Activation of the Kynurenine Pathway in Human Malignancies Can Be Suppressed by the Cyclin-Dependent Kinase Inhibitor Dinaciclib

Indoleamine 2,3-dioxygenase (IDO) and tryptophan 2,3-dioxygenase (TDO2) are the key enzymes of tryptophan (TRP) metabolism in the kynurenine pathway (KP). Both enzymes function as indicators of immunosuppression and poor survival in cancer patients. Direct or indirect targeting of either of these substances seems thus reasonable to improve therapy options for patients. In this study, glioblastoma multiforme (GBM) as well as head and neck squamous cell carcinomas (HNSCC) were examined because of their different mechanisms of spontaneous and treatment-induced immune escape. Effects on gene expression and protein levels were examined. Accompanying assessment of TRP metabolites from treated GBM cell culture supernatants was conducted. Our results show a heterogeneous and inversely correlated expression profile of TRP-metabolizing genes among GBM and HNSCC cells, with low, but inducible IDO1 expression upon IFNγ treatment. TDO2 expression was higher in GBM cells, while genes encoding kynurenine aminotransferases were mainly confined to HNSCC cells. These data indicate that the KP is active in both entities, with however different enzymes involved in TRP catabolism. Upon treatment with Temozolomide, the standard of care for GBM patients, IDO1 was upregulated. Comparable, although less pronounced effects were seen in HNSCC upon Cetuximab and conventional drugs (i.e., 5-fluorouracil, Gemcitabine). Here, IDO1 and additional genes of the KP (KYAT1, KYAT2, and KMO) were induced. Vice versa, the novel yet experimental cyclin-dependent kinase inhibitor Dinaciclib suppressed KP in both entities. Our comprehensive data imply inhibition of the TRP catabolism by Dinaciclib, while conventional chemotherapeutics tend to activate this pathway. These data point to limitations of conventional therapy and highlight the potential of targeted therapies to interfere with the cells' metabolism more than anticipated.

Indoleamine 2,3-dioxygenase (IDO) and tryptophan 2,3-dioxygenase (TDO2) are the key enzymes of tryptophan (TRP) metabolism in the kynurenine pathway (KP). Both enzymes function as indicators of immunosuppression and poor survival in cancer patients. Direct or indirect targeting of either of these substances seems thus reasonable to improve therapy options for patients. In this study, glioblastoma multiforme (GBM) as well as head and neck squamous cell carcinomas (HNSCC) were examined because of their different mechanisms of spontaneous and treatment-induced immune escape. Effects on gene expression and protein levels were examined. Accompanying assessment of TRP metabolites from treated GBM cell culture supernatants was conducted. Our results show a heterogeneous and inversely correlated expression profile of TRP-metabolizing genes among GBM and HNSCC cells, with low, but inducible IDO1 expression upon IFNγ treatment. TDO2 expression was higher in GBM cells, while genes encoding kynurenine aminotransferases were mainly confined to HNSCC cells. These data indicate that the KP is active in both entities, with however different enzymes involved in TRP catabolism. Upon treatment with Temozolomide, the standard of care for GBM patients, IDO1 was upregulated. Comparable, although less pronounced effects were seen in HNSCC upon Cetuximab and conventional drugs (i.e., 5-fluorouracil, Gemcitabine). Here, IDO1 and additional genes of the KP (KYAT1, KYAT2, and KMO) were induced. Vice versa, the novel yet experimental cyclin-dependent kinase inhibitor Dinaciclib suppressed KP in both entities. Our comprehensive data imply inhibition of the TRP catabolism by Dinaciclib, while conventional chemotherapeutics tend to activate this pathway. These data point to limitations of conventional therapy and highlight the potential of targeted therapies to interfere with the cells' metabolism more than anticipated.
In the KP, most studies focused on IDO1 because this molecule is amenable to pharmacological intervention (19)(20)(21)(22), and a couple of specific and global IDO inhibitors [including natural compounds (17,23,24)] already entered clinical trials, mostly reporting safe application and efficacy (stable disease at best outcome) (25). Current trials are evaluating the efficacy of IDO1 inhibitors in combination with chemotherapy, radiotherapy, and other immunotherapies including cytotoxic T-lymphocyteassociated protein 4 blockade (11,22). The latter is based on the observation of an enhanced lytic ability of tumor-antigenspecific T cells upon IDO1 inhibition and decreased numbers of local immunosuppressive cells such as regulatory T cells and myeloid-derived suppressor cells (20,26). The efficacy and toxicity data from recent clinical trials with IDO1 inhibitors is reviewed in Yentz and Smith (27). In most cases, however, overall survival was not significantly improved, leaving the future role for this combination therapy in question (28). More key enzymes are involved in TRP metabolism: tryptophan 2.3-dioxygenase (TDO2), a member of the oxidoreductases family, catalyzes the same initial step of the KP as IDO1 (2). Thus, TDO2 has been Abbreviations: CDKi, cyclin-dependent kinase inhibitor; GBM, glioblastoma multiforme; HNSCC, head and neck squamous cell carcinoma; IDO1, indoleamine 2,3-dioxygenase; IFN, interferon; KYAT, kynurenine aminotransferase; KP, kynurenine pathway; PBMC, peripheral blood mononuclear cells; SCC, squamous cell carcinoma; TDO2, tryptophan 2,3-dioxygenase.
shown to be constitutively and highly expressed in various cancer cells such as malignant glioma and HNSCC (29,30). More importantly, TDO2 also has immunomodulatory functions by promoting immune tolerance. This, in turn, promotes survival, growth, invasion, and metastasis and decreases patients' survival (just like IDO1) (13,22,31,32).
In this study, we performed a comprehensive analysis on the expression status of genes belonging to the KP. HNSCC and glioblastoma multiforme (GBM) were picked as prime examples for different spontaneous and treatment-induced immune escape mechanisms. Therefore, expression changes were determined under standard and targeted therapy, and results were compared among each other.

