ANGPTL8 negatively regulates NF-κB activation by facilitating selective autophagic degradation of IKKγ

Excessive nuclear factor-κB (NF-κB) activation mediated by tumor necrosis factor α (TNFα) plays a critical role in inflammation. Here we demonstrate that angiopoietin-like 8 (ANGPTL8) functions as a negative feedback regulator in TNFα-triggered NF-κB activation intracellularly. Inflammatory stimuli induce ANGPTL8 expression, and knockdown or knockout of ANGPTL8 potentiates TNFα-induced NF-κB activation in vitro. Mechanistically, upon TNFα stimulation, ANGPTL8 facilitates the interaction of IKKγ with p62 via forming a complex, thus promoting the selective autophagic degradation of IKKγ. Furthermore, the N-terminal domain mediated self-oligomerization of ANGPTL8 is essential for IKKγ degradation and NF-κB activation. In vivo, circulating ANGPTL8 level is high in patients diagnosed with infectious diseases, and the ANGPTL8/p62-IKKγ axis is responsive to inflammatory stimuli in the liver of LPS-injected mice. Altogether, our study suggests the ANGPTL8/p62-IKKγ axis as a negative feedback loop that regulates NF-κB activation, and extends the role of selective autophagy in fine-tuned inflammatory responses.

N F-κB plays a pivotal role in a variety of physiological and pathological processes, including inflammation, immunity and metabolism 1 . In non-stimulated cells, NF-κB is kept inactive in cytosol by binding to inhibitor of κBα (IκBα). Many agents, including pro-inflammatory cytokines, cause phosphorylation and degradation of IκBα, which results in releasing of NF-κB for translocation to the nucleus and initiating the expression of downstream genes 2 . Among these agentsmediated signaling, TNFα induction is a classical model to study the regulatory mechanisms of NF-κB activation. TNFα binds to its receptor TNF-RI to recruit the TNFR-associated death domain (TRADD), which recruits TNFR-associated factor 2 (TRAF2), TRAF5, and receptor interacting protein 1 (RIP1) to the receptor complex, then RIP1 further recruits and activates the transforming growth factor β activated kinase-1 (TAK1) complex and the IκB kinase (IKK) complex. The IKK complex consists of the catalytic subunits IKKα and IKKβ, and a regulatory subunit IKKγ. IKKγ, also known as NF-κB essential modulator (NEMO), is critical for the activation of the IKK complex; moreover, IKKγ also works as a scaffold which specifically channels the kinase activity of IKKβ to IκBα [3][4][5] . Finally, IκBα is phosphorylated and degraded, thereby releasing NF-κB to the nucleus.
Inappropriate NF-κB activation may cause immunodeficiency, chronic inflammation, autoimmunity and malignancy [6][7][8] . Therefore, IKK activation and IκBα phosphorylation, two key steps in NF-κB activation, should be tightly regulated, which highlights the physiological significance of IKKγ. Hypomorphic IKKγ mutations (IKKG gene located on the X chromosome) are lethal for male and lead to immune and developmental defects in female 9,10 ; while inactivation of the negative regulators of IKKγ, such as deubiquitinase A20 and CYLD lysine 63 deubiquitinase (CYLD), leads to serious disorders 11,12 ; similarly, males with a "gain of function" IKKγ mutant (ΔCT-NEMO, a C-terminal domain truncated mutant), which fails to recruit A20, develop autoinflammatory diseases 13 .
Proteolysis of signaling molecules is an important way to shut down signaling transduction. The ubiquitin-proteasome system (UPS) and autophagy are two major protein degradation pathways 14 . Previously, it was thought that UPS is highly selective while autophagy is a non-selective bulk process 14 . Recent studies suggest that autophagy can also target specific protein aggregates or other substrates for degradation, referred to as selective autophagy 15,16 . In general, proteins enter the selective autophagy are first K63-polyubiquitinated, then bound by one or more of autophagy receptors such as p62, neighbor of BRCA1 gene 1 (NBR1), nuclear dot protein 52 kDa (NDP52), TOLL interacting protein (Tollip), and optineurin (OPTN), followed by engulfment in autophagosomes 14,16 . Increasing evidences suggest that autophagy is important for the inflammation and immune responses 17,18 , however the role of selective autophagy in these critical physiological processes is poorly understood.
