Epstein–Barr Virus Induces Adipocyte Dedifferentiation to Modulate the Tumor Microenvironment

Epstein–Barr virus (EBV) is implicated in the development of several malignancies (1), including lymphoproliferative disorders (Burkitt's lymphoma, Hodgkin's disease, non-Hodgkin's lymphoma, HIV-associated lymphomas, leiomyosarcomas), epithelial malignancies (undifferentiated NPC and some forms of gastric carcinoma) as well as lymphoepithelioma-like carcinoma (LELC) arising in regions of the stomach (2, 3), breast (4), skin (5), thyroid (6), salivary (7), and kidney (8, 9). In terms of tropism, EBV can infect B cells and epithelial cells (10, 11), as well as non-B cell and non-epithelial cell populations, such as T cells and NK cells (10). However, the issue of whether other cell types are infected by EBV and the mechanisms by which these cell reservoirs of infection interact with tumor cells remain unclear.

Several tumors grow in the vicinity of adipocytes (such as breast and ovarian cancer) or metastasize to an adipocyte-dominated host environment (omentum or bone marrow; refs. 12, 13). Current evidence supports a model in which cancer cells reprogram adipocytes to generate cancer-associated adipocytes (CAA; refs. 14–17). CAAs release fatty acids (FA), mitogens, and proinflammatory adipokines, such as leptin (LEP), IL6, and TNFα, through lipolysis, which are transferred to tumor cells and cause activation of oncogenic pathways (17, 18) as well as cancer progression (19, 20). Remote metastasis is detected in ∼20% NPC cases. Half of these cases display metastasis to bone/bone marrow (21), an adipocyte-dominant region. Moreover, several EBV-associated LELC subtypes grow in the anatomical vicinity of fat tissues, such as breast and salivary gland. Several studies have reported the ability of EBV to alter metabolic pathways. For instance, EBV-encoded LMP1 modulates glycolysis and lipogenesis in NPC (22–24). Wang and colleagues (25, 26) further demonstrated that EBV reprograms the lipid biogenesis pathway for B-cell activation and transformation. However, to date, crosstalk among EBV, adipocytes, and tumor cells has remained largely unexplored.

Adipocytes represent an important long-term reservoir for certain types of virus [human immunodeficiency virus (HIV), adenovirus, and CMV; refs. 27–29] and parasites (Chagas disease; ref. 30), which may facilitate infection relapse. Patients infected with HIV experience many metabolic abnormalities (31). Thus, it is likely that other virus types, such as EBV, can similarly penetrate adipocytes and modulate adipocyte biology to benefit virus survival and replication.

In this study, we investigated the potential mechanisms by which EBV dysregulates adipocyte function within the tumor microenvironment. Our data clearly indicate that EBV is capable of infecting human adipocytes. Furthermore, infection induces expression of virus-encoded genes in adipocytes, triggering lipolysis, and dedifferentiation, enhances expression of proinflammatory adipokines and release of lipid mediators, and eventually leads to alterations in components of the tumor microenvironment. On the basis of the collective results, we propose that EBV efficiently reprograms adipocyte metabolic machinery to meet the energy and nutritional demands of rapidly proliferating cancer cells.

Primary human white preadipocytes (HWP) were purchased from PromoCell. Differentiation was conducted according to the manufacturer's instructions. Briefly, preadipocytes were grown in preadipocyte growth medium, which was replaced with preadipocyte differentiation medium upon growth to full confluence. Three days after induction of differentiation, the medium was replaced with adipocyte nutrition medium for 14–20 days. Mature adipocytes were obtained during this period. HK1, HK1EBV, and AKATA cells were cultured in RPMI1640 supplemented with 10% FBS (Gibco) and NPC-TW06 cells (32) grown in DMEM supplemented with 10% FBS. The authentication of NPC-TW06 cells, HK1 cells, HK1EBV cells, and AKATA cells was confirmed by short tandem repeat (STR) profile analysis (conducted by the Bioresource Collection and Research Center, Taiwan).