IFNγ Stimulation
Cells were cultured in six-well plates or ibidi chamber slides, incubated overnight and treated with IFNγ (50 ng/ml, Immunotools, Friesoythe, Germany) for 24 and 72 h, respectively. Thereafter, cells were harvested and further processed.

Apoptosis/Necrosis Assay
A Yo-Pro-1/PI-based assay for discriminating early apoptotic, late apoptotic, and necrotic cells was applied as described before (34).

Hemolysis Assay
Hemolytic activity of Dinaciclib was determined by hemoglobin release from whole blood cells after 2 h of incubation. Briefly, whole blood of healthy donors (N = 5) was seeded in 96well plates and treated with increasing Dinaciclib doses (ranging from 1, 5, and 10 µM). Negative controls were left untreated, and positive controls (=maximum lysis) were treated with 1% sodium dodecyl sulfate. Following the incubation period, cell-free supernatants were transferred into a new 96-well plate, and absorption was measured on a plate reader at 560 nm (reference wave length, 750 nm). Hemolytic activity was quantified according to the following formula and corrected for spontaneous hemolysis (=untreated controls): %Hemolysis = ((OD 560nm sample −OD 560nm buffer)/OD 560nm max − OD 560nm buffer)×100 In addition, peripheral blood mononuclear cells' (PBMC) viability (N = 5) were determined by Calcein AM staining. This was done upon 24 h incubation at the above-mentioned doses. Fluorescence measurement and quantification were done as described (34).

IDO1 Immunofluorescence
Tumor cells were treated with 50 ng/ml of IFNγ (Immunotools), TMZ, Cetuximab, or Dinaciclib for 24 h in chamber slides, respectively. Cells were washed with phosphate-buffered saline, fixed in 4% paraformaldehyde w/o methanol (Thermo Scientific, Darmstadt, Germany) for 20 min, washed again, followed by cell permeabilization in 0.3% Triton X−100/5% normal bovine serum in phosphate-buffered saline for 60 min. Cells were then incubated overnight at 4 • C in monoclonal rabbit IDO1 primary antibody (1:100; Cell Signaling Technology, Frankfurt/Main, Germany). Cells were washed, labeled with fluorochromeconjugated secondary antibody using goat antirabbit secondary antibody (1:250, Boster Biological Technology, Pleasanton CA, USA), and incubated in the dark for 2 h. Cell nuclei were stained with 4 ′ ,6-diamidino-2-phenylindole (DAPI), and cells were analyzed with a Zeiss LSM-780 Confocal Laser Microscope (Zeiss, Jena, Germany). Quantification of staining intensity was done using the ImageJ software. Therefore, channels were split into red, green, and blue. Subsequently, integrated density profiles of the same size were measured in the green channel.