ANGPTL8 (also called Lipasin, RIFL, TD26 or C19orf80) is known as a key regulator of plasma lipid metabolism, which functions mainly by inhibiting lipoprotein lipase 19,20 . Here, we demonstrate intracellular ANGPTL8 as a novel negative feedback regulator of TNFα-mediated NF-κB activation, which may work as a critical step to avoid excessive inflammatory responses by facilitating p62-mediated autophagic IKKγ degradation.

Results
Pro-inflammatory cytokines up-regulate ANGPTL8. ANGPTL8 regulates lipid metabolism, and the level of circulating ANGPTL8 is increased in type 2 diabetes (T2D) 21,22 . Since lipid toxicity and T2D are tightly correlated with inflammation, we investigated the level of ANGPTL8 upon stimulation of pro-inflammatory cytokines such as TNFα and IL-1β. In HepG2 cells, the transcription and expression of ANGPTL8 were both significantly increased after being treated with TNFα ( Fig. 1a, b), with a TNFα dose-dependent elevation in ANGPTL8 level (Fig. 1c). Similar results were observed in two additional cell lines, HEK293T (a human embryotic kidney cell line) and A549 (a human lung cancer cell line) ( Supplementary Fig. 1a), although their ANGPTL8 level was markedly lower than that of HepG2 cells ( Supplementary Fig. 1b). Consistently, IL-1β treatment induced the transcription and expression of ANGPTL8 in HepG2 cells ( Supplementary Fig. 1c, d). Collectively, these results indicate that the ANGPTL8 expression could be triggered by different inflammatory stimuli and in various cells.
Knockdown or knockout of ANGPTL8 potentiates NF-κB activation. We next determined whether ANGPTL8 regulates NF-κB activation. Three ANGPTL8-RNAi plasmids were generated, which efficiently inhibited the transcription and expression of ANGPTL8 in HepG2 cells (Fig. 2a). In luciferase reporter assays, knockdown of ANGPTL8 enhanced TNFα-or IL-1βinduced NF-κB activation ( Fig. 2b and Supplementary Fig. 2a), and the level of NF-κB activation was correlated with the knockdown efficiency. As a negative control experiment, ANGPTL8-RNAi did not affect the IFNγ-induced IRF1 activation, which is another pathway involved in immunity and separated from TNFα-mediated signaling (Fig. 2c). These data indicate that ANGPTL8 specifically potentiates the TNFα-triggered and IL-1β-triggered NF-κB activation.
ANGPTL8 regulates NF-κB activation at the IKK complex level. We next investigated the molecular mechanisms by which ANGPTL8 regulates the TNFα or IL-1β signaling mediated NF-κB activation. TNFα/IL-1β-mediated NF-κB activation includes three major steps: adaptors such as TRAF2/6-mediated or RIP1mediated recruitment of IKK complex to TNFR, activation of IKKβ, and activation of NF-κB. We cotransfected HEK293T cells with a vector expressing TRAF2, TRAF6, RIP1, IKKβ, or p65, together with an overexpression or knockdown vector of ANGPTL8 plus an NF-κB luciferase reporter vector. The overexpression/knockdown of ANGPTL8 induced changes of NF-κB activation upon overexpression of TRAF2/6 or RIP1, but not IKKβ and p65 (Fig. 4a, b). These data implicate that ANGPTL8 may involve in the recruitment or activation of IKK complex. Consistently, by co-IP experiments, we found that ANGPTL8 interacted with RIP1, IKKβ and IKKγ in the cells cooverexpressed ANGPTL8 and the regulator molecules involved in NF-κB activation cascade (Fig. 4c). Moreover, in untransfected cells, ANGPTL8 rapidly interacted with IKKβ/γ and RIP1 upon TNFα treatment (Fig. 4d), which may in part due to the TNFαtriggered ANGPTL8 expression. Collectively, these results indicate that ANGPTL8 may target IKK complex through participating in the recruitment or activation of IKKs.
Since ANGPTL8 is mostly known as a secreted protein, after we found its role in the regulation of intracellular signaling, we studied the localization of ANGPTL8. In ANGPTL8 overexpression HepG2 cells, a large amount of ANGPTL8 was detected in cell lysate, while a proportion of ANGPTL8 was secreted ( Supplementary Fig. 3a), this result was consistent with a recent study carried out in HEK293T cells 23 . Immunofluorescence experiments confirmed that endogenous ANGPTL8 had intracellular localization ( Supplementary Fig. 3b).
