AIM 2 inflammasomes regulate neuronal morphology and influence anxiety and memory in mice

Inflammasomes are the protein assemblies that consist of inflammasome sensors, adaptor apoptosis-associated speck-like proteins containing a CARD (ASC) and inflammasome caspase. Inflammasomes sense multiple danger signals via various inflammasome sensors and consequently use caspase to trigger proteolytic processing and secretion of IL-1β cytokines. Recent studies have suggested that neurons use their own innate immune system to detect danger signals and regulate neuronal morphology. Here, we investigate whether inflammasomes, the critical components of innate immunity, participate in regulation of neuronal morphology and function. Among various sensors, Absent in melanoma 2 (Aim2) expression in neurons is most prominent. Adding synthetic double-stranded DNA (dsDNA) to cultured neurons induces IL-1β secretion in an AIM2-dependent manner and consequently downregulates dendritic growth but enhances axon extension. The results of Aim2 knockout and knockdown show that AIM2 acts cell-autonomously to regulate neuronal morphology. Behavioral analyses further reveal that Aim2−/− mice exhibit lower locomotor activity, increased anxious behaviors and reduced auditory fear memory. In conclusion, our study suggests that AIM2 inflammasomes regulate neuronal morphology and influence mouse behaviors.

Innate immunity is a naive defense system that neutralizes and removes various foreign pathogens. It also acts as an alarm system to respond to intrinsic danger signals, such as stress and neurodegeneration 1,2 . In addition, immune responses are highly associated with neurodevelopmental disorders, such as autism spectrum disorders and schizophrenia [3][4][5][6] . In brains, microglial cells are recognized as the specialized immune cells that respond to infection and neurodegeneration [7][8][9] . However, neurons have also been shown to express critical molecules that are involved in innate immune responses, such as Toll-like receptors (TLRs) and the downstream adaptors myeloid differentiation primary response protein 88 (MYD88), the Toll/IL-1 receptor (TIR) domain-containing adaptor inducing IFN-β (TRIF) and the Sterile α and TIR motif-containing protein 1 (SARM1), as well as inflammasomes [10][11][12][13][14] . Neurons also produce inflammatory and antiviral cytokines, as well as corresponding receptors for these cytokines 13,15 . Overall, neurons have their own innate immunity, which responds to endogenous ligands 1,16 to regulate neurogenesis, neural differentiation and neurodegeneration 13,[16][17][18][19] .
Among various TLRs, TLR7 has been extensively studied in the nervous system. TLR7 recognizes single-stranded RNAs, including bacterial and viral RNAs, as well as endogenous mRNAs and miRNAs 1,16,20,21 . TLR7 activation delivers a signal to MYD88 and c-FOS, thereby inducing IL-6 expression to restrict axon and dendrite outgrowth of cultured neurons through both autocrine and paracrine signaling mechanisms 13,16 . Interestingly, TLR7 is activated in neuronal cultures in the absence of exogenous ligand. RNAs released from dead cells or miRNAs released through exosomes likely play roles in TLR7 activation of cultured neurons 13,16,19 . Echoing the role of TLR7 in neural differentiation, exploratory activity of Tlr7 knockout mice is altered at the juvenile stage 13 .
In addition to IL-6, neuronal TLR7 activation also increases the level of Il-1β mRNA expression, but the level of secreted IL-1β proteins in the supernatants of cultured neurons is not increased 13 . Cytoplasmic pro-IL-1β must be cleaved by an inflammasome-a composite assembly that consists of inflammatory caspases, adaptor apoptosis-associated speck-like proteins containing a CARD (ASC) and various sensors to detect inflammation-inducing stimuli 22 -to produce and secrete the mature form of IL-1β [23][24][25][26][27] . Different inflammatory sensors detect various secondary signals, rather than the signals that activate TLRs, to activate inflammasomes 27,28 .
