Hypoxia Induces Macrophage tnfa Expression via Cyclooxygenase and Prostaglandin E2 in vivo

Macrophage phenotypes are poorly characterized in disease systems in vivo. Appropriate macrophage activation requires complex coordination of local microenvironmental cues and cytokine signaling. If the molecular mechanisms underpinning macrophage activation were better understood, macrophages could be pharmacologically tuned during disease situations. Here, using zebrafish tnfa:GFP transgenic lines as in vivo readouts, we show that physiological hypoxia and stabilization of Hif-1α promotes macrophage tnfa expression. We demonstrate a new mechanism of Hif-1α-induced macrophage tnfa expression via a cyclooxygenase/prostaglandin E2 axis. These findings uncover a macrophage HIF/COX/TNF axis that links microenvironmental cues to macrophage phenotype, with important implications during inflammation, infection, and cancer, where hypoxia is a common microenvironmental feature and where cyclooxygenase and TNF are major mechanistic players.


INTRODUCTION
Innate immune activation during homeostasis and disease are tightly regulated, in part by the coordination of microenvironmental cues and cytokine signaling. Macrophages are important innate immune cells in disease and their activation status, commonly termed polarization, is especially important during the response to injury and infection. Mammalian macrophages have been classified into two broad polarization states: M1 (or classically activated) and M2 (alternatively activated) (1,2). M1 macrophages are highly antimicrobial and can phagocytose and efficiently kill bacteria. M2 macrophages are central players in tissue healing and restoration of homeostasis post injury/infection (3,4). M1/M2 polarization states have largely been determined in vitro by addition of exogenous factors, e.g., cytokines, to the growth media, however the complex regulation of the diverse macrophage phenotypes observed in vivo remains poorly understood. It is only in whole organism in vivo models, where tissues remain intact, that microenvironmental cues of macrophage activation can be fully investigated.
Leukocytes, including macrophages and neutrophils, are exquisitely sensitive to tissue hypoxia, a signature of many disease microenvironments (e.g., in inflammatory diseases, infections, and cancers) due to lack of local blood supply and a high turnover of local oxygen by pathogens (5,6). Tissue hypoxia can polarize macrophages, via stabilization of the Hypoxia Inducible Factor (HIF)-1α transcription factor (7). HIF-1α is a master transcriptional regulator of the cellular response to hypoxia (8) and has been shown to have activating effects on macrophages in vitro, increasing their pro-inflammatory cytokine profile and bactericidal capabilities, yet the molecular mechanisms behind these observations are not well-understood in vivo (6).
Eicosanoids are lipid signaling molecules and important pro-inflammatory mediators during homeostasis and disease (9,10). For example, tissue hypoxia increases cyclooxygenase (COX)/prostaglandin E2 (PGE2) production in epithelial cells in cancer models (11)(12)(13). Recent evidence indicates that hypoxia-induced tumor necrosis factor alpha (TNF) expression in osteoblasts (bone producing cells) is mediated through cyclooxygenase enzymes by an, as yet, unknown mechanism (14). This HIF/COX/TNF axis has not, to date, been observed in macrophages, and could be a potential novel pathway to macrophage activation in diseases where hypoxia is a microenvironmental factor.
Here, we investigated the regulation of expression of the important, primarily pro-inflammatory, cytokine tnfa by Hif-1α stabilization in vivo using optically transparent zebrafish larvae. We demonstrate that genetic, pharmacological, and hypoxia-mediated Hif-1α stabilization upregulated macrophage tnfa. We utilize well-characterized, non-invasive, zebrafish models of inflammation and infection to show that Hif-1αmediated tnfa acts via an alternative, cyclooxygenase dependent mechanism, that differs from better understood DAMP/PAMP mediated pathways that are cyclooxygenase independent (15,16). We show that the Hif-1α-mediated tnfa acts via active PGE2. Furthermore, this new macrophage HIF/COX/TNF axis is found in primary human macrophages. These findings have important implications in inflammation, infection, and cancer biology where macrophage phenotypes are influenced by microenvironmental hypoxia, and where HIF, COX, and TNF are major mechanistic players.

Tailfin Injury
Inflammation was induced in zebrafish embryos by tail transection at 2 days post-fertilization (dpf). Larvae were anesthetized and the tailfin was transected using a sterile scalpel blade (23). Larvae were imaged by confocal microscopy at 16 h post-wounding (hpw) on a Leica TCS-SPE confocal on an inverted Leica DMi8 base and imaged using 20× or 40× objective lenses.