Pretreatment NPC serum samples (n = 221) were collected for determining levels of free FAs and glycerol. All patients had fasting for about 8 hours before blood sampling. Patients who had any Charlson comorbid conditions such as chronic hepatitis infection or diabetes were excluded from this study. The characteristics of participants are summarized in Supplementary Table S1. A total of 65 pretreatment EBV-associated formalin-fixed paraffin-embedded tumor samples were obtained for IHC analysis. Among these, 28 samples were obtained from bone marrow metastatic NPC tumors, 14 samples from lung metastatic NPC tumors, 12 samples from liver metastatic NPC, and 11 from EBV positive LELC. All individuals were followed up for more than 3 years after treatment. All tumor samples were histologically confirmed by pathologists. Participants signed written informed consent for research purposes only.

Antibodies used for Western blotting, immunofluorescence, or IHC are as follows: phospho-p70S6K1 (T389; Cell Signaling Technology), p70S6K1 (Abcam), GAPDH (Abcam), p-HSL (Ser 563; Cell Signaling Technology), HSL (Cell Signaling Technology), EBV early antigen D (EaD; Abcam), EBV thymidine kinase (TK; LSBio), Rta (Argene), Zta (Santa Cruz Biotechnology), S100A4 (Cell Signaling Technology), phospho-AMPK (T172; Cell Signaling Technology), AMPK (Abcam), phospho-mTOR (S2448; Cell Signaling Technology), mTOR (Cell Signaling Technology), phospho-S6 (S240/244; Cell Signaling Technology), S6 (Cell Signaling Technology), ATGL (Abcam), MGL (Abcam,), phospho-ERK1/2 (T202/Y204), ERK1/2, phospho-GSK-3α/β (S21/9), GSK-α/β, and FAP-α. The anti-LMP1 mAb (S12) was affinity purified from a hybridoma. Trichrome Staining Kit (ScyTek Laboratories) was used to detect collagenous tissue fibers in tissue section. Phorbol 12-myristate 13-acetate (30 ng/mL, Sigma-Aldrich), sodium butyrate (3 mmol/L, Sigma-Aldrich), and human IgG (0.5%, The Jackson Laboratories) were used for EBV lytic induction in either HK1EBV cells or recombinant AKATA (rAKATA) cells. Everolimus (20 nmol/L, Sigma-Aldrich), rapamycin (20 nmol/L, Calbiochem), PD98059 (20 μmol/L, Calbiochem), GSK 2334470 (2 μmol/L, AdooQ Bioscience), and PF-4708671(5 μmol/L, Cayman) were used to block specific signaling pathways. Human recombinant adipokines, including LIF (20 ng/mL, Sigma-Aldrich), IL6 (30 ng/mL, R&D Systems), SCF (50 ng/mL, R&D Systems), MIP-1α (100 ng/mL, R&D Systems), IL8 (100 ng/mL, R&D Systems), and LEP (100 ng/mL, R&D Systems) were used to investigate effects on cell proliferation and activation of signaling pathways in NPC cells. Levels of adipocyte-secreted free FAs and free glycerol were determined using a Quantification Kit (BioVision). The human ELISA kits (R&D Systems) were used to detect levels of adipokines (LIF, IL6, IL8, Leptin, SCF, and MIP-1α) secreted into the culture medium of adipocytes.

IHC was performed using Leica BOND-MAX system and the Bond Polymer Refine Detection Kit (Leica Microsystems, DS9800) as described previously (33). In situ hybridization of EBV-encoded small RNAs (EBER) was performed according to the manufacturer's instructions (Novocastra, Leica Biosystems). The primary antibody or EBER oligonucleotide probe was omitted from negative controls (Supplementary Figs. S1A–S1E).

Fully differentiated human adipocytes were infected with recombinant EBV (tk_GFP_EBV) obtained from EBV-harboring Burkitt's lymphoma (rAKATA) or HK1EBV cells. Briefly, rAKATA cells were treated with 0.5% (vol/vol) goat anti-human immunoglobulin G (The Jackson Laboratory) or HK1EBV cells with TPA (30 ng/mL) and sodium butyrate (3 mmol/L) to induce lytic virus production. The supernatant of 4 day-activated cells was collected, centrifuged at 4°C, 3,000 rpm for 20 minutes, and filtered through a 0.45 μm membrane. EBV-containing supernatant (1 mL) was applied directly to adipocyte culture, followed by shaking at 200 rpm for 1 hour. The culture was washed four times with PBS and incubated at 37°C until harvesting. The same procedures were used to collect control supernatant derived from EBV-negative AKATA cells or HK1 cells. Adipocytes treated with an equal amount of control supernatant were used as uninfected controls.

