivatives with high efficacy against ovarian and prostate cancers in vitro [23, 26] and renal cancer both in vitro  and in vivo . We and others have also recently reported on the design of heterometallic Ru–Au complexes with in vitro efficacy against HCT 116 colon cancer cell lines . We found, in most cases, a synergistic effect of the heterometallic compound when compared to its monometallic counterparts (either alone or in combination) [23, 24, 25, 26, 27]. Most recently, we reported on the synthesis and anticancer activity of family of Ru–Au cationic complexes incorporating Au–N-heterocyclic carbene ligands of the type Cl2(p-cymene)Ru(μ-dppm)Au(NHC)]ClO4 . These compounds displayed high stability in physiological conditions, notable cytotoxicity in renal and colon cancer cell lines, and cytotoxic selectivity . Preliminary mechanistic studies indicated that these compounds behave unlike cisplatin and more like other Au(I) derivatives containing lipophilic ligands such as phosphanes (e.g., auranofin) and N-heterocyclic carbenes . The compounds did not exhibit interaction with DNA but inhibited mitochondrial thioredoxin reductase in Caki-1 renal cancer cell lines .
Materials and methods
Auranofin was purchased from Strem and used without further purification. [Cl2(p-cymene)Ru(μ-dppm)Au(IMes)]ClO4 (RANCE-1, 1) , monometallic [Ru(p-cymene)Cl2(dppm-κP)] (2)  and [AuCl(IMes)] ANCE-1 (3)  were prepared as described previously.
Human renal cell carcinoma line Caki-1 was newly obtained for these studies from the American Type Culture Collection (ATCC) (Manassas, Virginia, USA) and cultured in Roswell Park Memorial Institute (RPMI-1640) (Mediatech Inc., Manassas, VA) media containing 10% Fetal Bovine Serum, certified, heat inactivated, US origin (FBS) (Gibco, Life Technologies, US), 1% Minimum Essential Media (MEM) nonessential amino acids (NEAA, Mediatech), and 1% penicillin–streptomycin (PenStrep, Mediatech). IMR90 (human foetal lung fibroblast) cells were purchased from ATCC (Manassas, Virginia, USA) and maintained in Dulbecco’s modified Eagle’s medium (DMEM) (Mediatech) supplemented with 10% FBS, 1% NEAA, and 1% PenStrep. HUVEC (human umbilical vein endothelial) cells were obtained from American Type Culture Collection (ATCC) and cultured in Medium 200PRF (Gibco, Life Technologies, US) and Low Serum Growth Supplement (LSGS) (Gibco, Life Technologies, US).
Cell viability analysis
The cytotoxic profile (IC50) of RANCE-1 (1) was determined by assessing the viability of Caki-1 and IMR90 control cells treated with the appropriate cultured medium containing 0.1 μM, 1 μM, 10 μM, and 100 μM of RANCE-1 (1) or the monometallic compounds R (2) and ANCE-1 (3) alone and in combination for 72 h using the colorimetric cell viability assay PrestoBlue (Invitrogen). The cytotoxic profile of auranofin and cisplatin (for comparative purposes) was also determined. All compounds were dissolved in DMSO except cisplatin that was dissolved in H2O with a final DMSO concentration of 1%.
Cell death assay
The cell cycle profile in Caki-1 cancerous cells cultured in the appropriate medium containing IC50 concentrations of RANCE-1 (1), and auranofin for 24 h was analysed by flow cytometry stained with Annexin V (FITC) dye (Invitrogen). Staurosporine- and ionomycin-treated cells were also stained with Annexin V dye and served as positive controls for apoptosis and cell death. FITC fluorescence intensity was detected with a flow cytometry analysis which was conducted using a BD LSR II flow cytometer. 10 × 105 events per sample were recorded. The flow cytometer was calibrated prior to use.
Cell cycle profile
The cell cycle profile in Caki-1 cancerous cells cultured in the complete RPMI medium containing IC20 concentrations of RANCE-1 (1) and auranofin for 24 h was analysed by flow cytometry wherein total DNA was stained with FxCycle Violet (FCV; DAPI) dye (Invitrogen). DAPI fluorescence intensity was detected with a BD LSR II flow cytometer and flow cytometry analysis was conducted using BD FACSDiva 8.0.2 10*105 events per sample were recorded. The flow cytometer was calibrated prior to use.
