Activation of hepatic stellate cells (HSC) plays a crucial role in the liver disease progression from liver fibrosis/cirrhosis to cancer. Here, we found a rapid change of microenvironment after krasV12-induction in zebrafish liver with progressively increased stromal cell number and enlarged liver size. Neutrophils and macrophages exhibited a faster response than HSCs. By manipulating the numbers of neutrophils and macrophages through morpholino knockdown, we found that macrophages contributed to both HSC survival and activation while neutrophils appear to be only required for HSC activation. Serotonin, which is essential for HSC survival and activation, was found up-regulated in hepatocytes and macrophages, but not in neutrophils after krasV12 induction. Serotonin receptor was highly expressed in HSCs; increase of the receptor activity by an agonist stimulated HSCs and oncogenic growth of the liver while an opposite effect was observed with an antagonist. Activated HSCs promoted the pro-tumorigenesis functions of neutrophils and macrophages through secretion of Tgfb1. Overall, these observations elucidated a cellular interaction in microenvironment where that upregulated serotonin in hepatocytes and macrophages activated HSCs. Since the microenvironment crosstalk plays a vital role in manipulation of liver carcinogenesis, the underlying mechanism may provide potential therapeutic targets for liver diseases.
Tumor microenvironment is important in tumor initiation and progression. Hepatic stellate cells (HSCs), liver-specific mesenchymal cells, are an important component of tumor microenvironment and the stromal dynamics is also a key determinant of liver cancer progression1. In normal liver, HSCs are quiescent; upon liver injury, HSCs can be activated by various cytokines, such as TGFB (transforming growth factor b), IL1B (interleukin 1b) and TNF (tumor necrosis factor)2. HSC activation results in increasing extracellular matrix remodeling capabilities, as observed in non-alcoholic steatohepatitis patients3. There is also a strong relationship between activated HSCs and oncogenic hepatocytes. HSCs are the pre-eminent matrix-producing cells in liver fibrosis, potently secreting TGFB1 and PDGF, thus mediating liver fibrosis as well as hepatocyte proliferation4.
In hepatocellular carcinoma (HCC), activated HSCs induce phenotypic changes in oncogenic hepatocytes, notably through the production of growth factors and cytokines such as HGF (hepatocyte growth factor) and IL6 in favor of tumor cell proliferation5. Non-neoplastic stromal composition serves as a reliable prognostic predictor in HCC patient survival and recurrence6,7. Densities of intratumoral endothelial cells, tumor associated macrophages (TAMs), tumor associated neutrophils (TANs), CD8+ T-cells and activated HSCs are significantly higher in patients with liver diseases and they are also indicators of poor disease prognosis8,9,10,11. Curiously, various components of the tumor microenvironment are capable of influencing one another to build a complex signaling network. For example, TAMs and TANs are simultaneously activated to a pro-tumor gene expression profile during carcinogenesis and could affect each other12. Interestingly, both TAMs and TANs have been demonstrated to be able to influence HSCs in a non-neoplastic context. TAMs activate NFkB signaling in HSCs and thereby promote liver fibrosis via HSC-mediated extracellular matrix remodeling13. TAN-derived superoxide anion induces lipid peroxidation and stimulates collagen synthesis in HSCs14.
The zebrafish has been increasingly recognized as an alternative animal model in cancer studies15. Previously our laboratory has generated several inducible HCC models by transgenic expression of an oncogene in hepatocytes in zebrafish16,17,18,19. These zebraifsh models showed close histological resemblance to human HCC and share conserved molecular signatures with advanced human HCC20,21. Moreover, the inducible nature of these transgenic HCC model allows the oncogene to be switched on at a desired timing, providing an excellent model to study HCC initiation events. In addition, the transparent nature of zebrafish larvae and the availability of various transgenic reporter zebrafish with expression of a fluorescence protein marker in specific cell lineage such as HSCs, neutrophils or macrophages provide excellent tools to elucidate interactions of different microenvironment components in a neoplastic context22,23,24. In this study, we attempt to elucidate the microenvironment interaction among hepatocytes, HSCs, TAMs and TANs during early stages of liver tumorigenesis. By using several transgenic zebrafish lines, we found that along with krasV12-expressing oncogenic hepatocytes (OHs), macrophages could activate HSCs through serotonin-mediated signaling. Interestingly, HSC activation resulted in up-regulation of Tgfb1 expression, which in turn maintained pro-tumor gene expression in TANs and TAMs and thereby promoted carcinogenesis.