Scientific RepoRts | 6:32405 | DOI: 10.1038/srep32405 It is likely that a proper secondary signal is necessary to activate neuronal inflammasomes in order to induce IL-1β processing in neurons. The inflammasome sensors belong to two families: the nucleotide-binding oligomerization domain-like receptor (NLR) family and the PYHIN (hematopoietic interferon-inducible nuclear antigens with 200-amino-acid repeat-domain-containing protein) family. Among these various sensors, NLRP3 is the most studied inflammasome sensor in brain [29][30][31][32][33] . NLRP1, NLRC4 and AIM2 are also involved in brain injury and the innate immune response in brain 14,31,34,35 . Other sensors, such as NLRP6 and NLRP12, have been less extensively studied 36,37 . AIM2 is a member of the PYHIN family, while NLRP1, NLRP3, NLRP6, NLRP12 and NLRC4 all belong to the NOD-like receptor family 22 . In this report, we explore whether neurons are able to sense stimulation and activate inflammasomes and investigate whether, similar to TLRs, inflammasome activation regulates neuronal morphogenesis and mouse behaviors.

Results
Aim2 is expressed in developing neurons. To investigate the potential role of inflammasomes in neurons, we first examined expression of various inflammasome-related genes. Quantitative-PCR with the Universal ProbeLibrary system was used to detect the mRNA levels of Nlrp1a, Nlrp1b, Nlrp1c, Nlrp3, Nlrp6, Nlrp12, Nlrc4, Aim2 and Asc. The results showed that, in cultured mouse cortical and hippocampal neurons, Aim2 was the most highly expressed inflammasome sensor compared to the other inflammasome sensors (Fig. 1a). This result echoes a previous study showing Aim2 expression in neurons 14 . To further confirm this result, we performed quantitative-PCR using the SYBR Green system with another two different sets of primers for Aim2. Nlrp3, the most well-studied inflammasome sensor, was also included to compare with Aim2. Aim2 expression levels were much higher (~10-fold) than those of Nlrp3 (Fig. 1b). In addition to the cultured neurons, adult mouse brains were also analyzed. Although the expression levels of Nlrp3 and Nlrc4 were noticeably increased in adult brains compared with those of cultured neurons, Aim2 expression levels were still the most elevated among the different inflammasome sensors when detected with the Universal ProbeLibrary system, as well as with the SYBR Green system (Fig. 1c,d). To further confirm the higher expression levels of Aim2 in the brain, the actual copy number of Aim2 was measured by absolute quantitative RT-PCR. In adult mouse brains, the copy number of Aim2 mRNA was estimated to be ~190 per ng of total RNA, which was 6.2-and 18.2-fold higher than those of Nlrp3 and Nlrc4, respectively (Fig. 1e), further supporting the high expression of Aim2 in the brain. We then monitored the temporal expression of Aim2 and Nlrp3 by examining mouse brains at postnatal (P) day 0, 3 and 7, as well as at two months. Compared with P0 and P3, the brains of mice at two months of age had reduced (~50%) expression of Aim2 (Fig. 1f). In contrast, the mRNA expression level of Nlrp3 doubled in the brain during development from P0 to two months (Fig. 1g), though its expression level was still lower than that of Aim2 (Fig. 1e). These expression analyses suggest that AIM2 likely plays a role in brains.
IL-1β is released by cultured neurons upon TLR7 and AIM2 activation. To examine AIM2 function in neurons, we first monitored IL-1β secretion upon AIM2 activation. Since TLR7 activation increases IL-1β mRNA expression 13 , to maximize the effect of AIM2 activation, we applied CL075 and poly-deoxyadenylic-deoxythymidylic acid (poly dAdT) -the synthetic double-stranded DNA (dsDNA) -to activate TLR7 and AIM2, respectively, in cultured cortical and hippocampal neurons. The amounts of IL-1β in the supernatant of neurons treated with CL075 and poly dAdT increased approximately 5-6-fold when compared with the vehicle control (Fig. 1h). AIM2 activation by poly dAdT alone slightly induced IL-1β expression, but the difference was not significant (Fig. 1h). The combinatory effect of CL075 and poly dAdT on IL-1β production in the cultured neurons required AIM2 because Aim2− /− neurons did not produce IL-1β in the presence of CL075 and poly dAdT (Fig. 1h). These results suggest that AIM2 expression enables neurons to produce IL-1β under dsDNA stimulation and TLR7 activation.