Cell migration and invasion analysis
The antimigratory profile of RANCE-1 (1), the monometallic compounds R (2), and ANCE-1 (3) and auranofin was assessed by in vitro scratch assay using Caki-1 cells treated with the appropriate cultured medium containing IC20 concentration of the compounds. The diluting agent (0.1% DMSO) served as positive control. Briefly, using 6-well collagen-coated plate cells were allowed to seed for 24 h. After which, cells were serum starved for 24 h, scratched by 200 μL tips. 24 h after injury, cells were photographed using a Moticam camera mounted on a Zeiss microscope inverted microscope at 20×. The area invaded was measured in five randomly selected segments from each photo then averaged. Data were collected from two independent experiments performed.
The antiinvasion profile of RANCE-1 (1), the monometallic compounds R (2) and ANCE-1 (3) and auranofin was assessed by transwell assay by determining the invasion of Caki-1 cancerous cells treated with the appropriate cultured medium containing IC20 concentration of the compounds. The diluting agent (0.1% DMSO) served as positive control. Briefly, Geltrex® a Reduced Growth Factor Basement Membrane Matrix (Invitrogen) is thawed and added to a 24-well transwell insert and solidified in a 37 °C incubator for 30 min to form a thin gel layer. Cell solution of 5 × 105 cells in serum-free media is added on top of the Geltrex® coating to simulate invasion through the extracellular matrix. In insert was mounted into well containing complete media. Invasion of cells toward the chemotactic gradient membrane to the underside was assessed after 24 h when the membrane is fixed and stained with haematoxylin and eosin. Then, the invaded cells were captured with a Moticam camera mounted on a Zeiss microscope inverted microscope at 20×. Cell numbers were counted from 5 randomly selected fields of view from each photo then averaged. Data were collected from at least two independent experiments performed.
The antiangiogenic profile of RANCE-1 (1) and auranofin was determined by assessing the endothelial tube formation of human umbilical vein endothelial cells (HUVECs) treated with the appropriate cultured medium containing the above mentioned compounds. Briefly, 96-well plates were coated with Geltrex®, Reduced Growth Factor Basement Membrane Matrix (Invitrogen) (50 μl/well) and incubated at 37 °C for 30 min to allow gelation to occur. HUVECs were added to the top of the gel at a density of 6000 cells/well in the presence or absence of the RANCE-1 or its monometallic controls (1 μM). The diluting agent (1% DMSO) served as positive control. Cells were incubated at 37 °C with 5% CO2 for 24 h and pictures were captured with a Moticam camera mounted on a Zeiss microscope inverted microscope at 10X. Quantification of tube formation was assisted by SCORE, a web-based image analysis system (S.CO BioLifescience). Tube formation quantified by number of branching points (tube nodes, TN) and total length skeleton (tube length, TL). The data were obtained from the average of three wells per treatment condition.
Thioredoxin reductase activity assay
Caki-1 cells treated in vitro with IC50 concentrations of either RANCE-1 (1), the monometallic compounds R (2), and ANCE-1 (3) and auranofin or 0.1% DMSO were lysed after 24 h of treatment. The lysate was then mixed with Thioredoxin Reductase Assay buffer; after 20 min incubation DTNB (3,3′-Disulfanediylbis(6-nitrobenzoic acid)) was added; TrxR levels were detected according to the manufacturer’s instructions (Abcam kit, ab83463) using the BioTek Microplate Reader (BioTek U.S., Winooski, VT) at λ = 412 nm. Tests were done in duplicate. TrxR activity was calculated based on the linear amount of TNB produced per mg of total protein and adjusted for background activity from enzymes other than TrxR in the lysates.
Caki-1 cells treated in vitro with either IC20 concentration of RANCE-1 (1), auranofin or 0.1% DMSO were lysed after 72 h of treatment. Before application to the array, protein concentration was determined by BCA. Then, 150 μg of lysate was incubated for 24 h with the Proteome Profiler Human Protease Array Kit (ARY025, R&D Systems). The relative expression levels of the proteases were determined according to the manufacturer’s protocol, and signal intensities were compared using HLImage++ software (R&D).
Caki-1 cells treated in vitro with either IC20 concentrations of RANCE-1 (1), auranofin or 0.1% DMSO were lysed after 72 h of treatment, cell culture supernatant was collected and interleukin expression was determined by Multi-Target ELISA array kit (PathScan Cytokine Antibody Array Kit, Cell Signalling). The relative expression levels of the proteases were determined according to the manufacturer’s protocol, and signal intensities were compared using HLImage++ software (R&D).