Rapid modification of microenvironment following kras V12-induced liver tumorigenesis
In human HCC patients, there are high densities of TANs, TAMs and HSCs in the liver tumors and these correlate to poor prognosis8,9,10,11. To investigate if krasV12-intiated liver tumorigenesis would induce similar microenvironment modification in our transgenic zebrafish model, infiltration of neutrophils and macrophages as well as HSC density and activation were examined using different reporter transgenic lines as described in Methods. By crossing either lyz+ or mpeg+ transgenic reporter line with the kras+ zebrafish, infiltration of neutrophils and macrophages into krasV12-expressing livers was visualized at different time. Consistent with our previous study25, neutrophil infiltration was significantly increased from 8 hours post-induction (hpi), preceding the onset of krasV12-induced liver enlargement at 24 hpi (Fig. 1A,B). Similarly, macrophage infiltration became significant from 16 hpi (Fig. 1C). To examine HSC density and activation, a combination of HSC marker proteins, glial fibrillary acidic protein (Gfap) and α-smooth muscle actin (α-Sma), was used to identify HSCs in krasV12-expressing livers (Fig. 1D). Gfap+ cells in the liver demarcate both quiescent and activated HSCs22, while α-Sma marks only activated HSC26. Supplementary Figure S1A shows co-localization of hepatocyte-specific gfp, DAPI labeled nuclei and α-Sma. It is apparent that α-Sma is located in both cytoplasm and nuclei and this is consistent with an earlier observation in human mesenchymal stromal cells27. It is also interesting to note that α-Sma is also expressed at a low level in the liver (mostly in hepatocytes) (Fig. 1D). This is consistent with an earlier report in a mouse liver tumor model28. Quantification of Gfap and α-Sma staining is shown in Fig. 1E and it appeared that both total HSC density and the ratio of activated HSCs were increased significantly in krasV12-expressing livers, compared to wildtype livers at 24 hpi. In our previous study, we demonstrated that kras+ fish showed early HCC phenotype at 7 dpf and 96 hpi25. Hence, krasV12-expressing induced rapid microenvironment modifications, as indicated by sequential occurrence of increased infiltration of neutrophils and macrophages, followed by proliferation and activation of HSCs.
Regulation of HSC survival and activation by neutrophils and macrophages
In previous studies, both neutrophils and macrophages have been reported to affect HSC functions during liver disease progression14,29. However, their influence on HSCs during liver tumorigenesis is largely unknown. Since infiltration of these innate immune cells preceded HSC proliferation and activation in krasV12-induced liver tumorigenesis (Fig. 1), possible roles of neutrophils and macrophages in regulating state of HSCs were investigated. Splice-blocking morpholinos, MO-gcsfr, MO-irf8 and MO-pu.1 were used to inhibit differentiation of neutrophils or macrophages in krasV12-expressing larvae. Effects of these morpholinos were validated by injection in lyz+ or mpeg+ embryos (Supplementary Fig. S2). As reported by Feng et al.30, MO-gcsfr completely abrogated neutrophil differentiation but caused no change of the count of macrophages. Similarly as reported by Clay et al.31, MO-pu.1 repressed macrophage differentiation whereas neutrophil differentiation was unperturbed. As reported by Shiau et al.32, MO-irf8 favored neutrophil differentiation at the expense of macrophage differentiation, thereby increasing neutrophils and decreasing macrophages. As LPS induce macrophage activation33, a combination of MO-gcsfr morpholino injection and LPS treatment inhibited neutrophil differentiation while concurrently stimulating macrophage differentiation (Supplementary Fig. S2).
