Identification of a Novel Pathway of Transforming Growth Factor-β1 Regulation by Extracellular NAD+ in Mouse Macrophages

Background: Both extracellular NAD+ and the cytokine TGF-β1 are anti-inflammatory. Results: NAD+ increases both active and latent TGF-β1 in mouse macrophages. A mathematical model partially explains the complex effects of NAD+ on TGF-β1. Conclusion: NAD+ is a novel modulator of TGF-β1. Significance: Combined in vitro and in silico approaches may help elucidate novel pathways of TGF-β1 regulation. Extracellular β-nicotinamide adenine dinucleotide (NAD+) is anti-inflammatory. We hypothesized that NAD+ would modulate the anti-inflammatory cytokine Transforming Growth Factor (TGF)-β1. Indeed, NAD+ led to increases in both active and latent cell-associated TGF-β1 in RAW 264.7 mouse macrophages as well as in primary peritoneal macrophages isolated from both C3H/HeJ (TLR4-mutant) and C3H/HeOuJ (wild-type controls for C3H/HeJ) mice. NAD+ acts partially via cyclic ADP-ribose (cADPR) and subsequent release of Ca2+. Treatment of macrophages with the cADPR analog 3-deaza-cADPR or Ca2+ ionophores recapitulated the effects of NAD+ on TGF-β1, whereas the cADPR antagonist 8-Br-cADPR, Ca2+ chelation, and antagonism of L-type Ca2+ channels suppressed these effects. The time and dose effects of NAD+ on TGF-β1 were complex and could be modeled both statistically and mathematically. Model-predicted levels of TGF-β1 protein and mRNA were largely confirmed experimentally but also suggested the presence of other mechanisms of regulation of TGF-β1 by NAD+. Thus, in vitro and in silico evidence points to NAD+ as a novel modulator of TGF-β1.

Inflammation is a complex process in which various potent mechanisms that can control infection, injury, and proliferative diseases must be kept in check (1). Studies over the past decade have focused on the release of damage-associated molecular patterns (DAMPs) 4 from cells, a class of molecules that signal a disruption of cellular homeostasis. Prototypically, these DAMPs are proteins or other cellular constituents that carry out housekeeping functions normally but are released in settings of stress, inflammation, or injury. In turn, these agents stimulate, propagate, or potentiate both innate and adaptive immune responses (2). Recent studies have suggested that one such mediator may be ␤-nicotinamide adenine dinucleotide (NAD ϩ ), a ubiquitous cellular constituent that is used by cells as an electron acceptor (or, in its reduced form, NADH, as an electron donor) in a wide variety of enzyme-catalyzed redox reactions. These actions of NAD ϩ occur in multiple cell types secondary to the formation of cyclic adenosine dinucleotide ribose (cADPR) from NAD ϩ , with subsequent release of Ca 2ϩ (3). Importantly, NAD ϩ has been found to exert a profound anti-inflammatory activity that it appears to share with nicotinamide (4). The mechanisms by which these anti-inflammatory actions are carried out, however, remain poorly understood.
Transforming Growth Factor ␤1 (TGF-␤1) is a cytokine that belongs to a family of three related isoforms, all of which exert crucial biological functions. Of these three isoforms, TGF-␤1 is the most prominent in the control of inflammation and immunity (5). The numerous biological functions of all TGF-␤s require a set of post-translational modifications termed "activation." The bioactive forms of all TGF-␤s are 25-kDa homodimers produced from ϳ50-kDa monomers that dimerize to form the ϳ100-kDa TGF-␤ precursor. This precursor is cleaved intracellularly by furin proteases to yield the 25-kDa active TGF-␤ dimer, which remains associated with the remaining portion of its own pro-form, the latency-associated peptide (ϳ75 kDa). This complex is termed "latent TGF-␤" and is secreted in this form. Other proteins, such as latent TGF-␤binding proteins (LTBP, which targets TGF-␤ to the extracellular matrix) or ␣2 macroglobulin (which is associated with circulating TGF-␤1) can bind to this complex, creating the socalled large latent complex. Latent TGF-␤ is activated by a process that involves dissociation and degradation of latencyassociated peptide by proteins (e.g. plasmin and transglutaminase), heat, chaotropic agents, and acid as well as oxygen and nitrogen free radicals (6,7). Although TGF-␤1 can autoinduce its own expression at the mRNA level, the post-translational control of TGF-␤1 through activation is arguably the most potent regulatory mechanism for this cytokine (6,7).