Confocal microscopy experiments further suggested that in unstimulated cells, only a small fraction of IKKγ was co-localized with GFP-LC3 dots, a marker for autophagosome; whereas in cells overexpressing ANGPTL8, the overlap and correlation efficiency between IKKγ and GFP-LC3 dots were markedly increased (Fig. 6c, d). It has been observed that TNFα stimulation induces relocalization of IKKγ into punctate structures that are enriched in activated IKK kinases and IKKγ, which may be essential for NF-κB activation 26,27 . Consistently, we found that TNFα stimulation induces the recruitment of IKKγ into punctate structures, however, this TNFα-induced IKKγ punctual relocalization was dramatically decreased in ANGPTL8 −/− cells, and the co-localization of IKKγ with GFP-LC3 was also significantly reduced ( Fig. 6e, f).
ANGPTL8 and p62 co-mediate the autophagic degradation of IKKγ. Motivated by the observations that ANGPTL8 selectively mediated the autophagic degradation of IKKγ but not IKKα/β, we next studied whether ANGPTL8 mediates IKKγ degradation via selective autophagy. K63-linked ubiquitin chains have been reported to promote selective autophagy-dependent degradation for specific target proteins 14 , we found overexpression of wildtype or K63-linked ubiquitin (the ubiquitin in which all of the lysine mutated to arginine except the lysine 63) enhanced the ANGPTL8-mediated IKKγ degradation ( Fig. 7a and Supplementary Fig. 6); whereas CYLD, which removes the K63-linked ubiquitin chains of IKKγ 28 , attenuated ANGPTL8-mediated IKKγ degradation in a dose-dependent manner (Fig. 7b).
Typically, cargoes are selectively transferred by one or more of the known autophagy receptors (p62, Tollip, NDP52, NBR1 and OPTN etc.) for degradation 15,29 . We found that it was p62 and Tollip, but not NDP52, NBR1 or OPTN, that mediated the degradation of IKKγ, which was further markedly enhanced when co-expressed with ANGPTL8 ( Fig. 7c and Supplementary Fig. 7a). In wild-type cells, although p62 and Tollip both significantly and dose-dependently induced IKKγ degradation, ANGPTL8 deficiency only reversed p62-mediated, but not Tollip-mediated degradation of IKKγ ( Fig. 7d and Supplementary Fig. 7b). Moreover, ANGPTL8 could interact with p62, but not Tollip; and the interaction between ANGPTL8 and p62 was greatly enhanced after TNFα treatment (Fig. 7e). Also, knockdown of p62 dramatically attenuated ANGPTL8-mediated IKKγ degradation (Fig. 7f). The LC3 interacting region (LIR) of p62 is responsible for recruitment of LC3 to the autophagosome 30 . Here, we observed that p62ΔLIR (a LIR domain deleted p62 mutant) showed dramatically decreased ability in mediating IKKγ degradation ( Supplementary Fig. 7c). Notably, similar to p62 and IKKγ, we found that ANGPTL8 also underwent autophagic degradation ( Supplementary Fig. 7d). Altogether, these data indicate that ANGPTL8 and p62 work collaboratively to mediate the autophagic degradation of IKKγ.
Reducing ANGPTL8 attenuates the IKKγ-p62 interaction. In selective autophagy, substrate recognition by autophagy receptor is essential for cargo selection. By performing co-IP experiments in physiological conditions, we found that IKKγ, ANGPTL8 and p62 interacted with each other (Fig. 8a, b). We noted that only a small fraction of p62 was associated with IKKγ in resting cells, and this association was greatly increased after TNFα treatment, which is consistent with the observation that ANGPTL8 was simultaneously recruited to IKKγ (Fig. 8a, b). Consistently, TNFα-induced interaction between IKKγ and p62 was impaired by knockdown or knockout of ANGPTL8 (Fig. 8c, d). These data indicate that upon TNFα stimulation, IKKγ, p62, and ANGPTL8 form a complex in which ANGPTL8 plays an important role in mediating the interaction.
Oligomerization of ANGPTL8 is essential for IKKγ degradation. In selective autophagy, protein aggregation is a key event in mediating cargo selection and separation 31 . Interestingly, by co-IP analysis, we found ANGPTL8 can self-oligomerize (Fig. 9a).