AIM2 regulates neuronal morphology. A previous study suggested the role of IL-1β in controlling the cell morphology of cultured neurons 38 . It seems possible that AIM2 controls IL-1β secretion and thereby regulates neuronal morphology. We investigated the role of AIM2 in neuronal differentiation and studied whether IL-1β acts downstream of AIM2 in regulating neuronal development. We first manipulated the expression levels of Aim2 and monitored the dendritic and axonal growth of cultured neurons. Two different experiments were performed. The first experiment was to compare neuronal morphology of wild-type and Aim2− /− neurons. Total dendrite length, primary dendrite number and the number of dendritic branch tips were determined at 3, 5 and 7 DIV. The results showed that Aim2− /− neurons had longer total dendrite length at 5 and 7 DIV, more primary dendrites at 5 DIV and more branch tips at 7 DIV (Fig. 2a). In measuring axon formation, Aim2 deletion reduced both primary and total axon length and the number of axonal branch tips at 3 DIV (Fig. 2b). To confirm that the morphological changes in Aim2− /− neurons are due to Aim2 deletion, we reintroduced Aim2 to Aim2− /− neurons. The results showed that restoration of Aim2 expression reduced total dendrite length, primary dendrite number and the number of dendritic branch tips (Fig. 2c). Similarly, re-expression of Aim2 had the opposite effect on axonal growth-increasing both total and primary axon length (Fig. 2d). This suggests that the role of AIM2 in neurons is more complex than a simple control of cell growth or death, instead likely involving a specific function to influence dendritic and axonal differentiation.
The second experiment was to examine the effect of AIM2 in vivo using Golgi staining to analyze Aim2− /− mouse brains. In four-month-old mouse brains, Aim2− /− CA1 neurons exhibited a noticeably different pattern of dendritic arbors compared with WT neurons (Fig. 3a,b). For basal dendrites, both total dendrite length and the number of total dendritic branch tips were noticeably increased in Aim2− /− CA1 neurons (Fig. 3c). In contrast, Aim2 deletion did not influence the length and branching level of apical dendrites (Fig. 3c). Because neonatal brains had higher Aim2 expression levels compared to adult brains ( Fig. 1f), we also analyzed dendritic arbors in P7 brains. Similar to adult brains, total basal dendrite length was increased in Aim2− /− CA1 neurons at P7 (Fig. 3d,e), although the total number of branch tips did not differ between WT and Aim2− /− brains (Fig. 3d,e). Upon observing apical dendrites, we found that total dendrite length was not noticeably affected by Aim2 deletion. Nevertheless, the branching level of apical dendrites was higher in Aim2− /− CA1 neurons at P7 (Fig. 3d,e). These in vivo analyses support the role of AIM2 in controlling neuronal morphology at both neonatal and adult stages.

Activation of AIM2 inflammasomes influences neuronal morphology. In addition to manipulat-
ing Aim2 expression, we also investigated the role of AIM2 in neurons by activating AIM2 inflammasomes. We compared CL075 alone, poly dAdT alone, and a combination of CL075 and poly dAdT to the vehicle control. In our previous study, we showed that TLR7 activation by CL075 induces IL-6 expression, which ultimately restricts dendritic growth 13 . To exclude the contribution of IL-6, Il-6− /− cortical and hippocampal neurons were used. Consistent with our previous data 13 , CL075 alone did not alter the dendritic growth of Il-6− /− neurons (Fig. 4a,b). When CL075 was combined with poly dAdT, all dendritic parameters, including the total dendrite length, primary dendrite number and the number of dendritic branches, were noticeably reduced in Il-6− /− neurons (Fig. 4a,b), which suggests a negative role for AIM2 inflammasomes in dendritic growth. We also noticed that poly dAdT treatment alone was sufficient to downregulate the dendritic growth of Il-6− /− neurons (Fig. 4b), although it did not noticeably induce the amount of IL-1β expression observed for cultured neurons (Fig. 1h). To confirm the involvement of AIM2 in the effect of poly dAdT on neuronal morphology, poly dAdT was added into Aim2− /− neurons. Addition of poly dAdT did not influence dendritic growth of Aim2− /− neurons (Fig. 4c,d). These results suggest that AIM2 senses dsDNA in the environment to regulate morphology of cultured neurons.