IDO1 Immunohistochemistry on Patients' Tumor Samples
Primary antibody against IDO1 (rabbit IgG, clone D5J4E, Cell Signaling Technology, dilution 1:200) was used. All samples were pretreated for 20 min at 97 • C and pH 6.9. Standard immunoperoxidase technique was applied using an automated immunostainer (DAKO link) with diaminobenzidine as chromogen. IDO1 expression was defined as cytoplasmatic and membranous staining in >1% inflammatory cells.

Quantification of Tryptophan, Kynurenine, and Kynurenic Acid in Cell Culture Supernatant by Liquid Chromatography Tandem Mass Spectrometry System
The basis for the measurement was the method of Fuertig et al. which was adapted to the system used here (35).
The eluate between 0.5 and 9 min was introduced into the mass spectrometer and analyzed in MRM mode. The ion spray voltage (IS) was 4,000 V, the curtain gas flow was 40.0 psi, and the ion source temperature were set at 550 • C.
Internal standards were used for metabolite quantification ( Table 1). Data analysis, including peak integration and concentration determination, was performed with Analyst software (Version 1.5.1, AB Sciex, Darmstadt, Germany).

Statistical Analysis
All values are reported as mean ± SD. After proving the assumption of normality, differences between controls and treated cells were determined using the unpaired Student's ttest. If normality failed, the non-parametric Mann-Whitney Utest was applied. Statistical evaluation was performed using GraphPad PRISM software, version 5.02 (GraphPad Software, San Diego, CA, USA). In case of multiple comparisons, two-or one-way ANOVA on ranks (Bonferroni's multiple comparison test) was used. The criterion for significance was taken to be p < 0.05. Frontiers in Immunology | www.frontiersin.org

Basal IDO1 and Related Genes in GBM and HNSCC Cell Lines
While IDO1 itself is not the only mechanism by which tumors can resist immune-mediated killing, we studied the expression status of different KP-related genes on a panel of human GBM and HNSCC cell lines. These experiments revealed not only differences between both entities but also a heterogeneous profile of all tested genes among cell lines (Figure 1). IDO1 was differently expressed by most glioma samples (11/13) analyzed. In general, IDO1 was only detectable at very low levels (Figure 1). TDO2, the other rate-limiting enzyme of the KP (36), was constitutively expressed by all glioma samples, and expression was even higher in comparison to IDO1. Generally, expression status for TDO2 and kynurenine hydrolase (KYNU) was higher in GBM, while HNSCC expressed more kynurenine aminotransferases (KYAT) (Figure 1). Hence, these data indicate that the KP is active in both entities, with however different enzymes being involved in TRP catabolism.
Still, tumor cell lines grown in vitro not necessarily represent the in vivo situation; we therefore analyzed the IDO1 abundance in clinical resection specimens (Figure 2). In GBM, IDO1 was detectable in one of three cases (representative images are shown in Figure 2). By contrast, HNSCC samples presented with IDO1 but only on a small fraction of tumorinfiltrating lymphocytes (Figure 2). Although not analyzed systematically, the only HPV positive case in this small cohort showed highest IDO1 abundance, nicely reflecting the tumors' immunogenicity (11,37).

Gene Expression and Protein Changes Upon IFNγ Stimulation
IDO1 is an IFNγ-inducible enzyme. Upon stimulation, the KP is activated to induce immunosuppression. In vitro stimulation with IFNγ mimics the in vivo situation of an inflammatory microenvironment. Hence, upon immunemediated inflammation, IDO1-negative tumor cells may upregulate IDO1 as resistance mechanism. Using five individual GBM cell lines, IDO1 expression was inducible in all cases ( Figure 3A). Upregulation of IDO1 was high on protein levels in HROG05 cells and marginal in HROG63 (Figure 3B). TDO2 and KYAT3 were suppressed upon IFNγ stimulation in three of five samples and hardly detectable in one cell line, supporting data from a recent publication (38). KYNU was not affected by IFNγ stimulation (Figure 3A).
Just as in GBM, IDO1 was inducible in HNSCC cells ( Figure 3A). Immunofluorescence revealed focal expression of singular cells with different intensity (Figure 3B). Of note, IFNγ stimulation even induced upregulation of KYAT1, KYAT2, KYAT3, and KYAT4 ( Figure 3A and data not shown), most likely constituting a compensatory mechanism as described before in experimental autochthonous tumor models (39).