Confocal Microscopy of Transgenic Larvae
Larvae were mounted in 0.8-1% low melting point agarose (Sigma-Aldrich, A9414) and imaged on a Leica TCS-SPE confocal on an inverted Leica DMi8 base and imaged using 20× or 40× objective lenses.
For quantification purposes acquisition settings and area of imaging were kept the same across groups. Corrected total cell fluorescence was calculated for each cell using Image J by drawing around the cell and correcting for background levels of staining (27,28).

RNA Injections
Embryos were injected with dominant hif-1αb (ZFIN: hif1ab) variant RNA at the one cell stage as previously described (23). hif-1α variants used were dominant active (DA) that is a constitutively stable Hif-1α and dominant negative (DN) hif-1α that cannot signal due to lack of transactivation domain. Phenol red (PR) (Sigma-Aldrich, P0290) was used as a vehicle control.

Bacterial Pixel Count
Mm mCherry infected zebrafish larvae were imaged at 4 days post-fertilization (dpi) on an inverted Leica DMi8 with a 2.5× objective lens. Brightfield and fluorescent images were captured using a Hammamatsu OrcaV4 camera. Bacterial burden was assessed using dedicated pixel counting software as previously described (29).

Pathway Inhibitors
Unless otherwise stated, embryos were treated from 4 h pre-Mm infection to 24 h post-infection (hpi) by addition to the embryo water in six well plates and dimethyl sulfoxide (DMSO, Sigma-Aldrich, D8418), was used as a negative solvent control. The pan hydroxylase inhibitor, DMOG (dimethyloxaloylglycine, Enzo Life Sciences, BML-EI347-0010), was used at a 100 µM concentration by incubation in E3 embryo media (23). The selective PHD inhibitors JNJ-402041935 (Cayman Chemicals,

Hypoxia Hood
Embryos were incubated in 5% oxygen (with 5% carbon dioxide) in a hypoxia hood (Baker-Ruskinn Sci-tive UM-027) from 32 hpi for 6 h and were imaged at 2 dpf. Embryos from the same clutch kept in incubated normoxic room air were used as controls (37).
Cell supernatants were collected following 18 h incubation and were assayed in triplicate using a human TNF ELISA MAX (Biolegend, 430204).

Statistical Analysis
All data were analyzed (Prism 7.0, GraphPad Software) using unpaired, two-tailed t-tests for comparisons between two groups and ANOVA (with Bonferonni post-test adjustment) for other data. P-values shown are: * P < 0.05, * * P < 0.01, and * * * P < 0.001.