To investigate the effects of neutrophils and macrophages on HSCs, both total and activated HSCs were examined by IF co-staining with Gfap and α-Sma, under different knockdown conditions. Depletion of neutrophils in the kras+ larvae by MO-gcsfr or by a combination of MO-gcsfr and LPS treatment did not significantly reduce the numbers of total HSCs but resulted in a small and significant decreases of the portion of activated HSCs (Fig. 2A–C). In contrast, decreasing macrophage counts in the kras+ larvae by MO-irf8 and Mo-pu.1 resulted in a decrease in both total and activated HSCs (Fig. 2A–C). As it has been reported that activated HSCs secrete Tgfb1 and promote liver disease progression34, to further understand the molecular mechanisms of neutrophil- and macrophage-modulated HSC activity, co-IF staining of Gfap with Tgfb1 was carried out (Fig. 2D,E). As shown in Supplementary Fig. 1B, Tgfb1 is mainly localized in the cytoplasm. Depletion of either neutrophils or macrophages in kras+ larvae decreased the percentage of Tgfb1-expressing HSCs, suggesting that both neutrophils and macrophages contribute to Tgfb1-expressing capabilities of HSCs. In contrast, Gfap+/Caspase-3+ cells indicated apoptotic HSCs. Proportion of apoptotic HSCs increased only when macrophage differentiation is inhibited (MO-irf8 and MO-pu.1), thus explaining for the decreased HSC density observed with depletion of macrophages (Fig. 2F,G). Gfap−/Caspase3+ cells are mostly hepatocytes and there was no significant difference of hepatocyte apoptosis between different treatments. Therefore, during krasV12-induced liver tumorigenesis, neutrophils only appear to be required for HSC activation while macrophages contribute to both HSC survival and activation. It is interesting to note from Fig. 2G that kras expression, though causing increases of total and activated HSCs, also had a higher apoptotic rate of HSCs; thus it seems that kras-induced oncogenesis also leads to a higher turnover of HSCs.
Serotonin upregulation in hepatocytes, TAMs and Htr2b expression in HSCs
To explain how TAM maintain HSC survival and activation, possible underlying molecular mechanisms were interrogated. Serotonin served primarily as a neurotransmitter but is also required for HSC survival and activation during liver regeneration35. To examine the level of serotonin in hepatocytes, neutrophils and macrophages in wildtype and krasV12-expressing livers, IF staining of serotonin on liver sections of 6 dpf (days post fertilization) kras+/mpeg+, mpeg+, kras+/lyz+ and lyz+ larvae were conducted (Fig. 3A). As shown in Supplementary Fig. 1C, serotonin signal is mainly visiable in the cytoplasm. The serotonin level was significantly upregulated in hepatocytes and TAMs, but not in TANs after krasV12 induction (Fig. 3B). Meanwhile, the expressing of tph1a, which encodes the rate-limiting enzyme of serotonin synthesis, tryptophan hydroxylase 1a36, was examined and compared in FACS-isolated OHs, TANs and TAMs versus control hepatocytes, naïve neutrophils and macrophages respectively. Tph1b expression was found to be significantly up-regulated in OHs and TAMs but not in TANs (Fig. 3C), which was consistent with serotonin upregulation in OHs and TAMs. Serotonin has been showed to play an activating role on HSCs via 5-hydoxytryotamine receptor 2B (Htr2b) in a mouse liver regeneration model35. Consistent with this, htr2b expression was found to be predominantly in HSCs in zebrafish (Fig. 3D), being about 27 times higher than that in hepatocytes. Almost no htr2b expression was detected in neutrophils and macrophages (Fig. 3D).
Manipulation of Htr2b activity affect liver carcinogenesis and HSC density, activation and function
To examine the role of Htr2b-activated HSCs in krasV12-induced liver tumorigenesis, kras+ larvae was co-treated with doxycycline (dox) and an antagonist (SB204741) or an agonist (BW723C86) of Htr2b. SB204741 has been previously used as an antagonist of Htr2b35,37 and it has been shown to have 135 times higher selectivity over the most relevant serotonin receptors, Htr2c38. BW723C86 has also been widely used for investigating the effects of activation of Htr2b in disease progression39,40,41. Gross morphology was captured after the treatment and all the larvae were shown in the left lateral view (Fig. 4A). Inhibition of Htr2b in kras+ larvae displayed a deterrence of krasV12-induced liver enlargement while stimulation of Htr2b robustly accelerated liver tumorigensis (Fig. 4B). To elucidate the effects on HSC density and activation, IF co-staining of Gfap and α-Sma was performed in the liver section of the larvae fish (Fig. 4C). The antagonist treatment significantly decreased the total HSC density and the ratio of activated HSCs, while the agonist treatment increased HSC density significantly (Fig. 4C,F,G and Supplementary Fig. S3). It has been reported that activated HSCs secrete Tgfb1 to promote liver fibrogenesis42. IF co-staining of Gfap and Tgfb1 showed that the antagonist decreased Tgfb1 expression and the agonist promoted Tgfb1 expression (Fig. 4D,H). In addition, IF co-staining of Gfap and Caspase-3 indicated that the antagonist, but not the agonist, significantly induced HSC apoptosis (Fig. 4E,I). In sum, Htr2b was crucial for HSC survival and function; activation of Htr2b induced HSC proliferation and stimulated Tgfb1 expression, while inhibition of Htr2b caused HSC apoptosis.