Interference With the KP of Cytostatic and Targeted Therapies
Next, we examined whether cytostatic and targeted drugs have an influence on the KP. For GBM, TMZ was chosen, and for HNSCC, 5-FU, Cisplatin, Gemcitabine, as well as Cetuximab were used. As a targeted yet still experimental agent, the potent and specific CDKi Dinaciclib was applied to cells of both entities.
Before this experiment, drug doses were carefully tested in dose-response analyses (data not shown) along with discrimination of apoptosis and necrosis. Generally, drugs used in this study tended to induce necrosis, while apoptosis, if present, was only detectable at early time points. Exemplary results for the HNSCC cell line Detroit-562 are given in Supplementary Figures 1A,B. While cytostatics are well-known to affect normal cells' viability, the impact of the CDKi Dinaciclib on immune and red blood cells is less clear. We therefore performed a hemolysis and leukocyte viability assay. In this experiment, no toxicity was seen against normal cells (Supplementary Figure 1C). Even at high concentrations, Dinaciclib impaired cellular viability/integrity only marginally (Supplementary Figure 1C).
TMZ is an oral alkylating agent that methylates DNA at the O 6 position of guanine causing cell cycle arrest at G2/M. It is used as standard of care for GBM. However, acquired resistance, a process not fully understood, leads to major limitations in treatment. Here, TMZ downregulated IDO1 in three of five GBM cell lines but led to increased expression in HROG52 and HROG63-a paired GBM cell line established from the very same patient (primary lesion and upon relapse) (Figure 4). Gene expression of KYAT2, KYAT4, and KMO was heterogeneous. Generally, there was a trend toward higher expression of those genes but with cell-linespecific differences (e.g., KYAT3: p < 0.05 vs. control in HROG05 cells; Figure 4). KYNU expression was not affected by TMZ (data not shown). Interestingly, the combination of IFNγ and TMZ that mimics the in vivo situation led to similar or even stronger IDO1 upregulation compared to IFNγ alone in two out of four glioma samples (Figure 5). Adding Dinaciclib to either IFNγ or TMZ lowered the mRNA expression of IDO1 massively. Other KP-related genes like In HNSCC cells, Cetuximab was the only IDO1-inducing substance (exemplary results for Detroit-562 cells are given in Figure 6). Beyond that, the cytostatics as well as Cetuximab induced at least one of the KP-related genes (p < 0.05 vs. control), implicating activation of this pathway via different effectors. By adding Dinaciclib to cytostatic drugs, this effect was abrogated, even in the presence of IFNγ (Figure 6 and data not shown). Of note, Dinaciclib alone as well as in combination with other substances effectively suppressed all KPrelated genes, implying inhibition of the TRP catabolism by this CDKi.

Dinaciclib Blocks IFNγ-Induced IDO1 Expression in GBM and HNSCC Cells
Considering the active downregulation of KP-related genes by Dinaciclib, we investigated whether this CDKi is able to inhibit or reverse IFNγ-induced IDO1 upregulation in GBM and HNSCC cells on a protein level. TMZ and Cetuximab were included as active inductors of IDO1 and associated KP-related genes.
IFNγ and selected drugs were added simultaneously for 72 h. Dinaciclib effectively blocked IFNγ-induced IDO1 protein in both entities, while TMZ alone as well as the combination with IFNγ strongly enhanced IDO1 protein level (Figure 7). Hence, mRNA expression data were nicely confirmed.
When Dinaciclib was combined with IFNγ and TMZ, the IDO1-inducing stimulus of these latter substances was far too strong to be suppressed (Figure 7). However, the low number of residual cells in this combination hints toward additive or even synergistic effects independent from IDO1 (Figures 7A,B).
While IDO1 was highly inducible in GBM cells only, we then determined protein level upon IFNγ-prestimulation approaching the in vivo situation. The cytotoxic effect of Dinaciclib was preserved; however, levels of IDO1 enzyme were not significantly altered (Supplementary Figures 2A,B). Comparable results were obtained for TMZ. Virtually, all residual cells showed positive staining; still there was a trend toward lower intensity in monotherapy and in combination (Supplementary Figure 2B).
Taken together, the CDKi Dinaciclib is able to block IFNγmediated and thus most likely even chemotherapy-induced IDO1 upregulation in GBM and HNSCC cells. However, blunt interference with this TRP-metabolizing enzyme is unlikely.