As early as 1978, an "NAD ϩ -splitting enzyme" was reported in macrophages (8 -10). Because cytokines, radiation, and free radicals can lead to the activation and increased expression of latent TGF-␤1 in macrophages (11)(12)(13)(14), we hypothesized that extracellular NAD ϩ could exert a similar effect. We further hypothesized that the mechanism by which NAD ϩ would act would involve the generation of cADPR from extracellular NAD ϩ . Finally, we hypothesized that cADPR would exert its effects via the stimulation of Ca 2ϩ . To better understand the complex interplay among NAD ϩ , cADPR, Ca 2ϩ , and active and latent TGF-␤1, we constructed both statistical and mathematical models and validated some predictions from these models in vitro. These computational models also suggested the existence of as yet unknown mechanisms by which NAD ϩ can augment TGF-␤1. Taken together, our results demonstrate a novel pathway for TGF-␤1 activation via NAD ϩ and its metabolites and highlight the utility of mathematical modeling for discovering novel biological mechanisms.

EXPERIMENTAL PROCEDURES
Reagents-The DMEM and HEPES buffer used for cell culture were purchased from BioWhittaker-Lonza (Baltimore, MD). The penicillin/streptomycin and the L-glutamine media additives were purchased from Invitrogen. Fetal bovine serum was purchased from Gemini BioProducts (West Sacramento, CA) and was used due to its low TGF-␤1 content (data not shown). Antibodies to human TGF-␤1 and human latency-associated peptide were purchased from R&D Systems (Minneapolis, MN). The secondary antibodies, the normal serum, the ABC enzyme conjugate kit, and the diaminobenzidine brown stain kit were all purchased from Vector Laboratories (Burlingame, CA). ␤-Nicotinamide adenine dinucleotide (NAD ϩ ), 3-cADPR, 8-Br-cADPR, bradykinin acetate, and verapamil hydrochloride were purchased from Sigma. 1,2-bis-(o-Aminophenoxy)-ethane-N,N,-NЈ,NЈ-tetraacetic acid tetraacetoxy-Methyl ester (BAPTA-AM) was from Biomol Research Labs (Plymouth Meeting, PA). Ionomycin was purchased from Alomone Labs Ltd (Jerusalem, Israel).
Cell Culture and Experimental Treatments-RAW 264.7 mouse macrophage-like cells (American Type Culture Collection, Manassas, VA) and primary peritoneal macrophages isolated from C3H/HeJ (TLR4-mutant, n ϭ 8 animals) and C3H/ HeOuJ (wild-type controls for C3H/HeJ, n ϭ 8 animals) mice (The Jackson Laboratory, Bar Harbor, ME) were cultured and plated in DMEM ϩ 1% FBS containing L-glutamine and penicillin/streptomycin. The FBS used was previously determined to have the lowest levels of TGF-␤1 of various manufacturers and lots (data not shown) in order to minimize the exposure of cells to TGF-␤1 that could auto-induce further expression of TGF-␤1. The cells were cultured in 1% FBS as this was the lowest concentration of serum that allowed for cell proliferation while minimizing exposure to TGF-␤1. The passage number was kept Ͻ18 for the same reason; as the passage number increased, the basal immunocytochemical expression of both active and latent TGF-␤1 increased (data not shown). The cell culture and semiquantitative immunocytochemical detection of active and latent TGF-␤1 have been described previously (15). In brief, the cells were plated in eight-well Lab Tek TM Chamber Slide TM tissue culture plates (NalgeNunc International, Rochester, NY), which allowed for several experimental conditions per well and also for immunocytochemistry at the end of the culture period. The cells were plated at a concentration of 2 ϫ 10 5 cells/ml in 500 l and were maintained in 5% CO 2 in a humidified atmosphere until adherent. The cells were then treated with either NAD ϩ or 3-cADPR in the presence or absence of other pharmacological agents and analyzed for active/latent TGF-␤1 as indicated. In those experiments where Ca 2ϩ antagonists were used, the medium was removed before further treatment with the Ca 2ϩ agonists or NAD ϩ as indicated. For time-course experiments, the cells were treated with NAD ϩ and were left to incubate at 37°C for 1, 2, 6, 8, 12, or 24 h and then processed for immunocytochemistry as described previously (15). Each treatment was carried in duplicate to allow for the eventual parallel immunocytochemical detection of active and latent TGF-␤1 (see below).