Hypoxia and Hif-1α Stabilization Upregulate tnfa:GFP Expression in vivo
Tumor necrosis factor (TNF) is a central regulator of macrophage pro-inflammatory phenotypes during homeostasis, inflammation, and infection. Here, we use two transgenic zebrafish lines that express GFP under the control of a tnfa promoter region; TgBAC(tnfa:GFP)pd1028 and Tg(tnfa:eGFP-F)ump5Tg (18,20). The TgBAC(tnfa:GFP)pd1028 is a BAC (bacterial artificial chromosome) transgenic line that contains 50 kb of promoter region (18). The Tg(tnfa:eGFP-F)ump5Tg line has a smaller promoter region (3.8 kb) than the BAC line (20) and has previously been demonstrated to be upregulated in macrophages after tailfin injury in a zebrafish model (20). Here, we demonstrate that injury-induced upregulation of macrophage tnfa:GFP is also found in the BAC transgenic line at 16 h post-wound (hpw), shown in double transgenic TgBAC(tnfa:GFP)pd1028; Tg(mpeg1:mCherryCAAX)sh378 larvae ( Figure 1A).
TNF has been implicated in many bacterial infections, however the cell types involved have been difficult to observe in vivo. To investigate this we used a well-characterized zebrafish infection model, the M. marinum (Mm) model of tuberculosis infection (40). We used the tnfa BAC promoter P-values shown are: *P < 0.05, **P < 0.01, and ***P < 0.001, two way ANOVA with Bonferonni post-test adjustment. (E) TNF ELISA of human monocyte derived macrophages treated with LPS and incubated in normoxia or 0.8% hypoxia with or without treatment with NS398. LPS negative controls were performed but TNF produced in these groups was below detectable levels. Mean ± SEM, n = 5-8 biological repeats from 3 to 4 donors. P-values shown are: *P < 0.05, **P < 0.01, and ***P < 0.001, two way ANOVA with Bonferonni post-test adjustment. (F) Schematic of the arachidonic pathway indicating that stabilizing Hif-1α upregulates tnfa, an effect that is blocked using the Cox1/2 inhibitors SC560/NS398.
GFP line to establish the transcriptional regulation of tnfa during pre-and early-larval Mm granuloma formation stages. tnfa:GFP expression was induced in larvae following Mm infection before granuloma onset (at 1 day post-infection, dpi), and during granuloma formation (at 4 dpi) ( Figure 1B). tnfa:GFP expression was predominantly found in macrophages, shown by co-localization of fluorescence with mpeg-1:mCherry expressing macrophages (Figure 1C). Macrophage expression was confirmed using the other tnfa promoter driven line, Tg(tnfa:eGFP-F)ump5Tg; Tg(mpeg1:mCherry-F)ump2Tg ( Figure S1). Our data demonstrate that injury and Mm induced tnfa expression occurs in macrophages as part of an early response.
Hypoxia and Hif-1α stabilization are key features of diseased tissue microenvironments that can trigger macrophage pro-inflammatory responses. We first tested whether Hif-1α promotes tnfa in vivo, by stabilizing Hif-1α genetically in the tnfa:GFP BAC transgenic line (18,20). Dominant active (DA) Hif-1α RNA (23) resulted in upregulation of macrophage tnfa:GFP expression (Figures 1D-F) indicating a shift of macrophage phenotype toward a pro-inflammatory response. Dominant negative (DN) Hif-1α caused no change in tnfa:GFP expression (Figures 1E,F).
Finally, we subjected the tnfa:GFP line to physiological hypoxia (5% oxygen for 6 h at 32 hpf) and looked for GFP expression at 48 hpf. This level of hypoxia was sufficient to turn on phd3:GFP expression in Tg(phd3:GFP)i144 hypoxia reporter zebrafish (Figure S5) (37). Hypoxia induced tnfa:GFP expression, compared to normoxic controls at 48 hpf, to a similar level as that observed with genetic or pharmacological Hif-1α stabilization ( Figure 1J).
Together, these data indicate that tnfa expression is induced in macrophages in response to hypoxia and Hif-1α, a response that is targetable by pharmacological agents and has the potential to manipulate macrophage responses during disease.