ANGPTL8 is predicted to have an N-terminal signal peptide (residues 1-25) and two coiled-coil (CC) domains (residues 77-134 and 156-193) which are presumably associated with protein-protein interaction (Fig. 9b); however, the function of different domains/regions in ANGPTL8 is not clear 32,33 . By using domain mapping, we identified the region between residues 26-70 as an essential domain for the self-oligomerization of ANGPTL8 (Fig. 9a). Notably, the truncation mutants (71-C and Δ26-70) that lost the self-association capacity could not mediate the IKKγ degradation, while the CC domain that unrelated to the self-oligomerization of ANGPTL8 was dispensable in this process ( Fig. 9c and Supplementary Fig. 8). Consistently, ANGPTL8-Δ26-70 showed a significantly diminished interaction with IKKγ and lost the ability of inhibiting TNFα-induced NF-κB activation (Fig. 9d, e). Thus, oligomerization of ANGPTL8 mediated by its N-terminal 26-70 region is essential for the ANGPTL8-mediated inhibition of NF-κB activation by mediating the interaction and degradation of IKKγ. Moreover, we found that the coiled-coil domains were responsible for the ANGPTL8-p62 interaction ( Supplementary Fig. 9). Interestingly, compared with the fulllength ANGPTL8, ANGPTL8-Δ26-70 showed a stronger interaction with p62, suggesting the self-oligomerization is not required for ANGPTL8-p62 interaction (Fig. 9d). Gel filtration analysis was performed to explore the oligomerization status of ANGPTL8. The full-length ANGPTL8 was most abundant in a higher molecular mass (fractions 5-12) than ANGPTL8-Δ26-70 (fractions 10-15) (Fig. 9f). Transfection of ANGPTL8-Δ26-70 led to significantly decreased molecular weight of p62 and IKKγ-containing fractions in the same experiments, indicating the importance of ANGPTL8 oligomerization in the formation of IKKγ and p62-containing protein complex. We also expressed and purified the recombinant ANGPTL8 to assess its oligomerization tendency in vitro    Fig. 10a). The circular dichroism spectra revealed that in solution, ANGPTL8 adapted a mixed α-helix/β-structures/ random coil structure ( Supplementary Fig. 10b, c), which agrees with the structural prediction ( Supplementary Fig. 10d). ANGPTL8 showed tendency to aggregate into oligomers and large aggregates (fibrils) as measured by the dynamic light scattering (DLS) assays or by probing with the anti-oligomer and anti-fibril antibodies ( Supplementary Fig. 10e, f). Under transmission electronic microscopy, oligomers and fibrils were also identified ( Supplementary Fig. 10g).

Potential involvement of ANGPTL8 in acute inflammation.
Our in vitro studies suggest that ANGPTL8 degrades IKKγ, whose tight regulation is essential for the balance of inflammation, we next investigated the physiological relevance of ANGPTL8 in vivo. As demonstrated by qPCR analysis (Supplementary Fig. 11A), Angptl8 is relatively high in the liver and brown adipose tissue (BAT), and low in the spleen, lung and kidney of mouse. Next, we examined the tissue level of Angptl8 in mice challenged with lipopolysaccharide (LPS), a constituent of the Gram-negative bacteria outer membranes and an important microbial trigger that stimulates innate immunity. Interestingly, in tissues with high level of Angptl8, fast upregulation and downregulation of Tnfa transcription during the acute phase (0-1 h) and the resolution phase (1-6 h) was respectively observed upon LPS challenge; in contrast, they were relatively slow in tissues with low Angptl8 expression ( Fig. 10a and Supplementary Fig. 11b). This observation implicates that for tissues that are sensitive to the inflammatory stress, larger amount of "brake" molecules, such as ANGPTL8, may be demanded. Furthermore, upon LPS stress, the expression of Angptl8 was increased, while Ikkγ expression was decreased (Fig. 10b, c); however, the interaction among Ikkγ, p62 and Angptl8 was enhanced (Fig. 10d). These results are consistent with our in vitro experiments. Third, we measured the circulating ANGPTL8 level in the blood of two groups of patients with systemic inflammatory response syndrome. One group includes patients with positive detection of procalcitonin (PCT) which is a biomarker for early IKKγ-RNAi mock F l a g -I K K γ F l a g -I K K β F l a g -I K K α F l a g -R I P 1 F l a g -T R A F 6 F l a g -T R A F 6 F l a g -R I P 1 F l a g -I K K α F l a g -I K K β F l a g -I K K γ diagnosis of sepsis 27 , the other group includes patients with positive detection of endotoxin which is an important microbiological assessment for Gram-negative bacteria-mediated inflammation 34 . Compared to the healthy subjects, the circulating ANGPTL8 level was dramatically increased in both groups of patients ( Fig. 10e and Supplementary Table 1). Collectively, these results indicated that ANGPTL8 can be induced by inflammatory stimuli in mouse and human, and may thus play roles in shutting down acute inflammatory response.