Statistical analyses were performed using SPSS 17.0 (SPSS) or GraphPad Prism 5 (GraphPad Software). Univariate and multivariate Cox proportional hazards models were used to identify the factors related to prognosis. Kaplan–Meier survival and log-rank tests were applied to compare survival times between groups and the Mann–Whitney test used to evaluate differences between clinical samples. The Student t test was used to estimate the association between the experimental measurements. All statistical tests were two-sided. P values <0.05 were considered statistically significant.

This study was approved by the Institutional Review Board of Chang Gung Memorial Hospital (IRB104-2479B), Taiwan. Written informed consent was obtained from participants prior to inclusion in the study.

Adipocytes surrounding tumors exhibit profound morphological and functional alterations (34). Histologic examination revealed that adipocytes located at the invasive front of EBER⁺-tumors were smaller in size (Fig. 1A). To examine the possibility that EBV infects adipocytes, we evaluated the expression of EaD or small RNAs (EBERs) in adipocyte-rich EBV-associated malignancies. Results of immunohistochemical analysis revealed medium-to-strong immunoreactivity of EBERs or EaD in near 93% (50/54) of metastatic NPC tumors, and 66.7% (4/6) of salivary LELC tumors (Supplementary Table S2). Our results showed positive immunoreactivity of EaD in adipose tissues in close proximity to tumor beds (Fig. 1B, left and middle) but not in distal adipocytes (Fig. 1B, right). The immunoreactivity of EaD or EBERs in adipose tissues were found in approximately 45% (9/20) NPC tumors metastasizing to bone marrow and 50% (2/4) salivary LELC (Supplementary Table S2).

These histologic observations support the theory that interactions between adipocytes and tumor cells stimulate functional changes in adipocytes. To further investigate the EBV-mediated biological effects on adipocytes, we induced differentiation of human subcutaneous white preadipocytes into mature adipocytes by providing the necessary growth factors. As expected, primary preadipocytes underwent marked changes in morphology and were fully differentiated into adipocytes at day 9 (Supplementary Fig. S2A). Accumulation of lipid droplets (LD) was examined using BODIPY fluorescent dye in live adipocytes (Supplementary Fig. S2B) and Oil red O staining (Supplementary Fig. S2C). The in vitro differentiation efficiency of human preadipocytes differentiate into mature adipocytes was approximately 91.4% (mature adipocytes/total nuclei, 3,857/4,220) at passage 2, 85.4% (4,282/5,014) at passage 3, and 71.1% (2,791/3,921) at passage 4. Adipocytes differentiated from passage 2 and 3 preadipocytes were used in this study (Supplementary Fig. S2D). Data obtained from infection of fully differentiated adipocytes with cell-free GFP⁺ EBV from rAKATA cells (Materials and Methods) validated that EBV is capable of infecting human adipocytes (Fig. 1C). The infection efficiency of EBV to the adipocytes ranged from approximately 45% to 60% determined by dividing the number of GFP⁺ cells by the total number of nuclei stained with Hoechst 33342 (Supplementary Fig. S2E).

Next, we examined EBV-infected adipocytes for the presence of virus-encoded genes. Real-time RT-PCR experiments (Supplementary Table S3) revealed the presence of several EBV-encoded mRNAs in EBV-infected adipocytes, including nuclear antigen 1 (EBNA1), small noncoding RNA EBER1 (EBER1), thymidine kinase (TK), BZLF1 transcript (Zta), latent membrane protein 2A (LMP2A), and EaD (Fig. 2A–F). In addition, EBV-encoded products expressed in adipocytes infected with EBV derived from either HK1EBV (NPC) or rAKATA (Burkitt lymphoma) cells in a temporally regulated manner. To evaluate the presence of EBV episomes in adipocytes, we examined the levels of terminal repetitions (TR; Type I EBV, 166020–171823) linking the ends of the linear EBV genome. TR levels were increased at 24 hours after infection compared with DNA collected after 1 hour of infection (Fig. 2G). EBV episomal DNA was not detectable in uninfected adipocytes. The copy number of EBV episomal DNA in adipocytes (3 dpi) was about one tenth of the amount detected in HK1EBV cells. EBV-encoded TK, EaD, and latent membrane protein 1 (LMP1) proteins were also detected in adipocytes at 3 dpi, although LMP1 expression was relatively weak (Fig. 2H). However, expressions of TK and EaD were declined at 8 dpi, whereas expression of LMP1 remained detectable at 8 dpi (Fig. 2I). These data implied that EBV might switch program from lytic phase to latency at later time post infection. Data obtained with our model clearly indicate that EBV infects human adipocytes, at least in vitro. Notably, the culture supernatant of adipocytes appeared to induce expression of lytic proteins (EaD, Rta, Zta) but suppressed LMP1 expression in cancer cells harboring EBV (rAKATA and HK1EBV cells) (Supplementary Fig. S3). These findings support our hypothesis that interactions among adipose tissues, EBV, and tumor cells contribute to a protumorigenic niche and adipocytes may constitute an EBV reservoir in vivo.