IL-1β signaling is required for AIM2 to regulate neuronal morphology. The aforementioned results suggest that addition of poly dAdT is sufficient to activate AIM2 to restrict dendritic growth but is unable to noticeably increase IL-1β secretion into culture medium (Figs 1h and 4b). It is possible that the amount of IL-1β necessary to restrict dendritic growth was lower than the detection limit of our instruments. Alternatively, in addition  to IL-1β , AIM2 might use another effector to downregulate dendritic growth. To investigate whether IL-1β is involved downstream of AIM2 in regulating neuronal morphology, we examined the effect of poly dAdT on two mutant neurons: 1) Il1r1− /− neurons, which lack the most well-studied receptor for IL-1; and 2) Myd88− /− neurons, which lack the key downstream adaptor molecule MYD88 for IL-1 receptor signaling 39 . Both these mutant neurons are unable to respond to IL-1β . We found that, unlike wild-type neurons, poly dAdT did not influence axonal and dendritic growth of Il1r1− /− neurons (Fig. 5a,b). Similarly, Myd88− /− neurons also did not respond to poly dAdT in terms of controlling axonal and dendritic growth (Fig. 5c,d). These data support the notion that IL-1β is required for poly dAdT to control neuronal morphology.
AIM2 activation cell-autonomously regulates neuronal morphology. Although we found that AIM2 responded to poly dAdT stimulation and used the IL-1β pathway to regulate neuronal morphology, our data also show that poly dAdT alone did not significantly increase IL-1β secretion into the culture supernatant. It seems possible that a very low amount of IL-1β is sufficient to regulate neuronal morphology, in which case an autocrine mechanism is likely the most effective way for a low amount of neuronal IL-1β to control neuronal morphology. To investigate whether AIM2 uses a cell-autonomous mechanism to influence neuronal morphology, we  reduced Aim2 expression by transfecting an Aim2 knockdown construct. In our neuronal culture systems, transfection efficiency is usually less than 1%. Therefore, transfected neurons are surrounded by untransfected neurons, providing a good system to test the cell-autonomous effect of AIM2. If AIM2 acts cell-autonomously to regulate neuronal morphology, reducing Aim2 expression by an Aim2 knockdown construct should be sufficient to generate the abnormal neuronal morphology observed for transfected neurons. To investigate this possibility, we first constructed five different artificial miRNA knockdown constructs of Aim2. In HEK293T cells, miR-Aim2#2 exhibited the best effect in reducing the protein expression of Myc-tagged Aim2 (Fig. 6a). MiR-Aim2#2 (termed MiR-Aim2 hereafter) was then transfected into neurons. The results showed that miR-Aim2 expression in neurons resulted in longer total dendritic length and shorter total axonal length (Fig. 6b,c), which is similar to the phenotypes of Aim2− /− neurons (Fig. 2). Moreover, if AIM2 activation acts non-cell-autonomously to regulate neuronal morphology, Aim2 knockdown neurons should still be influenced by IL-1β secreted from neighboring neurons upon poly dAdT treatment. To investigate this possibility, poly dAdT treatment was included. We found that poly dAdT restricted dendritic arborization of neurons transfected with non-silencing control miR-Ctrl, but this effect did not occur in the Aim2 knockdown neurons (Fig. 6d). Taken together, these results suggest that the effect of poly dAdT on neuronal morphology is mediated by an AIM2-dependent cell-autonomous mechanism.

Aim2 deletion induces anxious behaviors and impairs auditory fear memory. The above results
suggest the role of AIM2 in regulation of neuronal morphology. We then wondered whether AIM2 is relevant for brain functioning. A series of behavioral paradigms, including open field, light-dark box and fear conditioning, were applied to analyze Aim2− /− and WT mice. In open field, Aim2− /− mice had noticeably lower locomotor activity, as shorter moving distances were recorded on all three examination days (Fig. 7a). Because the moving speed of Aim2− /− mice was comparable to that of WT mice (Fig. 7a), the shorter moving distance was not caused by lower moving speed. The total number of rearing events also tended to be lower in Aim2− /− mice, but this was only statistically significant for Day 2 (Fig. 7a). We noticed that Aim2− /− mice preferred to stay in the corners of their boxes (Fig. 7a), and also defecated and urinated more in the open field (Fig. 7a), both of which are suggestive of anxious behaviors in Aim2− /− mice. To confirm this point, we performed light-dark box experiments. Indeed, compared with WT animals, Aim2− /− mice spent more time in the dark box, though the numbers of transitions between light and dark boxes were comparable between WT and Aim2− /− mice (Fig. 7b). These data support that Aim2 deletion induces anxiety. In fear conditioning, the freezing responses during habituation and the period right after foot shocks were comparable between Aim2− /− and WT mice (Fig. 7c). However, one day after auditory fear conditioning, Aim2− /− mice had a noticeably lower freezing percentage in response to tone stimulus (Fig. 7c), suggesting a role for AIM2 in regulating associative memory.