Immunocytochemistry-In previous studies on TGF-␤1 activation in macrophages, we utilized dual immunofluorescence to detect active versus latent TGF-␤1 (13,14). However, we found that NAD ϩ caused macrophages to autofluoresce in a dose-dependent manner, and this artifactual effect confounded our ability to detect active and latent TGF-␤1 (data not shown). Accordingly, we carried out separate immunostainings for active and latent TGF-␤1 using the diaminobenzidine method as described previously (15). In brief, at the conclusion of the incubation period, the cells were fixed in 70% ethanol for 30 min. Endogenous peroxidases were inhibited using 0.3% hydrogen peroxide solution for 30 min. Nonspecific antibody reactivity was blocked as follows. The wells that were stained for active TGF-␤1 were blocked with 1.5% goat serum, whereas the wells stained for latent TGF-␤1 were blocked with 1.5% rabbit serum; all blocking was performed for 20 min. The cells were then incubated with primary antibody specific for either active TGF-␤1 (chicken anti-human active TGF-␤1) or latent TGF-␤1 (goat anti-human latent TGF-␤1) for 30 min. The cells were then incubated with secondary antibodies as follows. For wells stained for active TGF-␤1, the secondary antibody solution consisted of ϳ0.5% goat anti-chicken antibody mixed with 1.5% goat serum and 1 ml of PBS. For wells stained for latent TGF-␤1, the secondary antibody solution consisted of 0.5% rabbit anti-goat antibody mixed with 1.5% rabbit serum and 1 ml of PBS. The incubation time for the secondary antibody was 30 min. The cells were then exposed to the ABC enzyme con-jugate for 30 min, after which they were stained with a diaminobenzidine brown stain for 90 s. The cells were then incubated with a hematoxylin blue counterstain for 20 s. The slides were then dehydrated and cleared in a series of graded ethanol solutions (1 wash in 70%, 2 washes in 95%, and 2 washes in 100%) followed by 1 wash in xylene. The slides were then dried overnight and mounted with Permount TM (Fisher).
Quantification of Immunocytochemistry-Quantification of active/latent TGF-␤1 immunostaining was carried out as described previously (15). In brief, three images from each well were captured using a Zeiss Axioskop 40 (Göttingen, Germany) equipped with a digital camera and Motic Image 2000 software. The images were captured at 400ϫ magnification and analyzed using the Image J TM 1.35c freeware (NIH, Bethesda, MD) and Color Inspector 3D plugin. Because the two primary colors in each picture were brown and blue, data were gathered using a brown to blue ratio. Due to the diaminobenzidine brown stain, the color brown indicated the presence of TGF-␤1, whereas the color blue from the hematoxylin counter-stain indicated the absence of TGF-␤1. We used the histogram setting in Color Inspector 3D to cluster similar color values, thus simplifying the quantification process. We then determined different values to mean "blue" or "brown" and counted the percentage of the screen taken up by pixels that fell under the blue category or the brown category. The ratio of brown positive/blue-positive cells in 2-3 fields per image was calculated utilizing the following formula: brown-to-blue ratio ϭ (sum 2-3 fields brown/sum 2-3 fields blue) ϫ 100 [%].
Each complete immunocytochemistry assessment was repeated at least three times/experiment, and the mean brown positive/blue positive ratio was calculated. A higher brown-toblue value indicates a higher cell-associated expression of either active or latent TGF-␤1. In some experiments the final values for active or latent TGF-␤1 were determined as -fold change (versus non-treated control cells) Ϯ S.E. of at least three independent experiments as indicated.
Northern Blot Analysis-Total RNA was isolated from treated and control RAW 264.7 cells using an UltraSpec TM RNA isolation reagent from Biotecx Laboratories, Inc. (Houston, TX). Northern blot analysis of TGF-␤1 mRNA levels was carried out using a 1.6-kb TGF-␤1 cDNA probe derived from the mouse full-length sequence (Image clone #3586216 from Invitrogen, accession number BC013738) after Xba1 and Sal1 digestion as described previously (15). This probe cannot distinguish between active and latent TGF-␤1.
Statistical Analysis-All data are presented as the means Ϯ S.E. of n number of independent experiments as shown in each figure legend. Unless otherwise indicated the data were analyzed by one-way analysis of variance followed by a post hoc test (Student-Newman-Keuls Method) as appropriate using Sigma-Plot for Windows Version 11.0 (Systat Software Inc., San Jose, CA).

Data-driven Modeling and Mechanistic Mathematical
Modeling-A statistical model was generated from the dosecurve and time-course data as follows.