To identify the underlying mechanism for the change in liver size caused by inhibition and activation of Htr2b during krasV12-induced carcinogenesis, IF staining of hepatocyte proliferation (PCNA) and apoptosis (Caspase-3) was carried out (Fig. 5A,B and Supplementary Fig. S4A,B). The antagonist treatment significantly inhibited the proliferation of oncogenic hepatocytes and stimulated the hepatocyte apoptosis. The agonist treatment accelerated the hepatocyte proliferation (Fig. 5E,F). However, both antagonist and agonist treatments had no significant effects on cell proliferation and apoptosis in wildtype livers. Furthermore, whether or not the antagonist and agonist treatments affected the progression of liver fibrosis was also investigated. Progression of liver diseases is generally through liver fibrosis to cirrhosis and to carcinoma43 and collagen I and Laminin has been widely used as the liver fibrosis markers44. To examine the progression level of liver fibrosis, Collagen I and Laminin were detected by IF staining (Fig. 5C,D). The antagonist of Htr2b significantly inhibited the deposition of both Collagen I and Laminin, while the agonist treatment slightly stimulated the progression of liver fibrosis. There were no significant differences in wildtype liver among the two different treatments (Fig. 5G,H). Thus, it seems that the antagonist blocked liver carcinogenesis through induction of cell apoptosis and attenuate liver fibrosis while the agonist stimulated liver carcinogenesis by accelerating hepatocyte proliferation and liver fibrosis.
Reciprocal roles of HSCs on oncogenic hepatocytes, neutrophils and macrophages
As demonstrated in previous sections (Figs 3 and 4), serotonin was greatly increased upon activation of oncogenic kras in hepatocytes and this in turn activated HSCs through the receptor Htr2b. An obvious feature of the activation is the increase of Tgfb1 in activated HSCs, as previously reported for human cystic fibrosis-associated liver disease samples45. The serotonin-Tgfb signaling in HSC has been shown in a mouse model where seorotonin binding to Htr2b is found to stimulate TGFB1 expression in HSCs35. In our previous report, we have already shown the up-regulation of Tgfb in HSCs by serotonin treatment in zebrafish46. In the present study, the reciprocal role of HSCs on other cell types, such as hepatocytes, neutrophils and macrophages, was also investigated by examining gene expression changes following inhibition of Tgfb signaling. SB431542 is a specific inhibitor of Tgfb type I receptor and has been used in both human cells and zebrafish47,48. In our previous study, we also demonstrated the effect of SB431542 in zebrafish larvae by observing a significant decrease of phosphated Smad2, a well recognized downstream effector25. Inhibition of Tgfb signaling deterred the oncogene-induced liver enlargement (Fig. 6A,B). To further examine the effect of Tgfb1 on OHs, neutrophils and macrophages, gene expression of FACS isolated hepatocytes, neutrophils and macrophages from kras+, kras+/lyz+ and kras+/mpeg+ larvae treated with dox or dox/SB431542 was compared to FACS isolated control hepatocytes, naïve neutrophils and macrophages from fabp10+, lyz+ and mpeg+ larvae, respectively. In OHs, krasV12-induced up-regulation of pro-fibrosis (col1a1b, lama5), angiogenesis (vegfab, vegfc), epithelial-mesenchyme transition (snai1, slug) markers and down-regulation of anti-tumor cytokines (il12, tnfa) were significantly deterred by Tgfb inhibition (Fig. 6C). In TANs and TAMs, krasV12-induced up-regulation of pro-cancer inflammation genes (il1b and cxcl1) and down-regulation of anti-tumor cytokines (il12 and tnfα) and phagocytosis genes (mpx, lyz and mpeg) were also reverted with Tgfb signaling inhibition (Fig. 6D,E). Together, changes in gene expression profile showed that with Tgfb signaling inhibition, OHs, TANs and TAMs became less pro-cancer inflammation and more anti-tumor.