Caki-1 cells treated in vitro with either IC20 concentrations of RANCE-1 (1), auranofin or 0.1% DMSO were lysed after 72 h of treatment, cell culture supernatant was collected and VEGF expression was assessed by a VEGF human ELISA kit (100663 Abcam).
Results and discussion
Cytotoxicity, selectivity, cell death, and cell cycle arrest
Cell viability IC50 values (μM) in Caki-1 cells and IMR-90 fibroblasts for bimetallic RANCE-1 (1) and monometallic Ru R (2) Au ANCE-1 (3) compounds. Auranofin and cisplatin were used as controls
RANCE-1 (1) derivative is more cytotoxic to the Caki-1 cells than cisplatin, monometallic ruthenium R (2), and gold ANCE-1 (3) compounds. Importantly, RANCE-1 (1) is considerably less toxic to the nontumorigenic IRM-90 than cisplatin and ruthenium R (2) which makes RANCE-1 more selective. While auranofin is also cytotoxic with a low IC50 value in Caki-1 cells it is only moderately selective. We studied the effect of the combination of monometallic ruthenium R (2) and gold ANCE-1 (3) compounds (1:1 equivalents) on Caki-1 using the same conditions as for the bimetallic compound. The resulting IC50 values were larger than those of RANCE-1 (1), while remaining selective. This fact supports the idea that there is, indeed, a synergistic effect for RANCE-1 (1) on the renal cancer cells as described for other Ru-Au and Ti–Au compounds described by our group [25, 26, 27, 28]. In view of the results obtained, we decided not to explore the effects of cisplatin in Caki-1 cells any further.
Next (Fig. 3), we evaluated the effects of most cytotoxic RANCE-1 (1), and Auranofin for cell cycle arrest. We observed that cells treated with RANCE-1 (1) had the greatest percentage of cells in Sub G1 (26%) and more G1/G0 (37%) the fewest cells in S phase (16%), and similarly few cells in G2/M (20%). Similarly, for other p-cymene Ru derivatives such as RAED and RAPTA-C, it was reported that greatest percentage of cells was in G1/G0 (60%) and fewest cells in S phase (ca. 15%) and G2/M (5%) for ovarian cancer A278 cell lines . Auranofin-treated cells were almost exclusively in G0/G1 (82%), suggesting that the compound induced complete G1/G0 arrest, as has already been reported for AF in other cell lines [35, 36, 37, 38], and also in cancer cells treated with antiinflammatory drugs .
Inhibition of migration and invasion
RANCE-1 was observed to inhibit invasion in a 3D Transwell assay fitted with Geltrex® matrix an extracellular matrix analogue (Fig. 4b). RANCE-1 (1) reduced invasion by 66%, as we observed with migration the monometallic auranofin was an excellent inhibitor of invasion (54% of invasion is inhibited). The inhibition of invasion by the Ru monometallic compound R (2) at 72 % was quite robust (as previously described for another cancer cell line) , higher even than that of RANCE-1 or auranofin. The monometallic Au compound ANCE-1, on the other hand, only inhibited 18% of migration in agreement with the data obtained in the 2D scratch assay.
Inhibition of angiogenesis
The ruthenium containing derivatives RANCE-1 (1) and R (2) were observed to inhibit all tube formation in a 3D assay on Geltrex® matrix an extracellular matrix analogue (Fig. 5). The gold monometallic compounds ANCE-1 (3) and auranofin cause mild disturbances in tube arrangement, but there is viable tube formation. This indicates that the ruthenium component of the compound may be responsible for the antiangiogenic properties. However, RANCE-1 displays an impressive antiangiogenic effect (higher than that of the monometallic compound R (2)).
Inhibition of targets associated with cancer progression, cisplatin resistance, metastasis, and angiogenesis
Inhibition of thioredoxin reductase
Inhibition of VEGF
Vascular endothelial growth factor (VEGF) is the key mediator of angiogenesis in cancer, where it is up-regulated by oncogene expression, a variety of growth factors [47, 48, 49]. Before most solid tumors can grow beyond past the 0.5 cm diameter, they require blood vessels for nutrients and oxygen [39, 48, 49]. The production of VEGF and other growth factors by the tumor results in the ‘angiogenic switch’, where new vasculature is formed in and around the tumor, promoting exponential growth . In addition, ruthenium compounds have been reported to inhibit VEGF . Given the key tumor promoting properties of VEGF, we set out to evaluate the inhibitory effects of the RANCE-1 (1), the monometallic ruthenium R (2), and gold ANCE-1 (3) compounds and Auranofin on VEGF, an attractive target in cancer therapy. We found that VEGF secretion is significantly inhibited by both RANCE-1 (1) and auranofin after 72 h of incubation (70% and 55% reduction, respectively), while R (2) and ANCE-1 (3) lead to no notable inhibition of VEGF secretion (Fig. 6b). Angiogenesis is a complex process driven by diverse activities and an intricate sequence of factors. While VEGF is known to be a key regulator of angiogenesis and its downregulation is often correlated with reduced angiogenesis, such correlation is not absolute. As our data indicate while there is evident inhibition of tube formation by RANCE-1 and significantly less inhibition of tube formation by auranofin, they are both potent inhibitors of VEGF, this suggest that there might be more at play in RANCE-1 inhibition of VEGF and its effect on tube formation.