The responses of active and latent TGF-␤1 could be explained by two independent variables, concentration of NAD ϩ (values of 0, 10, 100, or 1000 M) and time (values of 1, 2, 4, 6, 8, 12, or 24 h). At each (concentration, time) pairing, 5-7 independent observations (experimental repeats) were ultimately obtained; some outliers were omitted. These responses were modeled as a bivariate gaussian vector, and the explanatory (independent) variables were interpreted as ordered factors. Orthogonal polynomial contrasts were used as columns of the factorial design matrix. Various possible model choices were explored, and second-degree polynomial fits with interactions provided a good explanation of the data. Specifically, we allowed linear and quadratic effects in concentration and time as well as possible interactions between these effects. In parallel, a mechanistic, equation-based model was generated from the dose-curve and time-course data, as follows.
The system was modeled in Matlab using ordinary differential equations (ODEs) with mass action kinetics for the following seven variables: N (NAD ϩ ), C (Ca 2ϩ ), Tl (latent TGF-␤1), Ta (active TGF-␤1), Tm (TGF-␤1 mRNA), and X 1 and X 2 (unknown intermediaries). All variables were presumed to have exponential decay rates. In addition, the following interactions were included in the ODEs. 1) NAD ϩ is consumed to produce Ca 2ϩ . 2) Latent TGF-␤1 is produced by translation from TGF-␤1 mRNA and consumed by the Ca 2ϩ -catalyzed conversion to active TGF-␤1. 3) Active TGF-␤1 is produced by the Ca 2ϩ -catalyzed conversion from latent TGF-␤1 and is consumed by a threshold-dependent constant rate. 4) The variables X 1 and X 2 represent a conglomeration of the intermediary steps going from NAD ϩ signaling to TGF-␤1 mRNA induction. 5) X 1 is produced according to a Hill function of NAD ϩ , and X 2 is produced by X 1 . 6) TGF-␤1 mRNA is subsequently produced by X 2 .
The parameters of the model were estimated such that the latent TGF-␤1 response becomes dose-dependent at ϳ12 h, as observed experimentally. The conceptual development of the ODE model and a complete simulated time-course resulting from the ODE model for active TGF-␤1, latent TGF-␤1, TGF-␤1 mRNA, NAD ϩ , and Ca 2ϩ are available as supplemental Figs. S1 and S2.

RESULTS
NAD ϩ Increases the Expression of Active and Latent TGF-␤1 in RAW 264.7 Macrophage-like Cells-We first tested the hypothesis that treatment of macrophages with extracellular NAD ϩ would lead to increased expression of both active and latent TGF-␤1 in macrophages, utilizing the RAW 264.7 mouse macrophage-like cell line. Various studies have suggested that the activation of latent TGF-␤1 is often best assessed immunocytochemically (13)(14)(15), and so we utilized this method initially. In prior studies (15), we have shown concordance between immunoblotting and this immunocytochemistry-based method of differential detection of active versus latent TGF-␤1. As shown in Fig. 1, treatment with 10 -1000 M NAD ϩ for 2 h led to a dose-dependent increase in immunocytochemically detectable latent (panels B-D) and active (panels F-H) TGF-␤1 compared with control resting cells (panels A and E).
NAD ϩ and Its Extracellular Metabolite cADPR Increase the Expression of Active and Latent TGF-␤1 in Primary Mouse Peritoneal Macrophages-We next sought to confirm in primary macrophages the basic finding of NAD ϩ /cADPR-mediated increase in TGF-␤1 and to determine whether or not the effect of NAD ϩ /cADPR was due to LPS contamination, again using our previously published immunocytochemical method coupled with image analysis and quantification (15).
As seen in Fig. 2, treatment with either 100 M NAD ϩ or 10 nM stable cADPR analog 3-deaza-cADPR (17) for 1 h led to increased active (Fig. 2, A and B) and latent TGF-␤1 (Fig. 2C) in isolated peritoneal macrophages from both wild-type (C3H/ HeOuJ) and LPS-hyporesponsive, TLR4-mutant (C3H/HeJ) mice at 1 h post-treatment. Moreover, there were no statistically significant differences between C3H/HeJ and C3H/HeOuJ macrophages with regard to expression of NAD ϩ /cADPR-induced active TGF-␤1 (Fig. 2B). Taken together, these results show that NAD ϩ and 3-cADPR can induce and activate TGF-␤1 in primary macrophages in vitro and strongly suggest that LPS contamination is at most an extremely minor contributor to the effects of NAD ϩ and 3-cADPR on TGF-␤1.
The Extracellular NAD ϩ Metabolite cADPR Increases the Expression of Both Active and Latent TGF-␤1 in RAW 264. 7 Macrophages-As seen in Fig. 3A, treatment with 100 M NAD ϩ for 2 h led to increased active TGF-␤1. To determine whether this effect of NAD ϩ depends on the prior conversion to cADPR (3), we treated RAW 264.7 cells with NAD ϩ in the presence of the stable, cell-permeable cADPR antagonist 8-Br-cADPR (16). As seen in Fig. 3A, 10 M 8-Br-cADPR antagonized the effect of NAD ϩ on both active and latent TGF-␤1. We also observed a significant expression of active (but not latent) TGF-␤1 induced by this dose of 8-Br-cADPR, which we hypothesize is due to an off-target effect of this compound.