In clinical studies, microenvironment inflammatory infiltrate has a positive correlation with HCC progression49. Liver fibrosis has a positive relevance in a long-term prognosis of nonalcoholic fatty liver disease and HCC50. We showed here that there is a rapid modification of microenvironment after oncogene activation. Significant infiltration of neutrophils and macrophages occurred first, followed by activation of HSC at least partially through TANs and/or TAMs. By knocking down the critical genes for neutrophil and macrophage differentiation, we find decreased numbers of neutrophils or macrophages to also cause a significant effect on HSC density and function. Depletion of macrophages decreased both total and activated HSCs, while depletion of neutrophils blocked HSC activation significantly, indicating different crosstalks between HSCs and these innate immune cells in the TME. This observation was reminiscent of previous studies where during liver fibrosis, neutrophils activate HSCs via Il17-dependent mechanism while macrophages is a stimulating factor for HSCs via NFkb activation29,51.
Molecularly, serotonin has been reported to activate stellate cells in a liver regeneration context35. In clinical studies, serotonin level is significantly increased during liver disease progression and is the highest during cirrhosis and early HCC52. Previously, we have found that serotonin is up-regulated in adult zebrafish HCC model and human liver disease samples46. Our current study recapitulated the increased serotonin level during krasV12-induced liver tumorigenesis in both OHs and TAMs, but not in TANs in zebrafish larvae. Serotonin are essential for HSC activation and Tgfb1 expression. The serotonin expression in TAMs and OHs, but not in TANs, explained that inhibition of macrophage differentiation had more crucial effects on HSC survival and function than inhibition of neutrophil differentiation. However, depletion of neutrophils also decreased the ratio of activated HSC, implying that there could be other signaling crosstalk between neutrophils and HSCs.
HSCs are well-established source of Tgfb1. In fact, HSC produced Tgfb1 is the pre-eminent mediator of liver fibrosis45. Here, we also observed upregulated expression of Tgfb1 after HSC activation. Tgfb1 is a known chemoattractant for both neutrophils53 and macrophages54 and it has been reported to increase hepatocyte proliferation and polarize neutrophils or macrophages to a pro-tumor profile during hepatocarcinogenesis55. After inhibition of Tgfb signaling, our zebrafish model showed downregulation of pro-tumor gene expression profiles in TANs and TAMs, indicating that the activated HSC-produced Tgfb1 could have feedback activation of TANs and TAMs.
In conclusion, with an inducible oncogene model in zebrafish larvae, we elucidated a regulatory mechanism of HSCs via upregulated serotonin in OHs and TAMs. Serotonin activated-HSC produced Tgfb, which in turn activated TANs and TAMs. Hence, microenvironment crosstalk between HSCs and TANs/TAMs could mutually activate one another, thus accelerating liver tumorigenesis. Inhibition of the crosstalk between these cell types could significantly deter rate of carcinogenesis. Hence, targeting these tumor-supporting components offers an attractive therapeutic strategy in the treatment of liver diseases.
Zebrafish were maintained in compliance with the guideline and protocol approved by Institutional Animal Care and Use Committee of National University of Singapore. Five transgenic lines, Tg(fabp10:rtTA2s-M2; TRE2:EGFP-krasG12V) (gz32Tg) in a Tet-On system for inducible, hepatocyte-specific expression of oncogenic krasG12V 16; Tg(lyz:DsRed2) (nz50Tg) with DsRed-labelled neutrophils under the lyz (lysozyme C) promoter23; Tg(mpeg1:mCherry) (gl22Tg), with mCherry-labeled macrophages under the mpeg1 (macrophage expressed gene 1) promoter24; Tg(fabp10:DsRed; ela3l:EGFP) (gz15Tg) with DsRed-labeled hepatocytes under the fabp10a (fatty acid binding protein 10a) promoter and EGFP-labeled exocrine pancreatic cells under the ela3l (elastase) promoter56 were used and referred to as kras+, lyz+, mpeg+ and fabp10+, respectively, in the present report.