Discussion
ANGPTL8 has been known as a potent regulator of lipid metabolism 35 . Circulating ANGPTL8 is increased in patients with T2D or non-alcoholic fatty liver diseases, which makes ANGPTL8 an attractive therapeutic target for metabolic syndromes 22,36,37 .
On the other hand, inflammation, especially the proinflammatory cytokines-mediated chronic inflammation, has been demonstrated to contribute to the development of metabolic diseases such as T2D. However, the role of ANGPTL8 in inflammation is unknown. As a central event of inflammation and immunity, TNFα-induced NF-κB activation must be tightly controlled to avoid inflammatory diseases, autoimmunity and cancers [6][7][8] . Here we present the first evidence that multiple inflammatory stimuli, including TNFα, induce the transcription and expression of ANGPTL8 in vitro and in vivo; the latter then forms a protein complex with p62 and IKKγ, in which ANGPTL8 works as a co-receptor of p62 and facilitates the autophagic IKKγ degradation, thereby inhibiting the TNFα-induced NF-κB activation (working model shown in Fig. 10f). The ANGPTL8/p62-IKKγ axis thus serves as a negative feedback loop to restrict the TNFα-trigged NF-κB activation and inflammation.
Since ANGPTL8 is mostly known as a secreted protein, it is interesting that our results suggest it also plays intracellular roles and has intracellular location. Actually, other studies have demonstrated that some secreted proteins have intracellular functions. For example, ISG15, an interferon-induced modifier, is found both intracellularly and extracellularly; the secreted ISG15 has cytokine like activities 38 ; whereas the intracellular ISG15 can conjugate various proteins via ISGylation, and prevents the IFN-  6). e, f The representative images (e) and quantitative results (f) of effects of ANGPTL8 deficiency on TNFα-induced co-localization of IKKγ with the autophagosomes. The scale bar represents 10 μm (n = 6). Data are shown as the mean ± SEM, unpaired two-tailed student's test was used for statistics (d, f). *p < 0.05, **p < 0.001. Data are representative of three independent experiments NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02355-w ARTICLE α/β-dependent autoinflammation 39 . PCSK9 is another secreted protein, and binds hepatotic LDLRs both extracellularly and intracellularly which leads to LDLRs degradation 40,41 . Similarly, the intracellular function of ANGPTL8 has been reported. It can enhance the cleavage of ANGPTL3, a molecule involved in the triglyceride metabolism 20 ; and it is also involved in the lipolysis of adipocytes 42 . ANGPTL8 has been reported to be induced by thyroid hormone, and modulates autophagy mainly via enhancing autolysosome maturation 43 . In this study, we found that ANGPTL8 specifically mediates the autophagic IKKγ degradation, which can be promoted by overexpression of K63-linked ubiquitin, and chemical or genetic blockage of autophagy could inhibit this IKKγ degradation (Figs. 5-7). Furthermore, p62, a classical autophagy receptor, works cooperatively with ANGPTL8 to mediate the proteolysis of IKKγ (Figs. 7 and 8 and Supplementary  Fig. 7). Although it has been reported that IKKγ is mainly degraded by lysosomal pathway 44 , this is the first report that IKKγ can be degraded by ANGPTL8/p62-mediated selective autophagy.
The specificity of selective autophagy for the degraded cargoes is mainly attributed to autophagy receptors. While there are numerous molecules or cell organelles need to be degraded, only a handful of known autophagy receptors are responsible for this process 15,31,45 . There is a general model behind the selectivity: the substrate needs to be recognized by other proteins (e.g., molecular chaperones) before interacting with autophagy receptors, which can be termed as "co-receptor" 31 . In this study, several lines of evidence indicate ANGPTL8 as a co-receptor of p62 to mediate the IKKγ selection. First, p62 and many other chaperones are stress-responsive, possibly to ensure different substrates are properly degraded under certain conditions 46,47 ; similarly, ANGPTL8 is a stress-responsive molecule with enhanced expression under TNFα, IL-1β, LPS, or infection (Figs. 1 and 10 and Supplementary Fig. 1). Second, the interactions between p62 and ANGPTL8 are mutually required for IKKγ degradation (Fig. 7). Third, TNFα treatment promotes the formation of the IKKγ-p62-ANGPTL8 complex, which is impaired in the ANGPTL8 knockdown or knockout cells (Fig. 8). Consistently, enhanced binding affinity between Ikkγ, Angptl8, and p62 in the liver of LPS-treated mice was also observed (Fig. 10d). Our results thus suggested a novel role of ANGPTL8 as a co-receptor of p62 in selective autophagy.