Adipocytes store energy in the form of triglycerides (TG) in LDs to maintain energy homeostasis. During energy deprivation, TGs are hydrolyzed by lipases, including adipocyte TG lipase (ATGL), hormone-sensitive lipase (HSL), and monoglyceride lipase (MGL), to free FAs (FFA) and glycerol (35). Here, we investigated the biological effects of EBV infection in adipocytes. EBV-infected adipocytes exhibited reduced LD sizes relative to unchallenged adipocytes (Fig. 3A). Quantification of LD areas revealed that EBV infection led to approximately 50% LD reduction, compared with that observed in the uninfected adipocytes [control vs. EBV; mean (SEM): 889.2 (50.7) vs. 485.2 (23.9); Fig. 3B and C]. Reduction of LDs was concomitant with increased mRNA expression of lipolytic genes (ATGL and HSL; Fig. 3D–F) and decreased expression of FA synthase (FASN; Fig. 3G). The AMP-activated protein kinase (AMPK) cascade and ribosomal protein S6 kinase 1 (p70S6K1) pathways are known to regulate energy homeostasis (36) and adipogenesis (37, 38). Accordingly, we assessed the expression of key molecules involved in lipogenesis (AMPK, mTOR, and p70S6K1) and lipolysis (ATGL, HSL, and MGL) in EBV-infected adipocytes. Results of Western blot analysis demonstrated inhibited regulation of lipogenesis and increased expression of lipolytic enzymes (Fig. 3H). Immunofluorescence data showed enhanced delipidation and activation of HSL (HSL_Ser563) in EBV-infected human adipocytes compared with that of non-EBV infected adipocytes (Fig. 3I). Furthermore, enhanced expression of activated HSL was observed in the vicinity of EBER-positive tumors (Fig. 3J). EBV-induced lipolysis was further confirmed by increased release of FFAs and glycerol in culture supernatant compared with non-EBV infected adipocytes (Fig. 3K). These results provide evidence that EBV infection modulates lipid metabolism in the tumor microenvironment. Next, we evaluated whether the FFA levels in serum samples of patients with NPC are correlated with prognosis. The clinicopathologic characteristics of our study population are summarized in Supplementary Table S1. The survival analysis showed poorer local recurrence-free survival of patients with higher levels of FFA (P = 0.034; Fig. 3L, left) compared with patients with lower serum FFA levels. However, no significant correlation was evident between FFA level and metastasis-free survival (P = 0.723; Fig. 3L, right). Moreover, serum FFA was identified as a predictive marker for poorer local control rate in both univariate [P = 0.034, HR = 2.997; 95% confidence interval (CI), 1.089–8.253] and multivariate (P = 0.023; HR = 3.694; 95% CI, 1.197–11.40) Cox regression analyses (Table 1). Our collective findings suggest that higher amounts of systemic FFA have adverse effects in patients with NPC, supporting the utility of serum FFA as an independent prognostic predictor of recurrence-free survival.