We next sought to further define the TGF-␤1 response of RAW 264.7 cells to cADPR. As seen in Fig. 3B, authentic cADPR (10 M) led to increased expression of both active and latent TGF-␤1. Similarly, the stable cADPR analog 3-deaza-cADPR (17) (which is much more stable than authentic cADPR) also led to increased expression of both active (Fig. 3C) and latent TGF-␤1 (Fig. 3D) at concentrations of 1-50 nM.
The Effects of NAD ϩ and cADPR on TGF-␤1 Require Ca 2ϩ Mobilization, Are Mimicked by Ca 2ϩ Agonists, and Are Inhibited by Ca 2ϩ Antagonist-cADPR has been reported to lead to the release of Ca 2ϩ from ryanodine-sensitive intracellular stores (3). Accordingly, we further probed this pathway in the response to NAD ϩ . We found that blocking Ca 2ϩ with the Ca 2ϩ chelator BAPTA (10 M) inhibited the activation of TGF-␤1 by 100 M NAD ϩ (Fig. 4A) at 1 h and to a lesser degree at 3 h (Fig. 4B). Furthermore, treatment with the cADPR analog 3-cADPR (1 nM, Fig. 5A) and the Ca 2ϩ agonists bradykinin (10 M, Fig. 5B) or ionomycin (1 M, Fig. 5C) for 1 h induced increased active and latent TGF-␤1, an effect inhibited by pretreatment for 30 min with the Ca 2ϩ chelator BAPTA. The involvement of Ca 2ϩ mobilization in the activation of TGF-␤1 was further investigated using the L-type calcium channel blocker verapamil. We found that pretreatment of macrophages with 100 M verapamil for 30 min resulted in a significant inhibition of NAD ϩ -induced activation of TGF-␤1 (Fig.  5D), although this drug had no effect on the immunostaining for latent TGF-␤1.
The Effect of NAD ϩ on Active and Latent TGF-␤1 Follows a Complex Dose and Time Course-We next carried out a detailed dose-and time-course study of the effects of NAD ϩ on the immunocytochemically detectable expression of active (Fig.  6A) and latent TGF-␤1 (Fig. 6D). This study suggested that the effect of NAD ϩ on both active and latent TGF-␤1 was complex and possibly biphasic. NAD ϩ led to the activation of latent TGF-␤1 at early time points (1-2 h), which declined by 6 h and then appeared to rise again toward 24 h. The effect on latent (total) TGF-␤1 was similar but shifted in time, with the peak effect of NAD ϩ on latent TGF-␤1 occurring at 6 h and then declining by 12 h. This study also suggested that the effect of NAD ϩ on active TGF-␤1 was dose-dependent from 10 to 1000 M at late time points (8 -24 h) post-stimulation but peaked at 100 M and declined at 1000 M at early time points (1-6 h). The effects on NAD ϩ on latent TGF-␤1 were similar but again appeared to be shifted in time, with dose-dependent increases of latent TGF-␤1 apparent at 1, 12, and 24 h but with a peak at 100 M at 2-8 h.
Statistical and Mathematical Modeling of the Complex Effects of NAD ϩ /cADPR/Ca 2ϩ on TGF-␤1-To gain insight into the complex dose-and time-courses described above, the responses were modeled statistically (see Figs. 6, B-E) as a bivariate gaussian vector (see "Experimental Procedures"). The expected value of active TGF-␤1 was modeled as Dependence of latent TGF-␤1 on NAD ϩ was deemed to be simpler in the sense that interactions between concentration and time were not required in the model. Indeed, in this case we obtained a purely linear dependence. The following equation describes the relationship between latent TGF-␤1, NAD ϩ concentration, and time, The analogous p values, read from left to right, are lower than the respective values of 0.0001, 0.0001, 0.005, and 0.108. The model-based S 2 was 1.08 on 142 degrees of freedom with an R 2 of 0.238. The hypothesis of no effects of concentration and time upon latent TGF-␤ was rejected based on an F value of 5.554 on 8 and 142 degrees of freedom with a p value of 0.000004051. This statistical model suggested that the effects of NAD ϩ on TGF-␤1 were nonlinear, and this model was capable of predicting not only the levels of active (Fig. 6, B versus A) and latent TGF-␤1 (Fig. 6, E versus D) but also the biphasic dose effect of NAD ϩ .