Treatment Induced Influence on KP-Related Metabolites
Our data revealed IDO1 induction by TMZ, which is reversible by Dinaciclib. Thus, we examined the influence on KP-related metabolites in GBM cell lines.
TRP, KYN, and the downstream metabolite kynurenic acid (KYNA) were quantified by MS using cell culture supernatants of GBM cell lines (Figures 8A,B). TRP was catabolized after 24 h from all cell lines among all treatment regimens. Adding TMZ or Dinaciclib in monotherapy marginally affected TRP consumption as well as KYN and KYNA production. Stimulation with either IFNγ or a combination of TMZ resulted in greatly enhanced TRP depletion and increased KYN levels, although to varying degrees in the different cell lines (Figure 8A). Small amounts of KYNA were produced constitutively and to a greater extent after IFNγ mono-and TMZ combination in all cell lines (Figure 8A). In contrast, KYNA level remained unchanged upon Dinaciclib in combination with IFNγ, confirming immunofluorescence results (please see Figure 7 for details). The same was true for the KYN/TRP ratio, being only affected in samples treated with IFNγ as well as the combination of IFNγ and TMZ ( Figure 8B).
These data underline our gene and protein expression data. The CDKi Dinaciclib is directly or indirectly capable of blocking the KP. TMZ particularly in combination with the proinflammatory cytokine IFNγ accelerates TRP consumption accompanied by KYN and KYNA production in GBM cells.