EBV-induced release of FFA and glycerol during lipolysis provides crude materials for promoting cancer cell growth. We further evaluated whether NPC cells engulf lipids released from EBV-induced delipidation of adipocytes using the coculture system. Our results revealed that NPC cells could take up prelabeled neutral lipids released from EBV-challenged adipocytes (Supplementary Fig. S4A). NPC cells grown in supernatant derived from EBV-infected adipocytes grew faster than those derived from unaffected control adipocytes or cancer cell culture medium (DMEM supplemented with 5% FBS; Fig. 4A). EBV-mediated NPC cell proliferation was further confirmed via EdU incorporation assays (Supplementary Fig. S4B). Quantification of the results revealed increased DNA synthesis by the supernatant derived from EBV-infected adipocyte cultures (Supplementary Fig. S4C). This finding suggests that NPC cells utilize environmental lipid mediators as a fuel resource to benefit cancer growth. Interestingly, supernatant fractions of EBV-stimulated adipocytes enhanced activation of the mTOR/p70S6K1/S6 signaling axis, a hallmark for active cell growth, in NPC cells (Fig. 4B). In addition to FFAs, CAA-released pre-inflammatory adipokines are reported to facilitate cancer growth (19, 39). We measured the expression of nine pro-inflammatory adipokines, including IL6, LIF, IL8, TNFα, IL4, SCF (KITL), adiponectin (ADIPOQ), leptin (LEP), and C-C motif chemokine ligand 3 (CCL3; MIP-1α), in adipocytes. EBV infection enhanced the mRNA expression of LIF, IL6, IL8, LEP, SCF, and MIP-1α compared with unchallenged adipocytes (Supplementary Figs. S5A–S5I). Likewise, results of ELISA assays showed that the amount of LIF, IL6, IL8, LEP, and SCF secreted into the culture medium of adipocytes were increased by EBV (Fig. 4C–H). Western blot analysis revealed that treatment of NPC cells with LIF, IL6, IL8, SCF, MIP-1α, or LEP induced activation of p70S6K1 and its downstream molecules (GSK-3α/β and EKR1/2) in NPC cells (Fig. 4I). To determine the upstream regulator of adipokine-mediated p70S6K1 activation, we treated cells with everolimus, rapamycin (two mTOR inhibitors), PD98059 (a MEK1/2 inhibitor), GSK 2334470 (a PDK1 inhibitor), or PF-4708671 (a p70S6K1 inhibitor) prior to adipokine stimulation to block the specific upstream regulator. Results demonstrated that adipokine-mediated activation of p70S6K1 occurred via mTOR (Fig. 4I; Supplementary Fig. S6). To further investigate the role of adipokine in NPC cell growth, we added back selected adipokines to culture medium and evaluated NPC cell proliferation. Our results showed that stimulation with these adipokines substantially enhanced cell proliferation, whereas costimulation with everolimus suppressed adipokine-induced enhancement of cell growth (Fig. 4J). We also examined the expression of adipokine receptors in human NPC (n = 14) and normal nasopharyngeal biopsies (n = 22) by using quantitative real-time PCR assays to assure that tumor cells could respond to adipokine stimulation. Results showed that both NPC tumors and normal nasopharyngeal tissues expressed LIF receptor (LIFR), IL6 receptor (Gp130 and IL-6R1), IL8 receptor (CXCR1 and CXCR2), SCF receptor (c-KIT), MIP-1α receptor (CCR1), LEP receptor (LEPR), and ADIPOQ receptor (ADIPOR1 and ADIPOR2; Supplementary Figs. S7A–S7J). Except for the IL8 receptor, levels of adipokine receptors were slightly higher in tumor cells compared with adjacent normal biopsies. On the basis of these results, we propose that the proinflammatory adipokines or lipid mediators released from EBV-challenged adipocytes have substantial influence on cancer cell growth through activation of p70S6K1 signaling.

Our experiments showed that EBV infection causes delipidation in human adipocytes. The biological consequences of delipidation of adipocytes in the tumor microenvironment (TME) were therefore of considerable interest. In terms of morphology, EBV-challenged adipocytes gradually dedifferentiated to a fibroblast-like phenotype following infection (Fig. 5A). The majority of intracellular LDs vanished on day 20 after EBV stimulation. IHC analysis of EBV⁺-bone marrow metastatic NPC and LELC sections disclosed the presence of elongated adipocytes located close to the invasive fronts of tumors (Fig. 5B), supporting the interesting possibility that fibroblast-like cells are derived from EBV-induced dedifferentiated adipocytes. S100A4 is regarded as a marker of fibroblasts undergoing tissue remodeling and its overexpression is associated with metastasis and poorer prognosis in several types of human malignancies (40, 41). EBV infection enhanced mRNA expressions of S100A4 and fibroblast activation protein α (FAPα) in adipocytes, which increased with after infection time (Supplementary Figs. S8A and S8B). Stronger S100A4 protein signals were detected in dedifferentiated adipocytes expressing LMP1 protein at 8 and 20 dpi compared with preadipocytes (Fig. 5C). Stromal expression of S100A4 in bone marrow metastatic NPC appeared to correlate with FAP-α expression as well as the extent of fibrotic response (Fig. 5D; Supplementary Table S2). Our collective data support the possibility that fibroblasts within the TME are partly derived from adipocyte dedifferentiation induced by EBV. In other words, interactions between EBV and adipocytes aid in remodeling a microenvironment that benefits virus persistence and tumor growth.