The inhibition of VEGF-2 has been recently reported for a ruthenium (III) macrocyclic cationic compound described by Che et al. which inhibited MS-1 endothelial and HUVEC tube formation (a model for angiogenesis) and inhibited angiogenesis in chicken chorioallantoic membrane (CAM) .
Inhibition of interleukins
Inflammatory cytokines such as interleukins (IL) are known to be associated with malignant progression of breast and lung cancers [50, 52, 53]. IL-5, IL-6, IL-8, IL-12, and IL-17A are of particular relevance as their elevated expression has been detected in multiple epithelial tumors and has been shown to be associated with tumor metastasis [52, 53]. ILs also play an important role in tumor promotion and metastasis through various proteolytic interaction and through control of matrix metalloproteinases (MMP) expression and the expression of angiogenic proteins growth factors such as VEGF . We studied the inhibition of ILs by selected most active compounds bimetallic RANCE-1 (1) and auranofin after 72 h of incubation at IC20 concentrations (Fig. 6c). We found that RANCE-1 (1) inhibited IL-6 expression by 60%. RANCE-1 (1) inhibited IL-5 expression by 37% and had no effect on IL-13 expression, and resulted in a complete inhibition of IL-17A expression. The inhibition of IL-17 is of great clinical interest, because increased IL-17A expression is associated with ER(−) and triple-negative tumor hyper proliferation and poor prognosis in breast cancer. IL-17 is also known to drives several pathogenic processes during breast cancer progression to metastasis; IL-17 not only promotes tumor cell survival and invasiveness, it also contributes to the promotion of tumor angiogenesis. Auranofin, as previously reported, inhibits IL-6, but also completely inhibits expression of IL-5, IL-8, and IL-13, all key players in inflammatory signalling . Given the potent inhibition of IL expression observed upon treatment with RANCE-1 (1), this compound might be a good candidate for combination therapy with a more cytotoxic agent. The great value of IL inhibitors is further increased, because ILs are known inducer of MMPs which are critical in metastasis [50, 55].
Inhibition of metalloproteases
Matrix metalloproteinases (MMPs) are a multigene family of zinc-dependent extracellular proteins. MMP are remodelling endopeptidases implicated in pathological processes, such as carcinogenesis [50, 55, 56, 57].
Accumulating evidence demonstrates the crucial role of proteolytic enzymes such ADAMs, which are closely related to MMPs facilitate tumor cell invasion and metastasis. In particular, ADAM-9 is reported to be highly expressed in invasive renal cell cancer, while ADAM8 is considered a robust hemo-histo-chemical marker for lung cancer and correlated with poor prognosis for patients with pancreatic ductal adenocarcinoma [60, 61]. From our study, we observed that treatment with RANCE-1 (1) ablated all expression of ADAM-8, and inhibited ADAM-9 expression by 37%, while auranofin had no effect on the expression levels of those two proteases (Fig. 7b). Such inhibition is particularly robust and in combination with the VEGF and IL(s) secretion’s inhibition that RANCE-1 (1) displays, this compound presents a molecular target profile that would potentially make it a very efficient antimetastatic and antiangiogenic agent.
Inhibition of cathepsin proteases
Cysteine cathepsin proteases (Cts) are known regulators of cancer progression and therapeutic response. Some members of the cathepsin family are highly expressed in metastatic tumors, including CatsB, CtsL, CtsS, and CtsX [62, 63, 64]. Cathepsin B is of significant importance to cancer therapy as it is involved in various pathologies and oncogenic processes in humans [50, 63, 64, 65]. Therefore, a pan-cathepsin inhibitor is of great clinical interest. Some ruthenium p-cymene compounds (RAPTA family) have been shown to be inhibitors of cathepsin B and this inhibition has been correlated to their antimetastatic properties . We reported on the inhibition of cathepsin B (purified) by a neutral ruthenium–gold compound . We observed that RANCE-1 (1) ablated CtsB and CtsD expression (100% suppression), while inhibiting CtsL, CtsS, and CtsX/Z by 38, 35, and 53%, respectively, while auranofin had no inhibitory effect on the expression of the five Cts evaluated.