DISCUSSION
The finding that high IDO1 expression is associated with shorter survival in cancer patients made IDO1 a promising target either by specific inhibitors or indirectly by immunomodulation.
A recent study described dramatically suppressed tumor growth upon IDO1 knockdown by increasing the number of CD4 + and CD8 + T cells in murine GBM models (9). However, the exact mechanisms underlying IDO1 and thus TRP metabolism along the KP remain unclear. Therefore, we focused on the expression of IDO1 and IDO-related KP genes and their potential involvement in immune evasion in experimental models of HNSCC and GBM.
We were able to show that the KP is active in both entities, with different enzymes involved in TRP catabolism. Of note, basal IDO1 expression was low and inversely correlated with TDO2. In the only prior study on primary GBM cultures, similar results were described with constitutive TDO2 expression in most GBM cell cultures (29). In here, TDO2 likely promotes tumor growth by suppressing antitumor immune responses (2,31). KP products are considered as therapeutic targets because IDO1 and other genes of the TRP metabolism are not expressed in healthy brain tissue, but gradually increase with GBM dedifferentiation (low vs. high grade GBM). In HNSCC, different results on IDO1 are documented, and expression is heterogeneous among different HNSCC cell lines. Of note, IDO1 abundance of primary resection specimen and cultured cells seems to be independent from anatomical site and HPV status (40). Still, IDO1 is a useful marker for progression of in oral squamous cell carcinoma (41). In esophageal squamous cell carcinoma, progression and metastasis correlates with strong inflammation at the tumors' invasive front and disturbed TRP metabolism (42). These cumulative data highlight the biological relevance of the KP in malignancies and may explain why IDO1 is barely detectable upon long-term in vitro culture. By mimicking the inflamed microenvironment and thus taking a step closer to the in vivo situation, IFNγ was added as strong IDO1 inductor (43). While GBM cells responded with the expected IDO1 upregulation on mRNA expression and protein level as well as accelerated TRP consumption, this molecule was barely inducible in HNSCC cells. It is conceivable that this is due to the duration of in vitro culture. GBM cells were established recently and thus used in defined low passages (<P40), whereas half of the HNSCC cell lines were long-term cultures with more or less unknown passage [Detroit-562 as well as UT-SCC14 and UT-SCC15 (44) are the only exceptions; <P40]. Cell lines may acquire additional mutations overtime changing their protein expression. Another in vitro limitation is that experiments were conducted without immunological pressure. In vivo studies are desirable to verify the results.
Indirect effects of TRP metabolism include interference with other biological functions like migration, angiogenesis, and cell growth regulation (18,40). To investigate the influence of anticancer drugs on TRP catabolism, we performed a comprehensive analysis using conventional chemotherapy (TMZ, 5-FU, Cisplatin, Gemcitabine) and targeted drugs (Cetuximab, Dinaciclib). The KP-related gene expression and metabolites were determined in residual cells. In GBM, the standard of care drug TMZ was applied either with or without IFNγ stimulation. While this substance affected IDO1 on the expression level, the amount of the resulting protein increased. This may be explained by either increased protein's half-life due to a reduced rate of degradation or the preferential translation during cellular stress. In previous studies, exposure of several cultured human malignant glioma cell lines, primary neurons, and a neuroblastoma cell line to IFNγ reduced TRP levels in culture medium accompanied by increased IDO1 expression and KYN production (29,45). Our results confirm these data, and in addition, we were able to demonstrate that IFNγ stimulation in combination with TMZ stimulated KYN and KYNA production and TRP catabolism in GBM cell cultures. The increase in TRP catabolism and KYN production (KYN/TRP ratio) is widely used as indirect indicator of the cumulative activities of TDO2, IDO1, and IDO-2 (38,46). The KP in brain tumors is likely triggered by IFNγ from immediate surrounding tissue (29,47,48). Thus, IDO1 expression in brain tumor cells is likely to be triggered when IFNγ is produced from activated T cells and/or microglia and neurons. Furthermore, gliomas and glioneuronal tumors have an elevated tryptophan uptake and catabolism in vivo (49).
Given our observation on a further enhanced KP activity upon TMZ treatment, this might provide an explanation of (acquired) drug resistance and final relapse. Hence, IDO1 blocking agents should be investigated in TMZ-tailored therapeutic approaches.
In HNSCC cells, KP activation was different. KP-related genes were exclusively induced by standard drugs, and only Cetuximab induced IDO1. Additional upregulated genes involved kynurenine aminotransferases, responsible for synthesizing a neuroprotectant, and KMO. While the specific biochemical activity of these molecules and biological relevance in cancer is barely examined, we interpret this result as one possible mechanism of resistance upon therapy-a finding quite common after conventional chemotherapy and usually also being associated with poor response toward neoadjuvant therapy in other entities (50).
Mechanistically, this can be attributed to the secretion of proinflammatory substances, such as prostaglandin E2 or highmobility group protein B1 by dying tumor cells, secondary contributing to KP activation. By accumulating TRP, toxic metabolites of tumor cells actively shape an immunosuppressive microenvironment. Breaking down this shield is one of the main objectives in pharmacological inhibition of KP. Questions remain why most inhibitors failed in clinical trials, and mechanisms are only just beginning to become clear. A fact worth mentioning is the functional redundancy of IDO1, IDO-2, and TDO2 (51), augmenting the risk of mechanistic bypass.
Dinaciclib is a potent and specific CDK inhibitor of CDK1, CDK2, CDK5, and CDK9. Preclinical studies showed that this inhibitor is capable of decelerating tumor growth in numerous cancer entities via cell cycle arrest and apoptosis induction (52,53). In our study, Dinaciclib was the only KP-inhibiting substance tested here. Of note, impairment of the KP was independent from the combination partner, and this CDKi effectively suppressed IFNγ-induced IDO1 upregulation after simultaneous treatment. While this result was completely unexpected and has-to the best of our knowledge-not been described previously, our data do not support the idea of blunt interference with the KP. GBM cells with strong IDO1 expression showed only marginally reduced IDO1 protein level after Dinaciclib treatment. This effect might be boosted after long or repeated treatment cycles. In line with these findings, several preclinical studies already proposed synergistic effects of selective and unselective IDO1 inhibitors when administered in conjunction with chemo-and/or radiotherapy (4). This may finally have impact for secondor third-line immunotherapeutic approaches. Therefore, the late KYN/TRP index is indeed a relevant clinical benchmark providing prognostic value for GBM patients (54).
Summarizing our findings, we provide evidence for the relevance of TRP catabolism in malignancies especially in the context of standard therapy. The CDKi Dinaciclib was identified as indirect KP inhibitor. Lastly, specific KP inhibition may increase the efficacy of standard drugs by restoring immune function and thus improve patients' outcome.

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

AUTHOR CONTRIBUTIONS
CR performed experiments, analyzed data, and participated in manuscript writing. BS and AZ performed immunohistochemistry and analysis, and provided images. HK, JG, NI, and FS performed experiments and analyzed data. GD performed LC-MS analyses. CC and CJ participated in paper finalization and critically revised the manuscript. DS and EW critically revised the manuscript. CM designed study, the outline of the manuscript, performed data interpretation, and wrote the manuscript.