All chemical treatments were conducted from 3 dpf for 4 days. The chemicals used included doxycycline (dox) (20 µg/ml; Sigma, D9891), BW723C86 (15 µM; Sigma, B175), SB204741 (15 µM; Sigma, S0693), SB431542 (2.5 µM; Tocris, 1614), LPS (5 ng/µl; Sigma, L4391). The dosages were selected based on our previously validated experiments25,46,57.
For knockdown of Gcsfr, Irf8 and Pu.1, previously validated gene-specific morpholinos were used, including MO-gcsfr (5′-GAAGCACAAGCGAGACGGATGCCAT-3′)30, MO-irf8 (5′-AATGTTTCGCTTACTTTGAAAATGG-3′)58 and MO-pu.1 (5′- GATATACTGATACTCCATTGGTGGT-3′)59. A standard control morpholino, MO-control (5′-CCTCTTACCTCAGTTACAATTTATA-3′) targeting a human beta-globin intron (Gene Tools, Philomath, OR), was used as a negative control. Aliquots of morpholino (1 mM) and 1% (wt/vol) phenol red in Danieau solution were injected into embryos at the 1-cell stage. Dox was added from 3 dpf to 6 dpf.
Photography and image analysis
Prior to imaging, zebrafish larvae were anesthetized in tricaine (0.08%; Sigma, E10521) and immobilized in methylcellulose (3%; Sigma, M0521). For each time point of treatment experiments, 20 larvae were used for imaging analyses in each group. Each larva was photographed individually with Olympus microscope. Livers in wildtype fish were manually circled based on bright field images and measured using ImageJ. Tumor livers in kras+ fish were imaged for GFP expression and GFP expressing areas were measured by ImageJ.
Isolation of hepatocytes, HSCs, neutrophils and macrophages by FACS
To enrich HSCs, neutrophils, macrophages and hepatocytes, central part of 6-dpf larvae (after removal of the head and tail regions) were used for FACS using a cell sorter (BD Aria, 643245). The central part of larvae were dissociated into single cells using a 40-µm mesh (BD falcon, 352340) following digestion with trypsin (0.05%; Sigma, T1426), as previously described60. KrasV12 expressing hepatocytes (kras+) and HSCs (hand2+) were isolated based on EGFP expression, neutrophils (lyz+) and normal hepatocytes (fabp+) were isolated based on DsRed expression and macrophages (mpeg+) were isolated based on mCherry expression. Expression of fabp10a, lyz and mpeg was further used to confirm the purity of sorted hepatocytes, neutrophils and macrophages (Supplementary Fig. S5). Indeed, fabp10a, lyz and mpeg were highly and specifically expressed in hepatocytes, neutrophils and macrophages, respectively in both wildtype and kras+ fish.
RNA extraction, cDNA amplification and RT-qPCR (reverse transcription-quantitative PCR)
Total RNA was extracted using RNeasy mini kit (Qiagen, 74104). A total of 5 ng RNA was used as a template to synthesize and amplify cDNA using QuantiTect Whole Transcriptome Kit (Qiagen, 207043). Amplified cDNA was thus used for real-time quantitative PCR with LightCycler 480 SYBR Green I Master (Roche, 04707516001). Interested genes were amplified for 40 cycles (95 °C, 20 seconds; 65 °C, 15 seconds; 72 °C, 30 seconds). The sequences of primers used are presented in Supplementary Table S1.
Histological and cytological analyses
Zebrafish larvae were fixed in 4% paraformaldehyde in phosphate buffered saline (Sigma, P6748) and cryo-sectioned at 8 µm thickness using a Leica Cryostats CM1950, followed by immunofluorescence (IF) staining. The primary antibody was derived either from rabbits (anti-PCNA, Santa Cruz, FL-261; anti-Caspase-3, BD biosciences, C92-065; anti-Serotonin, Sigma, C5545; anti-Tgfb1, Anaspec, 55450; anti-Collagen I, Abcam, ab23730; anti-laminin, Sigma, L9393; anti-α-Sma, Abcam, ab15734) or from mice (Anti-Gfap, Abcam, 154474). Anti-mouse or anti-rabbit secondary antibodies were purchased from Thermo Fisher Scientific.
Statistical significance between two groups was evaluated by two-tailed unpaired Student t-test using inStat version 5.0 for Windows (GraphPad, San Diego, CA). Statistical data are presented as mean values ± standard error of mean.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by grants from Ministry of Education of Singapore (R154000667112 and R154000A23112).