Therefore, our studies suggest that bimetallic ruthenium–gold compound RANCE-1 (1) is a potent pan-MMP and pan-cathepsin inhibitor with notable inhibition of members of the ADAM protease family. This inhibition may lay the mechanism by which RANCE-1 (1) inhibits tumor growth, invasion, and denovo vasculature.
To summarize, this study demonstrates the synergism of a heterometallic Ru-Au (RANCE-1, 1) compound designed to harness the cytotoxic and apoptotic effects of Au(I) lipophilic cations as well as their TrxR inhibition properties with the potential antimetastatic effects of a Ru p-cymene derivative containing a phosphine. In addition to be cytotoxic and apoptotic, RANCE-1 (1) displays relevant inhibition of migration, invasion, and angiogenesis, while also inhibiting molecular pathways associated with these processes. The molecular targets inhibited include different interleukins, metalloproteases, and cathepsins, which are involved in tumor metastasis and angiogenesis, and to an even higher degree the angiogenic factor VEGF (also involved in angiogenesis). It is noteworthy that the inhibition observed is in general better than that of the individual monometallic fragments present in the heterometallic compound. In some cases, this inhibition can be correlated with a particular metallic fragment of the bimetallic compound (reinforcing the idea of the positive synergistic effect caused by the two distinct metals).
During this study, we evaluated auranofin and observed it had a very similar effect to RANCE-1 (1) in the renal cancer cell line Caki-1 in terms of its antiproliferative and antimetastatic properties and some of the sets of targets inhibited. RANCE-1 (1), however, was a better VEGF inhibitor than auranofin and a much better pan-MMP and pan-cathepsin inhibitor. Moreover, RANCE-1 (1) block all angiogenic formation, while Auranofin merely induced branching disturbances of de novo angiogenesis in an in vitro model.
These results are very relevant in the search for multi-targeted therapies that would best avert target-specific induced resistance, and thus warrant further evaluation of the efficacy and mechanism of RANCE-1 (1) in renal cancer in vivo. In addition, the relevant results found for auranofin (an old antirheumatic drug being currently ‘repurposed’ as chemotherapeutic for different diseases including cancer) also warrant further exploration of this agent in renal cancer treatment.
We thank the National Cancer Institute (NCI) for grant 1SC1CA182844 (M.C.) and Dr. Karen Hubbard for providing infrastructure and advice. We are grateful to Dr. Natalia Curado and Yu-Fung Mui for their assistance in the preparation of the metallic compounds, Mike A. Cornejo for his assistance with some IC50 assays, and Ciara Bagnall for her technical expertise in imaging and immunocytochemistry.
- 25.Fernández-Gallardo J, Elie BT, Sadhukha T, Prabha S, Sanaú M, Rotenberg SA, Ramos JW, Contel M (2015) Chem Sci 6: 5269-5283. (b) Contel M, Fernández-Gallardo J, Elie BT, Ramos JW. US Patent 9,315,531 (04/19/2016)Google Scholar
- 37.Fan C, Zheng W, Fu X, Li X, Wong Y-S, Chen T (2014) Cell Death and Disease 5:e191Google Scholar
- 42.DeCicco-Skinner KL, Henry GH, Cataisson C, Tabib T, Gwilliam JC, Watson NJ, Bullwinkle EM, Falkenburg L, O’Neill RC, Morin A, Wiest JS (2014) J Vis Exp 91:e51312Google Scholar
- 53.Welte T, Zhang XH (2015) Mediators Inflamm 2015:804347. https://doi.org/10.1155/2015/804347
- 59.Woessner JF, Nagase H (2000) Matrix metalloproteinases and TIMPs: protein profile. Oxford University Press, OxfordGoogle Scholar
- 61.Schlomann U, Koller G, Conrad C, Ferdous T, Golfi P, Garcia AM, Höfling S, Parsons M, Costa P, Soper R, Bossard M, Hagemann T, Roshani R, Sewald N, Ketchem RR, Moss ML, Rasmussen FH, Miller MA, Lauffenburger DA, Tuveson DA, Nimsky C, Bartsch JW (2015) Nat Commun 6:6175CrossRefPubMedPubMedCentralGoogle Scholar