After cargo selection, an important role for autophagy receptors such as p62 is to sequester cargoes into larger aggregates The effect of deubiquitinase CYLD on the ANGPTL8-mediated IKKγ degradation. c The effects of different autophagic adaptors on the ANGPTL8-mediated IKKγ degradation. d Overexpression of p62 but not Tollip-mediated IKKγ degradation is attenuated in ANGPTL8 −/− cells. e ANGPTL8 interacts with p62 but not Tollip in the mammalian overexpression system. f ANGPTL8-mediated IKKγ degradation is decreased in p62-RNAi stable cell line. Data are representative of three independent experiments before they being degraded 31,48,49 . During this process, p62 forms oligomers, and such suprastructure enables its interaction with LC3B and ubiquitinated cargoes 49,50 . It is an interesting and open question that whether other p62-related proteins work in similar ways 50 . Here, we demonstrate that ANGPTL8 self-oligomerizes through its N-terminal region and such oligomerization is essential for the interaction/degradation of IKKγ, thereby affecting the ANGPTL8-mediated inhibition of TNFα-induced NF-κB activation (Fig. 9), indicating the oligomerization is important to the interaction between the co-receptors and cargoes. However, ANGPTL8 seems to directly interact with p62, since ANGPTL8-Δ26-70 showed stronger interaction with p62 compared with the full-length ANGPTL8 (Fig. 9).
As major protein degradation pathways, UPS and autophagy have been proven to play significant roles in inflammation 17,18 . While there have been dozens of regulators identified to influence NF-κB activation by UPS, reports on how autophagy degrades specific signaling molecules in inflammation are rare [50][51][52][53] , which may in part due to the concept that UPS is highly selective but autophagy is a bulk process. However, recent studies suggested both process can be specific, UPS is involved in the rapid degradation of single proteins, while autophagy can selectively remove protein aggregates and damaged/excess organelles that are too big in size for proteasomes 14 . Interestingly, signaling components, especially receptors and scaffold proteins, tend to form oligomers to mediate the signaling 54 , implicating that these aggregation-prone proteins may be the appropriate substrates for autophagy. IKKγ is a well-known scaffold protein and TNFα stimulation can induce the IKKγ translocation to supramolecular structures which is important to the NF-κB activation 4,5,27,55 .
Here, we demonstrated that IKKγ could undergo the ANGPTL8/ p62-facilitated selective autophagy, we also observed that the percentage of punctual IKKγ can be enhanced by ANGPTL8, these findings suggested a possibility that aggregated IKKγ is not only important for signaling transduction, but also a precondition for its degradation. Moreover, our study implicates that additional scaffold proteins in this or other pathways may also be degraded with similar mechanism.
Multiple studies have confirmed the induction of circulating ANGPTL8 in different metabolic syndromes 36,37 . Here we found that the level of circulating ANGPTL8 was dramatically increased in severe infection (Fig. 10e). To our knowledge, it is the first report on the relationship between ANGPTL8 and acute inflammation in clinical patients. We also observed a correlation of Angptl8 level with LPS-induced acute inflammatory response in different tissues of mouse ( Fig. 10a and Supplementary  Fig. 11). Future experiments using Angptl8 knockout or transgenic mice will be helpful to reveal the mechanisms underlying its physiological and pathological roles in inflammatory diseases.
Collectively, our results uncover an important fine-regulation mechanism for NF-κB activation. Notably, the observations that ANGPTL8 also can be induced by additional factors such as IL-1β, LPS, insulin resistance and feeding from this study and from literature 56 , implicate a more broaden role of ANGPTL8 in the autophagic degradation of other inflammation or metabolism associated proteins, a question awaits further exploration.