Hif-1α Dependent tnfa:GFP Transcription Requires Cyclooxygenase
Eicosanoids are lipid signaling molecules and can be produced by macrophages as part of a pro-inflammatory response (9, 10). We tested whether cyclooxygenase inhibition affected Hif-1αmediated macrophage tnfa:GFP expression, using Cox-1 and Cox-2 inhibitors (SC560 and NS398, respectively) (32). First, we showed that neither SC560 nor NS398 caused increased Hif-α transcriptional activity, using the Tg(phd3:GFP)i144 hypoxia reporter zebrafish (Figure S6). In the absence of Hif-1α stabilization tnfa:GFP was not altered by SC560 or NS398 compared to DMSO treated negative controls (Figures 2A,B). Strikingly, the DA Hif-1α-induced tnfa:GFP expression was reduced to basal levels by both SC560 and NS398 (Figures 2A,B). These findings were replicated using the hydroxylase inhibitors FG4592 (Figures 2C,D) and DMOG with NS398 ( Figure S7), (N.B. DMOG with SC560 caused toxicity and developmental delay therefore data is not shown for this group). The production of an unrelated Hif-1α target gene phd-3 by hydroxylase inhibitors, shown in the Tg(phd3:GFP)i144 transgenic line, was not altered by SC560 or NS398 treatment, demonstrating that Hif-1α transcriptional potential is not affected by cyclooxygenase inhibition (Figure S8).
We tested whether HIF-1α stabilization induced TNF in human monocyte derived macrophages (hMDMs). We found that human monocyte derived macrophages produced higher levels of TNF protein in hypoxia (0.8% oxygen) than those in normoxia, measured by an anti-human TNF ELISA ( Figure 2E). Furthermore, treatment with the COX-2 inhibitor, NS398, reduced this hypoxia-induced TNF to equivalent levels found in normoxia (Figure 2E). This was replicated when HIF-1α was stabilized in hMDMs using the hypoxia mimetic FG4592 ( Figure S9). These data indicate that a HIF/COX/TNF axis is found in zebrafish and human macrophages, and could be a druggable target in disease situations ( Figure 2F).
We have previously shown that DA Hif-1α induces the expression of another important pro-inflammatory cytokine, il-1β, shown by an il-1β:GFP transgenic line ( Figure S10A) (37).  Cox-1 inhibition by SC560 did not abrogate Hif-1α-induced il-1β:GFP, while Cox-2 inhibition with NS398 caused a small decrease that did not reach basal levels (Figures S10B,C). These data indicate that Hif-1α-induced tnfa expression is caused by a product of the cyclooxygenase arm of the arachidonic acid pathway via a novel macrophage HIF/COX/TNF axis, while Hif-1α-induced il-1β does not fully act via this pathway suggestive of complex regulation of inflammatory cytokines by Hif-1α stabilization. We next tested whether injury-and infection-induced tnfa are Hif-1α and cyclooxygenase dependent processes. Induction of tnfa:GFP expression after injury was not blocked by dominant negative Hif-1α (Figures 3A,B), and this was also the case for Mm infection-induced tnfa:GFP expression (Figures 3C,D). Similarly, neither genetic nor pharmacological Hif-1α stabilization were additive to the tnfa:GFP expression caused by either injury (Figures 3E,F) or Mm infection (Figures 3G,H; (Figures S11A,B). These data indicate that the early tnfa response of injury and Mm infection do not require active Hif-1α. Macrophage tnfa:GFP expression induced after injury was not abrogated by cyclooxygenase inhibition using NS398 (Figures 3E,F). Similarly, cyclooxygenase inhibition using either SC560 or NS398 did not alter the expression of Mminduced tnfa:GFP (Figures 3G,H). SC560 and NS398 mediated inhibition of Hif-1α-induced tnfa expression did not diminish the host-protective effect of DA Hif-1α (Figures S12A-C) nor DMOG treated larvae (Figures S12D-F), suggesting that the presence of Mm induced tnfa is sufficient for protection. Our data indicate that Hif-1α-induced macrophage tnfa expression via cyclooxygenase differs from DAMP-(injury) or PAMP-(Mm) induced tnfa mechanisms.

Blocking Cyclooxygenase Independent Arachidonic Acid Pathways Do Not Abrogate Hif-1α Upregulation of tnfa:GFP
To investigate whether the effect of the cyclooxygenase inhibitors on tnfa was specific to the prostaglandin path of arachidonic acid signaling, we targeted the lipoxin and leukotriene producing arms using the 15-Lipoxygenase inhibitor PD146176 and leukotriene B4 receptor antagonists ( Figure 4A) (33,34). The 15-Lipoxygenase inhibitor PD146176 did not block the tnfa:GFP induced by DA Hif-1α (Figures 4B,C). Furthermore, PD147176 increased tnfa:GFP expression on its own in the absence of infection, although not to the same extent as DA Hifα (Figures 4B,C). PD146176 also did not affect tnfa:GFP expression after Mm infection (Figures 4B,C), nor did it block the protective effect of DA Hif-1α. Treatment with PD146176 was sufficient to decrease Mm burden, but not to the same extent as DA Hif-1α ( Figure 4D). Leukotriene B4 inhibition, using the BLTR1/2 antagonists U75302 and LY255283, did not increase tnfa:GFP levels and also did not block DA Hif-1α-induced tnfa:GFP (Figures 4E,F). These data indicate that blocking components of the lipoxygenase dependent arms of the arachidonic acid pathway does not block the Hif-1α effect on tnfa expression and do not replicate the effects observed by blocking the cyclooxygenase dependent arm.