To help define the mechanism by which NAD ϩ , cADPR, and Ca 2ϩ led to the induction and activation of TGF-␤1, we sought FIGURE 2. Increased expression of active and latent TGF-␤1 protein induced by NAD ؉ and its extracellular cADPR in primary mouse peritoneal macrophages. Isolated peritoneal macrophages from both wild-type (C3H/HeOuJ) and the TLR4-mutant (C3H/HeJ) mice were either incubated with medium alone (Control) or were treated with NAD ϩ (100 M) or the cADPR analog 3-cADPR (10 nM) for 1 h as indicated and subsequently immunostained for active (panels A and B) or latent (panel C) TGF-␤1 as described under "Experimental Procedures." Panel A, shown is a representative immunostaining for active TGF-␤1. Panel B, *, p Ͻ 0.05 versus Ctrl HeJ; **, p Ͻ 0.05 versus Ctrl HeOuJ. Panel C, *, p Ͻ 0.05 versus Ctrl HeJ; **, p Ͻ 0.05 versus Ctrl HeOuJ; #, p ϭ 0.005 versus NAD ϩ HeJ, analyzed by one-way analysis of variance followed by Tukey test).

JOURNAL OF BIOLOGICAL CHEMISTRY 31007
to create a mechanistic mathematical model of the presumptive interactions among these variables. Based on these data-driven modeling studies, we inferred that the effects of NAD ϩ on TGF-␤1 are nonlinear. Accordingly, we created a nonlinear ODE model of interactions we considered the most parsimonious and yet still capable of recapitulating the complex biological phenomena described above. Our initial ODE model included the interactions depicted in supplemental Fig. S1B (see "Experimental Procedures"), which shows NAD ϩ signaling through cADPR and Ca 2ϩ to activate TGF-␤1, which can then autoinduce its own mRNA production. However, this initial model was unable to account for the apparent second rise in TGF-␤1 observed at the later time points in Fig. 6, A and D. Importantly, we sought to include a mechanism by which BAPTA could suppress the NAD ϩ -induced increase in both active and latent TGF-␤1 at 1 h (Fig. 4A), but to a lesser extent at 3 h (Fig. 4B), arriving at the model depicted in supplemental Fig. S1C. Simulations from this subsequent model were able to reproduce the experimentally observed time-dependent response of TGF-␤1. This model could account for the timing-based difference in the effect of BAPTA on NAD ϩ -induced active and latent TGF-␤1 but could not recapitulate the attenuated response seen at high dose at the early time points (Fig. 6, A and D). To explain this phenomenon, we hypothesized a threshold-dependent inhibitory effect of Ca 2ϩ on TGF-␤1 mRNA that was included in our final model (supplemental Fig. S1D) described by the following equations: Parameter values for this model were estimated manually and are presented in Table 1. An initial simulation was carried out with no NAD ϩ input, and the resulting values were used as the initial conditions for each of the subsequent simulations with varying concentrations of NAD ϩ input. This procedure allows the system to equilibrate to a steady state before making any perturbations and corresponds to the culture of RAW 264.7 cells to allow for adherence (supplemental Fig. S2).
The ODE model simulations were generally in good concordance with experimental data for both active TGF-␤1 (Fig.  6, C versus A) and latent TGF-␤1 (Fig. 6, F versus D), although we observed a discrepancy at early time points (1 and 2 h) for latent TGF-␤1 (Fig. 6, F versus D).
We next attempted to validate the predictions of both the statistical and ODE models experimentally at both the protein and mRNA levels. We first subjected RAW 264.7 cells to 10, 100, or 1000 M NAD ϩ for 16 h, a time point not used for the calibration of either model. As seen in Figs. 7, A and B, the statistical model was able to accurately predict the levels of active and latent TGF-␤1. The ODE model was able to recapitulate the increase in active and latent TGF-␤1 induced by 10 and 100 M NAD ϩ , respectively, but was unable to capture the attenuated response to the 1000 M NAD ϩ dose observed at 16 h.