On the basis of the current findings, we propose a model linking EBV, adipocytes, and tumor progression (Fig. 6). Stimulation with EBV triggers lipolysis and dedifferentiation of adipocytes that constitute a new fibroblast-like population in TME. Furthermore, released lipid mediators provide an energy resource for tumor cells and proinflammatory adipokines act to activate oncogenic signaling in cancer cells. These factors act in concert to promote a protumorigenic microenvironment, facilitating cancer progression.

It is well recognized that tumorigenesis involves constant communication between cancer cells and surrounding stromal cells. Of the stromal cell types, the most abundant and perhaps most significant are the CAFs. Potential sources of CAFs include local activated fibroblasts, mesenchymal stem cells, and cancer cells that have undergone epithelial-to-mesenchymal transition. CAFs may express fibroblast specific protein 1 (FSP-1/S100A4), FAP-α, smooth muscle actin alpha (α-SMA), PDGF receptor alpha (PDGFRα), and PDGF receptor beta (PDGFRβ), depending on the types of cancer (42). The finding that the LMP⁺- dedifferentiated adipocytes expressed stronger S100A4 supports that EBV-induced dedifferentiated adipocytes may be one of the resources of CAFs associated with fibrosis in TME.

EBV-associated tumors share similar histologic features, including the presence of tumor-infiltrating leukocytes that secrete inflammatory cytokines to facilitate a tumor-promoting microenvironment. Evidence have shown that LMP1 modulates TME through release of cytokines, chemokines, and tumor-promoting factors, such as LIF, IL6, IL8, TGFβ, VEGF, FGF-2, IL1, IL8, CXCL1, and MMPs (43, 44). Importantly, IL6 and LIF have been implicated in cancer cachexia-associated adipose wasting in mouse model (45–47). The increased levels of these inflammatory adipokines could progressively trigger systemic metabolic disorders, such as insulin resistance, lipoatrophy, and cancer cachexia that adversely affect prognosis.

The immune system plays a pivotal role in limiting and controlling the expression of viral genes during EBV latent infection. Adipose tissue is considered as an immune organ that adipocyte-derived factors exert critical influence on immune system. In NPC, EBV assists cancer cells to circumvent local immune surveillance by expressing factors such as vIL-10 and LMP1, which counteract the immune system and activate NF-κB and STAT3 signaling pathways (43). It is reasonable to speculate that EBV hijacks lipid metabolic machinery to create a more permissive proinvasive microenvironment from which reactivation and relapse of infection occur.

On the basis of our findings, EBV infection alters the energy homeostasis of adipose tissue. Yet, viral infection can also trigger innate immune response, releasing cytokines such as IFNs, TNFα, and IL1β, which promote fat depletion (48, 49). Therefore, it is also likely that the lipolytic phenomenon is a consequence resulting from expressions of viral genes and the virus-induced innate immune response.

Fat depletion has been reported to be associated with poorer prognosis of head and neck cancers, including NPC (50). Our novel findings provide information on the mechanisms by which EBV dysregulates metabolism in adipocytes within tumor microenvironment. A number of important issues, such as whether fat dedifferentiation is correlated with an increased level of desmoplastic response and drugs modulating lipid metabolic pathways could prevent metastasis of EBV-associated malignancies, remain to be explored.

No disclosures were reported.

S.-C. Liu: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. N.-M. Tsang: Resources, data curation, formal analysis, validation, methodology. P.-J. Lee: Validation, investigation, methodology. Y.-H. Sui: Validation. C.-H. Huang: Investigation, visualization. T.-T. Liu: Validation.

This work was funded by the Ministry of Science and Technology, Taiwan (MOST 107-2314-B-008-003, MOST 108-2314-B-008-002), a grant from the VGHUST Joint Research Program (VGHUST108-G4-3-1).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.