Methods
Reagents, antibodies and cell lines. Immunoblots with ANGPTL8/Angptl8 were done by mouse anti-ANGPTL8 monoclonal antibody (1:300), a kind gift from Dr. Y. Wang (Wuhan University) 19,20 . Immunofluorescence with ANGPTL8 was done by an anti-ANGPTL8 monoclonal antibody (Sigma, SAB3501080, 1:100). Co-IP  Fig. 8 ANGPTL8 is important for the IKKγ-p62 interaction. a, b ANGPTL8 forms a complex with IKKγ and p62 upon TNFα stimulation. HepG2 cells were stimulated with TNFα or left unstimulated for indicated time (2.5 × 10 7 cells per group). And then cells were lysed and cell lysate was subjected to co-IP analysis by anti-IKKγ. (a) or anti-ANGPTL8 (b) followed immunoblots. c The effects of ANGPTL8-RNAi on the IKKγ-related protein complex formation. The experiment was performed as in a except the ANGTPL8-RNAi or control cells (2.5 × 10 7 ) were used. d The effects of ANGPTL8 deficiency on the IKKγrelated protein complex formation. ANGPTL8-deficient or control cells (2.5 × 10 7 ) were left untreated or treated with TNFα for 1 h before subjected to Co-IP analysis and immunoblots. Data are representative of three independent experiments experiments with ANGPTL8/Angptl8 were done by mouse anti-ANGPTL8 polyclonal antibody which was raised against recombinant human full-length ANGPTL8 with standard protocols. A list of commercial reagents, and other antibodies and dilutions used in the present study was provided in Supplementary  Table 2. HepG2 (CL-0103000), HEK293T (CL-0005), and A549 (CL-0016) cells, which were analyzed with authenticated STR locus and tested for mycoplasma contamination, were purchased from Procell Biotech. (Wuhan, China).
Transfection and luciferase reporter gene assays. Cells (5 × 10 4 ) were seeded on 48-well plates and transfected on the following day, empty control plasmid was added to ensure that each transfection receives the same amount of total DNA. To normalize transfection efficiency, 0.02 μg of pRL-TK Renilla luciferase reporter plasmid was added to each transfection. Luciferase assays were performed using a dual-specific luciferase assay kit (Promega), the firefly luciferase activities (NF-κB or IRF1 firefly luciferase reporter) were normalized based on Renilla luciferase activities.
RNAi experiments. Double-strand oligonucleotides corresponding to the target sequences were cloned into the pSuper plasmids (Oligoengine). The target sequences for human ANGPTL8, p62, IKKγ, ATG5, and ATG7 cDNA are listed in Supplementary Table 3.
Retrovirus-mediated stable RNAi cell lines. The packaging cell line HEK293T was transfected with retroviral vectors by calcium phosphate precipitation. Twelve hours later, cells were washed by PBS, and antibiotics-free medium was added for another 24 h. The filtered supernatant was used to infect HepG2 or ANGPTL8 −/− cells in the presence of 6 μg/mL polybrene. The infection was repeated twice to enhance transduction efficiency.
Quantitative real-time PCR. Total RNA was isolated from cells using RNAiso Plus reagent (Takara) and subjected to qPCR analysis. The mRNA levels of specific genes were normalized to GAPDH. The gene-specific primer sequences for ANGPTL8, IKKG, Angptl8, and Tnfa are listed in Supplementary Table 3 . f Analysis of protein complex containing ANGPTL8, p62 and IKKγ by size-exclusion chromatography. HEK293T cells (2 × 10 6 ) were transfected with indicated plasmids for 24 h later before cells were lysed and subjected to size-exclusion chromatography. Data are shown as the mean ± SEM in e, unpaired two-tailed Student's test was used for statistics (e). ***p < 0.0001, NS > 0.05. Data are representative of three independent experiments. FL full length, TM truncation mutants primers for CXCL2, IL8, NFKBIA, TNFA, and GAPDH were as we previously described 58,59 .
CRISPR-Cas9-mediated knockout of ANGPTL8. The CRISPR-Cas9 based protocols for genome engineering were used as described 60 . PGL-U6-GRNA and PST1374-Cas9 plasmids were gifts of Dr. X. Zhang (Wuhan University). The ANGPTL8 gRNA target sequence and the identification primers for ANGPTL8 knockout are listed in Supplementary Table 3.