NAD ϩ Increases the Expression of TGF-␤1 mRNA in a Manner Predicted in Silico-We next examined the effects of NAD ϩ on TGF-␤1 mRNA. TGF-␤1 can auto-induce its own mRNA expression (6), and this auto-induction is presumably driven by exposure of cells to active TGF-␤1. Because extracellular NAD ϩ led to the activation of latent TGF-␤1 protein and the increased expression of latent (total) TGF-␤1 protein, we hypothesized that exposure to NAD ϩ would lead to increased TGF-␤1 mRNA as well. Moreover, because the maximal activation of TGF-␤1 by NAD ϩ occurred at 1 h, whereas maximal NAD ϩ -stimulated latent TGF-␤1 expression was seen at 6 h but persisted to 24 h, we examined the effects of NAD ϩ on TGF-␤1 mRNA at 21 h. As seen in Fig. 7C, increased expression of TGF-␤1 mRNA was observed with as little as 10 M NAD ϩ and persisted at levels of NAD ϩ up to 1000 M. As described in supplemental Fig. S1B (see above), a mathematical model that invoked only the auto-induction of TGF-␤1 was incapable of replicating the persistence of elevated TGF-␤1 mRNA and protein at later time points observed experimentally. In contrast, a mathematical model that included a hypothetical, Ca 2ϩ -independent pathway acting at the level of TGF-␤1 mRNA induction (supplemental Fig. S1D) accurately predicted the experimental data on NAD ϩ -mediated TGF-␤1 mRNA induction (Fig. 7C).

DISCUSSION
A large and growing class of DAMPs has emerged over the past decade as proinflammatory endogenous cellular products that stimulate inflammation (e.g. the production of TNF-␣ when released in settings of cell stress or cell death) (2). Canonical DAMPs include high mobility group box-1 (HMGB1), S100A, uric acid, and heat shock protein-70 (HSP-70) (2). However, relatively little attention has been paid to endogenous anti-inflammatory compounds derived from cells, mediators that include adenosine (18), ubiquitin (19), and hemopexin (20). Herein, we have identified and partially characterized a novel pathway of TGF-␤1 regulation by extracellular NAD ϩ in mouse macrophage-like cells using combined biochemical and in silico methods (Fig. 8), and our results suggest that NAD ϩ may be another member of this emerging group of anti-inflammatory DAMPs.
Indeed, emerging literature raises the possibility that NAD ϩ could be released by parenchymal cells in the setting of cellular stress or injury (3) and to exert various effects on inflammation. Granulocytes treated with NAD ϩ exhibited increased production of reactive oxygen species and other features consistent with activation (21). There is also extensive evidence that T-cell signaling is affected by NAD ϩ , with ultimate effects on proliferation and apoptosis (22)(23)(24)(25)(26)(27)(28). Furthermore, several studies have shown that NAD ϩ can protect against inflammatory stress both in vitro and in vivo (29,30), in part via reduced activity of the central inflammatory transcription factor NF-B. 5 We have also found that treatment of endotoxemic mice with NAD ϩ could protect from lethality in the case of mice injected with high dose LPSs and greatly reduced plasma TNF-␣ and NO b Ϫ / NO 3 Ϫ and elevated IL-10 in mice subjected to low-dose LPS. 5 Based on these data, we hypothesized that NAD ϩ , like other DAMPs, would have to act on nearby macrophages to induce classical anti-inflammatory cytokines. Given the long-recognized, potent and generally localized effects of TGF-␤1 in inflammatory settings (5), we hypothesized that one mechanism by which NAD ϩ could exert such potent effects would be via modulation of this cytokine.
The canonical signaling intermediary that controls the extracellular functions of NAD ϩ appears to be cADPR, which has been reported to lead to the release of Ca 2ϩ from ryanodine-sensitive intracellular stores (3). In most models of extracellular signal transduction, NAD ϩ is converted to cADPR and nicotinamide by extracellular ADP-ribosyl cyclase (3). The cADPR then enters cells through either a CD38-dependent (31) or -independent (32) mechanisms. The resulting increase in intracellular cADPR concentration leads to binding of cADPR to ryanodine-sensitive calcium channels on endoplasmic reticulum membranes. This binding increases the probability that the channel will be in an open conformation, allowing the release of endoplasmic reticulum calcium stores to the cytoplasm (3).
Although a role for extracellular cADPR in intracellular signaling events has been reported, there are other mechanisms through which extracellular NAD ϩ may act on cells. It has been proposed that extracellular NAD ϩ may modulate cellular responses by acting as a substrate for endogenous ADP-ribosyltransferases (33,34). Five mammalian ADP-ribosyltransferases (ART1-5) have been identified (35). Interestingly, CD38 has also been shown to possess ADP-ribosyl transferase activity (36). The ADP-ribosyltransferases are structurally and functionally related to cholera and pertussis toxins, which interfere with signal transduction in human host cells by ADP-ribosylating regulatory G-proteins (3). Several structurally unrelated inhibitors of ADP-ribosyltransferase were found to inhibit LPSinduced production of reactive oxygen intermediates and TNF-␣ by human mononuclear cells (37,38).