Hif-1α-Induced tnfa:GFP Requires Active Prostaglandin E2
A key family of immune regulators downstream of arachidonic acid and cyclooxygenases are prostaglandins. The best characterized of these as a regulator of macrophage function is prostaglandin E2 (PGE2) (41). We tested whether PGE2 was a mediator in the HIF/COX/TNF pathway by addition of exogenous PGE2 to DA Hif-1α larvae (35). Exogenous PGE2 had no effect on tnfa:GFP expression in the absence of infection (Figures 5A,B). However, PGE2 was able to rescue the decrease in tnfa:GFP expression after inhibition of either Cox-1 ( Figure S13) or Cox-2 (Figures 5A,B) in DA Hif-1α larvae. Furthermore, addition of PGE2 alone was sufficient to increase the DA Hif-1α-induced tnfa:GFP expression (Figures 4A,B). These tnfa:GFP activating effects were not observed by addition of exogenous 15-keto prostaglandin E2, an immunologically inactive degradation product of PGE2 (Figures 5C,D) (35,42,43). These data indicate that Hif-1α-induced tnfa:GFP requires active prostaglandin E2.
PGE2 signals via prostanoid receptors. In human and zebrafish there are four different receptors, EP1-4 (44). Of these, EP4 is the best characterized in immune cells and is expressed on human macrophages. When EP4 is blocked, macrophage proinflammatory cytokine release is decreased (45). We therefore tested a previously published EP4 antagonist AH23848 (36). We first demonstrated that AH23848 did not activate Hif-1α transcription using the Tg(phd3:GFP) hypoxia reporter line ( Figure S6). The increased level of phd3:GFP induced by hydroxylase inhibitors was also not altered by AH23484 treatment, demonstrating that Hif-1α transcriptional potential is not affected by EP4 inhibition (Figure S8). Addition of AH23848 abrogated dominant active Hif-1α-induced tnfa:GFP, indicating that Hif-1α stabilization acts through PGE2 signaling via EP4 to upregulate Tnfa expression (Figures 5E,F).