In conclusion, our analyses suggest that Aim2 deletion influences brain functions.

Discussions
In this report, our data suggest that neuronal AIM2 inflammasomes are able to sense dsDNA to promote IL-1β secretion by neurons. IL-1β then acts as an autocrine signaling mechanism to downregulate dendritic growth, whilst promoting axonal growth. In classical innate immune responses, the combinatory signals that activate TLRs and inflammasomes provoke proteolytic processes and massive secretion of the IL-1 family of cytokines 27,40 .
In cultured neurons, we also found that, by adding exogenous agonists, activation of both TLR7 and the AIM2 inflammasomes promotes IL-1β secretion. However, in the absence of exogenous ligands, manipulation of Aim2 expression levels in cultured neurons was sufficient to regulate neuronal morphology, which is perhaps due to endogenous ligands released from dead cells or healthy neurons in cultures, echoing previous studies showing that endogenous ligands of TLR7 regulate neuronal death and morphology 13,16,19 . In this scenario, it is likely that there is a basal level of IL-1β in cultures, which is also able to regulate cell growth of neurons. Manipulation of Aim2 expression levels may influence the processing of this basal level of IL-1β . Since the level of secreted IL-1β is so low, it can only act as an autocrine signal to cell-autonomously control neuronal morphology. This mechanism is likely involved in regulation of neural development in vivo. As for our analysis of Aim2− /− brains, we found that without any immune challenge, Aim2− /− CA1 neurons already had a more complex dendritic arbor at P7. Perhaps, apoptotic cells during development serve as a source of endogenous ligands to activate the AIM2 inflammasomes of neurons. These neuronal innate immune responses allow neurons to detect their growing environment and avoid extending their dendrites to regions with dead cells.
Our study suggests that AIM2 activation restricts dendritic growth but promotes axonal extension. Thus, the effect of AIM2 on neuronal morphology is unlikely to be due to the induction of cell death. In fact, IL-1β has been shown to have a differential effect on dendrites and axons. It can downregulate the dendritic growth of cultured neurons 38 . For axonal growth, however, the reported function of IL-1β is controversial. During the development of sympathetic neurons, 100 ng/ml IL-1β inhibits nerve growth factor-induced axonal growth 41 . In contrast, IL-1β at a concentration of 50 ng/ml stimulates neurite growth of enteric neurons 42 . High doses (500 ng/ml) of IL-1β trigger axonal growth of entorhinal cortex slice cultures, but not transverse spinal cord slice cultures 43 . Thus, it seems likely that different types of neurons respond differently to IL-1β in terms of axonal growth. Nevertheless, taken with our findings, these studies suggest IL-1β may play a role in the regulation of neuronal development, not just induction of pyroptosis (a type of cell death induced by inflammation).
In addition to IL-1β , there are several examples showing that cytokines or growth factors are able to act as autocrine signals regulating neuronal growth. Our previous study showed that TLR7 uses IL-6 to restrict axonal and dendritic growth 16 . Depending on the concentration, IL-6 may act as an autocrine or paracrine signal to regulate neuronal morphology 13,16 . Neurons are able to secrete brain-derived neurotrophic factor (BDNF) at the growth cone. Locally-secreted BDNF then cell-autonomously regulates axonal differentiation 44 . Taken together, autocrine signaling may be a general mechanism for neurons to sense environmental cues and to control their growth and differentiation.