Our results demonstrate that mouse macrophage-like cells treated with NAD ϩ exhibit elevated expression of both active and latent TGF-␤1, with a biphasic time course that also depends on the specific concentration of NAD ϩ . In an attempt to discern the mechanisms responsible for this complexity, we utilized both data-driven and mechanistic computational modeling. Complex, multidimensional features of inflammation and associated processes (e.g. apoptosis) have been elucidated using such computational methods, for example, suggesting that the regulation of the transcription factor NF-B involves oscillations in the expression of its inhibitor, IB (39), and that signaling for apoptosis can be reduced to two principal components (40). An initial statistical analysis suggested that the interactions among NAD ϩ , latent TGF-␤1, and active TGF-␤1 were nonlinear. A series of nonlinear mechanistic mathematical models of increasing complexity were generated in an attempt to determine the mechanisms responsible for these nonlinear interactions. When combined with our biochemical data, our models suggested that NAD ϩ mediates its complex effects on

JOURNAL OF BIOLOGICAL CHEMISTRY 31011
TGF-␤1 via cADPR, secondary to stimulation of Ca 2ϩ fluxes. Our results do not rule out a role of ADP-ribosyltransferases in the effect of NAD ϩ on TGF-␤1 however. In fact, the inability of our mechanistic mathematical model, which does not incorporate the ADP-ribosyltransferases, to fully predict the multiphasic behavior of TGF-␤1 in response to NAD ϩ may suggest that we have not fully accounted for all the relevant biological mechanisms in modeling this process.
In our studies we found that blocking Ca 2ϩ with the Ca 2ϩ chelator BAPTA inhibited the activation of TGF-␤1 by NAD ϩ or cADPR and that treatment with the Ca 2ϩ agonists ionomycin and bradykinin both induced increased expression as well as activation of TGF-␤1. These studies suggest that modulation of Ca 2ϩ is a central mechanism responsible for the elevation of TGF-␤1. However, in multiple attempts we could not reproducibly demonstrate Ca 2ϩ fluxes in RAW 264.7 cells (data not shown). Interestingly, whereas many cell lines exhibit Ca 2ϩ fluxes in response to extracellular NAD ϩ (41,42), there are no published reports of ryanodine-sensitive Ca 2ϩ flux measurements in RAW 264.7 cells. It is, therefore, unclear if the effects of NAD ϩ and cADPR require the activation of ryanodine-sensitive Ca 2ϩ channels. Moreover, our data suggest that the effects of NAD ϩ on TGF-␤1 requires L-type Ca 2ϩ channels, although we have not defined the specific channels involved.
There are several limitations to our study. The first is that although the main finding of NAD ϩ /cADPR-induced induction of TGF-␤1 was shown both in RAW264.7 cells and in primary macrophages, the detailed time-and dose-curve experiments were carried out only in the former. We note that the extensive time-and dose-curve studies performed over many repeats would have been extremely cumbersome in primary cells. Moreover, we expect to have observed much greater variability in the biological responses studied. Another limitation is the extensive use of semiquantitative immunocytochemistry In both panels gray bars show predictions using a statistical model, and black bars show predictions using an ODE model as described under "Experimental Procedures." Panel C, total RNA was isolated from untreated (control) or RAW 264.7 cells treated with 10, 100, or 1000 M NAD ϩ for 21 h followed by Northern blotting and analysis for TGF-␤1 mRNA as described under "Experimental Procedures." The relative amount of mRNA is presented as the ratio of mRNA to 18S RNA (a representative blot from three independent experiments is shown in the inset). for these studies. Indeed, the expression of latent and active TGF-␤1 can be assessed by various methods, including ELISA, cell-based bioactivity assays, Western blotting, and immunocytochemistry to detect cell-associated TGF-␤1 (43). In our prior work establishing the utility of this assay (15), however, we showed the concordance between this method and other methods of determining TGF-␤1. Furthermore, prior studies had shown that TGF-␤1 is predominantly cell-associated (13,14,44,45).
In conclusion, our results thus point to NAD ϩ as one of a growing group of agents (proteases, plasmin, chaotropic agents, acid pH, radiation, and oxygen-and nitrogen-free radicals) (7) capable of activating TGF-␤1 and have at least partially elucidated the novel biochemical pathway by which these effects on TGF-␤1 occur. Our studies may further suggest that one mechanism by which anti-inflammatory DAMPs exert their actions is via activation of TGF-␤1. Moreover, our studies highlight the utility of traditional biochemical/pharmacological studies coupled with computational modeling in defining novel biological mechanisms.