DISCUSSION
Control of macrophage function during homeostasis and disease are critical for maintaining healthy tissues and must integrate changes in the local microenvironment with cytokine signaling. Understanding of the signaling pathways that link microenvironment with macrophage phenotypic outcomes may identify novel therapeutic avenues for control of macrophages during disease.
Here, we show that a disease relevant microenvironmental cue, hypoxia signaling via Hif-1α, upregulates macrophage tnfa expression in a cyclooxygenase dependent. TNF upregulation by hypoxia has been previously demonstrated in a range of mammalian cells, both directly, through HIF responsive elements (HREs) in its promotor region, and indirectly (46)(47)(48)(49). Here, we addressed hypoxia-induced tnfa levels using promoter driven GFP expression in two zebrafish transgenic lines as a proxy for endogenous tnfa expression (18,20). Changes in endogenous tnfa expression on a wholebody-level after Hif-1α stabilization were not observable by qPCR (data not shown), presumably because the changes in transcription at the macrophage level are too small to be detected using these methods due to macrophages representing a small proportion of the zebrafish larvae, a challenge that was previously reported in wholebody detection of another pro-inflammatory cytokine, il-1β (37). These promoter driven tnfa transgenic lines have therefore opened up opportunities to study tnfa induction in situ in the in vivo zebrafish. We observed increased tnfa expression with stabilized Hif-1α, demonstrating that hypoxia signaling alone can lead to a change of macrophage phenotype, consistent with our previous observations with il-1β expression, suggestive of a pro-inflammatory response (37). Our findings are also in line with HIF-1α stabilization being important for the production of TNF and other proinflammatory molecules, such as nitric oxide (NO), leading to host-protective effects against group A Streptococcus in in vivo murine models (50). Not only could tnfa activation be achieved by genetic stabilization of Hif-1α, but also using hypoxia mimetics (hydroxylase inhibitors) and physiological hypoxia, that stabilize endogenous Hif-1α. These findings indicate that the Hif-1α-Tnfa axis is targetable by pharmaceuticals and could be druggable during disease.
A similar tnfa induction was induced by infection and inflammation in our in vivo zebrafish models. It has been widely demonstrated that Tnfa is required for control of early Mm infection, with perturbation of Tnfa signaling leading to high infectious burdens (51). While Hif-1α-induced tnfa was downregulated by cyclooxygenase inhibition, it was notable that the high level of tnfa expression after inflammation or infection was not affected by cyclooxygenase inhibition. These findings are consistent with the novel HIF/COX/TNF macrophage axis being independent of infection/inflammationinduced macrophage Tnfa. Mm-induced Tnfa is widely reported as being toll-like receptor (TLR) mediated, and our data indicate that Mm-induced levels of Tnfa are sufficient for control of bacteria in the presence of stabilized Hif-1α (28,52). In the absence of the TLR adapter protein MyD88, bacterial induction of Tnfa is lost, demonstrating the central role of TLR pathways in post-infection Tnfa induction (28,(52)(53)(54). The best characterized exogenous TLR signal, LPS via TLR4, can induce Tnfa production via activation of the central pro-inflammatory regulator NFκB (54,55). As cyclooxygenase inhibition did not alter either inflammation or infection driven tnfa it suggests that these complex damage/infection driven mechanisms of Tnfa induction are not dependent on the HIF/COX/TNF pathway. Our data demonstrate that Hif-1α stabilization can stimulate a macrophage tnfa producing response in the absence of a TLR stimulus, which may have important implications in sterile diseased tissue and infers an alternative macrophage tuning mechanism.
Hypoxic regulation of TNF via COX has previously been demonstrated in mammalian osteoblasts, however, little is known about this interaction (14). Here, we demonstrate that Hif-1α upregulation of macrophage tnfa is likely to be via the production of PGE2 signaling via the EP4 receptor. PGE2 is an inflammatory mediator that is known to affect macrophage polarization states during infection, and synergises with cytokines to amplify proinflammatory responses (41,56). The degradation metabolite 15-keto-PGE2 did not rescue the loss of tnfa expression after cyclooxygenase inhibition, consistent with previous reports that 15-keto-PGE2 is unable to bind the prostaglandin E2 EP receptors, thus demonstrating a requirement for active PGE2 (57).
Here, we demonstrate that a macrophage HIF/COX/TNF pathway is found in human monocyte derived macrophages, indicating that this pathway may have conserved roles in macrophage homeostasis and disease, and opening up possibilities for pharmaceutical intervention to promote proinflammatory macrophage phenotypes in human disease. Our findings in human MDMs indicate that the HIF/COX/TNF pathway may, at least in part, be cell-autonomous to macrophages, however, this was not tested in vivo where it is likely that contributing signals from other celltypes, for example endothelial cells, may play important roles in macrophage phenotype outcomes (58).
Although the effect of the HIF/COX/TNF axis in early Mm infection was negligible, this pathway is likely to have important roles in other disease situations. Mycobacterial infection, along with many other bacterial pathogens, stimulates early TNF production, however some pathogens remain undetected by macrophages during early disease. An example of this is in Lyme disease, caused by Lyme borreliosis, where early low TNF production is associated with worse disease outcomes (59). Multiple infectious/inflammatory disease pathologies have hypoxia as a feature of the tissue microenvironment, concurrent with the presence of macrophages. For example, Hif-1α mediated TNF activation may be pertinent in later TB infection situations where granulomas have stabilized Hif-1α due to necrotic and hypoxic centers (60,61). Hypoxia is a key hallmark of cancers with high levels of HIF-1α widely found in those that produce large tumors where the center is hypoxic (62). These tumors can also produce cyclooxygenase/PGE2, where these important immunomodulatory signals are required for cellular adaption to the tumor microenvironment (63). Macrophages play central roles in cancer-inflammation that could be exploited utilizing pharmaceutical control of this novel HIF/COX/TNF mechanism. Further investigation of the HIF/COX/TNF axis in models where hypoxia is a key hallmark of disease pathology is required to uncover the full therapeutic relevance of this potentially important novel macrophage pathway.
In conclusion, we have identified a novel mechanism for macrophage tnfa production via Hif-1α and cyclooxygenase that is found in zebrafish and humans. Importantly, this axis links a commonly found microenvironmental cue, hypoxia, to a macrophage cytokine, TNF. We provide strong evidence to show that this response acts via the cyclooxygenase/PGE2 arm of the arachidonic pathway. Due to the central roles of these modulators in disease microenvironments we anticipate that this HIF/COX/TNF pathway may have important implications in conditions such as inflammation, infection and cancer.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the South Sheffield Research Ethics Committee (07/Q2305/7), University of Sheffield, Firth Court, Western Bank, S10 2TN. The healthy participants provided their written informed consent to participate in this study. Animal work was performed following UK law: Animal (Scientific Procedures) Act 1986, under Project License P1A4A7A5E. Ethical approval was granted by the University of Sheffield Local Ethical Review Panel, University of Sheffield, Firth Court, Western Bank, Sheffield, S10 2TN.

AUTHOR CONTRIBUTIONS
PE and AL conceived and designed the experiments, performed the experiments, and analyzed the data. PE wrote the paper.