Confocal microscopy. The WT and ANGPTL8 −/− stable cell lines expressing GFP-LC3 were constructed by lentivirus mediated gene transfer. At 20 h after transfected with IKKγ-Cherry, cells were treated with or without TNFα for 2 h. After fixing with 4% (W/V) formaldehyde, the nuclei were stained by DAPI (1 μg/mL) and the cells were imaged with a Zeiss LSM 880 confocal microscope. The plugin JACoP of Image J was used to calculate the colocalization rate for the red pixels and green dots (green dots represent LC3-II, a marker for autophagosome 61 ) as described 62 . Colocalization of signals from IKKγ-Cherry and GFP-LC3 dots was evaluated using Manders' overlap coefficient and the Pearson's correlation coefficient.
Size-exclusion chromatography. Cells (2 × 10 6 ) were transfected with indicated plasmids for 24 h before being lysed in 500 μL of pre-lysis buffer. The cell lysate was then incubated on ice for 30 min followed by sonication and was spun down at 12000×g for 10 min. The supernatant was filtered with a 0.45 μm filter (Millipore) before being loaded onto a Superose 6 size-exclusion chromatography column (GE Healthcare, 1 × 30 cm), which was pre-equilibrated with Triton and EDTA free pre-lysis buffer. The samples were eluted at 4°C by lysis buffer at a flow rate of 500 μL/min and collected in fractions of 500 μL. The fractions were precipitated with 20% trichloroacetic acid and analyzed by immunoblots with antibodies against Flag, IKKγ, and p62.
Human studies. To compare circulating ANGPTL8 levels between patients with inflammation and healthy controls, patients with positive detection of procalcitonin (PCT > 0.5 μg/L, n = 18), positive detection of endotoxin (LPS > 0.1 EU/mL, n = 10) and healthy controls (n = 30, from physical examination center) were included in the study. The detail sample information was listed in Supplementary Far-UV circular dichroism and structural modeling. A JASCO-810 circular dichroism spectropolarimeter (Tokyo, Japan) was used to monitor secondary structures. Incubated samples were diluted to a final concentration of 10 μM and detected in a 1 mm path length at 25°C. Circular dichroism (CD) spectra were . f A working model of ANGTPL8 mediated regulation of TNFα-induced signaling. TNFα upregulates the expression of ANGPTL8, and then aggregated ANGPTL8 facilitates the ANGPTL8-p62-IKKγ complex formation, which promotes the autophagic IKKγ degradation and inhibits TNFα-induced NF-κB activation. Data are shown as the mean ± SEM in c, e, unpaired two-tailed Student's test was used for statistics. ***p < 0.0001, **p < 0.01, *p < 0.05 obtained from 260 to 200 nm at a 200 nm/min scanning speed and a 2 nm bandwidth. All samples were measured in triplicates and the averages were taken. The data were converted to mean residue ellipticity and the secondary structural contents were further calculated with the software CDPro 63 . Structural modeling of ANGPTL8 was conducted using the online server I-TASSER.
Dynamic light scattering analysis. The sizes of ANGPTL8 aggregates were measured by dynamic light scattering in a zeta pals potential analyzer (Brookhaven Instruments, USA). Samples were vortexed and detected at room temperature, the scattering angle was set at 90°. Each measurement was repeated three times and the average mean particle size was recorded. The data was analyzed by the multimodal size distribution (MSD) software 64 .
Dot blot assays. Sample aliquotes (2 μL) obtained at indicated time points were blotted onto a nitrocellulose membrane (Bio-Rad, USA). Dried membrane was blocked with 5% non-fat milk for 1 h at room temperature and then incubated with anti-oligomer antibody (A-11) or anti-fibril antibody (OC) at 4°C overnight. The membrane was incubated with anti-rabbit IgG for 2 h at room temperature later. An ECL chemiluminescence kit (Advansta, USA) was used for the development.
Transmission electronic microscopy. Incubated solution was applied onto a 300mesh formvar-carbon coated copper grid and sit for 5 min. Freshly prepared uranyl formate (1%) was dropwise added for staining. Dried samples were observed under a transmission microscope (Hitachi, Japan) operating at an accelerating voltage of 200 kV 65 .
Statistical analysis. Sample sizes, as described in figure legends, were selected based on effect size and availability as per usual standard. Randomization was done by selecting animals of similar age and weight. No blinding was involved in animal studies. Statistically significant differences between the mean values were determined by two-tailed Student's t-test (*p < 0.05, **p < 0.01, ***p < 0.001, NS > 0.05). Data are presented as the mean ± SEM.