Our study shows that Aim2 deletion induces anxious behaviors and impairs locomotion and associative memory. Previous human genetic and knockout mice have indicated the roles of IL-1 signaling in cognition 45 . The IL-1 receptor is heterodimeric, containing a ligand-binding subunit (IL-1R1) and essential accessory subunits. IL1RAPL1, one of the accessory subunits, regulates postsynaptic signaling 46 and is associated with mental retardation 46,47 . Deletion of Il1rapl1 in mice results in learning deficiency 48 , which is consistent with our observation that Aim2 deletion impairs auditory fear conditioning. However, in contrast to our finding, reduced anxiety behaviors were found in Il1rapl1− /− mice 48 . For Il1r1− /− mice, hyper locomotion and decreased anxiety in open field were observed 49 , which also contrast with the phenotype of Aim2− /− mice. Several possibilities may explain the conflicting observations in these mutant mice. For instance, there are at least nine different IL-1 receptors and accessory subunits 50,51 . Different subunits have different expression patterns and vary in regulation or function. It is not clear if there is complex interplay between these different subunits, which could manifest in altered mouse behavior. It may also explain why many controversial results were observed for the IL-1 study 45 . Moreover, AIM2 is just one of various inflammasome sensors that processes pro-IL-1β and Aim2 deletion can only impair IL-1β production in response to dsDNA and perhaps only in a subset of cell types. Thus, the impact of Aim2 deletion and IL-1 receptor deficiency is likely to be different. To resolve this issue, it will be necessary to delete Aim2 or IL-1 receptors in a cell type-specific manner. Conditional knockout mice are therefore needed to address the contribution of different types of cells to the AIM2-dependent inflammatory response. Chemicals and antibodies. We used rabbit polyclonal GFP (Invitrogen), mouse monoclonal MAP2 (Sigma-Aldrich), mouse monoclonal SMI-312R (Covance), mouse monoclonal Myc (Cell Signaling, 9B11), rabbit monoclonal β -tublin (9F3, Cell Signaling), Alexa Fluor 488-and Alexa Fluor 594-conjugated secondary antibodies (Invitrogen), and HRP-conjugated secondary antibodies (GE Healthcare), as well as CL075 (InvivoGen) and poly dAdT (InvivoGen). Poly dAdT is known to activate AIM2 inflammasomes 14,55,56 . The length of poly dAdT used in this report is 1184 bps.

Methods
Plasmids. The full-length of the Aim2 construct, pUNO1-Aim2, was purchased from InvivoGen and further subcloned into GW1-myc vector. Aim2 miRNA constructs were generated using the BLOCK-iT TM Pol II miR RNA1 Expression Vector Kit (Invitrogen). A plasmid pcDNA 6.2-GW/EmGFP-miR-neg (miR-ctrl, Invitrogen), which was predicted to not target any vertebrate gene, was used as a negative control. The detailed plasmid construction strategy and primer sequences are available in the Supplementary Materials. Cell culture, Transfection, Neuronal Morphometry, Q-PCR and ELISA. Cortical and hippocampal neurons were collected from E17.5 mouse embryos and cultured in Neurobasal medium/DMEM (1:1) with B27 supplement, penicillin, streptomycin and glutamine (Invitrogen). For most transfections, the calcium phosphate precipitation method was used as described 57 . For poly dAdT transfection, we used the Lipofectamine 2000 system (Invitrogen). To investigate neuronal morphology, cells were seeded at a density of 2.5 × 10 5 cells/well in 12-well plates with 18 mm poly-L-lysine-coated coverslips. For ELISA and Q-PCR, cells were seeded at a density of 2.5 × 10 6 and 1 × 10 6 cells/well in poly-L-lysine-coated 6-well plates, respectively. The detailed methods for neuronal morphometry, Q-PCR and ELISA 13 are available in the Supplementary Materials.
Animals and behavioral analyses. The animals used in the behavioral assays were offspring of Aim2 knockout or wild-type C57BL/6J. Male mice were used for behavioral assays to rule out a hormonal effect of female mice on offspring behaviors. All of the procedures for the behavioral analyses have been described Scientific RepoRts | 6:32405 | DOI: 10.1038/srep32405 previously 15,58,59 with minor modification. Behavioral analyses were performed in sequence as follows: week 12, open field; week 13-14, light-dark box; week 14-16, auditory fear conditioning. The animals were moved from the housing room to the behavioral room for accommodation for 1 week prior to the behavioral tests. Before tests, the mice were put into a new cage for 10 min habituation. After each test, the mice were put into another new cage until all mice had finished the test. Therefore, the tested mice did not affect the non-tested mice. All tests were performed from 14:00 to 19:00. The detailed protocols for each behavioral paradigm are available in the Supplementary Materials. Statistical analysis. Statistical analyses were performed using GraphPad Prism. For data in Figs 2,3,5 and 7, an unpaired two-tailed t-test was used. For data in Figs 1 and 4, a one-way ANOVA with Bonferroni correction was used. For data in Fig. 6, a two-way ANOVA with variance was used. The data are presented as the mean ± s.d. (Fig. 1)