An XBP1s-GFP screen to identify microbiome molecules with UPR-modulating activity
Gut metabolites are jointly derived from diet, modified human metabolites and microbially derived compounds that act as key mediators between the gut microbiome and host biology. For example, short chain fatty acids (SCFAs) such as butyrate, acetate, and propionate are produced by gut bacteria and modulate host cell functions such as histone deacetylase (HDAC) activity, gene expression, cell proliferation, and immune response. Commensal microbes can also alter pools of available host-generated metabolites such as tryptophan derivatives. Previous work from our group has identified differences in fecal and serum metabolites in the context of IBD [25]. To investigate how these metabolites impact mucosal homeostasis and inflammation, we selected a library of 399 molecules (Additional file 1, Table S1). The library includes 37 amino acid derivatives and 32 short chain fatty acid derivatives predicted from metagenomes/KEGG mapping, 93 bile acids with structures similar to cholic acid, 118 metabolites that were differently abundant between the stool samples of IBD patients and healthy individuals, as well as 119 bacterial derived metabolites. We henceforth refer to these molecules as microbiome box (MBB) compounds.
To monitor ER stress responses, we measured XBP1s since all three UPR signaling pathways upregulate XBP1s expression upon protein-folding stress. We generated an XBP1s-GFP knock-in reporter in a human colorectal adenocarcinoma cell line (HT-29) by CRISPR gene editing (Fig.1a). HT-29 in vitro culture is commonly used to model absorption, transport, and secretion by intestinal cells. As expected, addition of 2 μg/mL Tm (~ 2 μM) induced fluorescence by 3-fold in comparison to DMSO control after 12 h of incubation (Fig. 1b). The GFP induction by Tm is dose dependent with an EC50 of 0.2 μg/mL (Fig. 1b). This knock-in XBP1s-GFP reporter enables screening for molecules that modulate the ER stress pathway, as single-cell fluorescence can be measured in 384-well titer plates by high-content imaging and CellProfiler image analysis. We conducted two screens to identify ER stress induction activities, or ability to dampen UPR response induced by Tm treatment, using the MBB compounds (Fig. 1c).
We investigated the ability of MBB compounds to activate ER stress at four concentrations (50 μM, 15 μM, 5 μM to 1.5 μM), as indicated by an increase in green fluorescence of the XBP1s-GFP cells 12 h after compound treatment. Eight compounds were identified to induce fluorescence (Fig. 1d). For signal quantification, fluorescence was measured in the XBP1s-GFP cell line as well as the parental HT-29 cell line, which represents a baseline control. The ratio of fluorescence in the two cell lines in response to compound treatment (XBP1s-GFP/WT) therefore indicates the increase in fluorescence specifically due to XBP1s expression, correcting for chemical autofluorescence. As expected, the addition of 0.5% v/v DMSO does not induce relative fluorescence (Fig. 1e), whereas the addition of 1 μg/mL Tm activated relative fluorescence by 2.5-fold. Five of the 8 hit compounds (A1, A2, A5, A6, A7) were considered active (Fig. 1e, Additional file 2, Fig. S1a). Intriguingly, A1 (7-dehydrocholesterol), A2 (carnosol), and A6 (epoxysqualene) were found in our previous studies to be decreased in the stool of IBD patients compared to that from healthy individuals [25]. A5 and A7, two bacterial metabolites that induced the strongest response in the XBP1s-GFP cells after 12 h also resulted in a similar response after 36 hrs incubation (Additional file 2, Fig. S1b). None of the molecules affected HT-29 cell viability as measured by CellTiter-Glo assay, except for A7 treatment at 50 μM, which reduced viability by 50%.
Since activation of terminal UPR by ER stressors could lead to apoptosis and inflammation, we rationalized that dampening down the UPR response using bacterial-derived molecules could be a viable therapeutic strategy. Therefore, we also screened MBB compounds for inhibition of Tm-induced fluorescence in the XBP1s-GFP reporter. Cells were pre-treated with compounds for 2 h before incubating with 0.75 μg/mL of Tm for 12 h. DMSO pretreatment followed by Tm resulted in 2.5-fold GFP induction compared to no Tm addition (Fig. 1f). As a control, pre-treatment with 5 μM of the XBP1 inhibitor mycotrienin II (denoted Myc II) almost completely abolished the Tm-induced GFP fluorescence. Our screen identified three MBB compounds that dampened Tm-induced XBP1s-GFP (Fig. 1d, f). BZI687 (B1) and BZI688 (B2) reduced GFP signal by 80% and 50%, respectively, albeit only when supplied at 50 μM. Soraphen A (B3, denoted SoraA), however, inhibited XBP1s-GFP at 15 μM and 50 μM concentrations (Additional file 2, Fig. S1c). We then tested whether the efficacy of these inhibitors could be enhanced using a low level of Tm. Indeed, with 0.2 μg/mL Tm, these three molecules significantly reduced XBP1s-GFP when supplied at 15 μM and 50 μM (Additional file 2, Fig. S1d).
To validate the activity of these compounds in wild-type HT-29, we measured mRNA levels in these cells. As expected, Tm induced the spliced-form of XBP1 (XBP1s) relative to unspliced XBP1u by up to 50-fold (EC50 0.3 μg/mL) compared to DMSO (Additional file 2, Fig. S2a). HSPA5 and DDIT3, two UPR target genes, were induced up to ~ 30-fold (Additional file 2, Fig. S2a, b). Addition of A2 and A6 induced these two transcripts by only 2–3-fold (Additional file 2, Fig. S2b). By contrast, 50 μM A5 or A7 increased both transcript levels by ~ 15 to 30-fold. A5 (N-octanoyl-Met-Phe-H, structure shown in Fig. 2a) is an acylated Phe-Met dipeptide aldehyde, whereas A7 is trisindoline (structure shown in Fig. 3a).
Consistently, the inhibitor molecules B1 and B3 decreased Tm-induced expression of HSPA5 and DDIT3 (Additional file 2, Fig. S2c). 4μ8C, an IRE1 RNase-selective inhibitor, on the other hand, only reduced XBP1s but not HSPA5 or DDIT3 (Additional file 2, Fig. S2d). Taken together, our screen identified several MBB compounds with ER stress modulation activities; among these, we chose to focus on the most active molecules in this screen, the activators A5 and A7 and the inhibitor SoraA, for subsequent studies.
Dipeptide aldehydes induce UPR pathways
A5 is produced by expression of a Clostridium non-ribosomal peptide synthase gene cluster BGC0001686 in E. coli [34]. Dipeptide aldehydes similar to A5 are commonly made by gut bacteria containing such biosynthetic gene clusters [35]. For instance, IJA655 is co-produced with A5 by BGC0001686 over-expression (personal communication with Voigt lab) [34]. Its acyl chain is 4 carbons longer than that of A5 and harbors an internal double bond (Fig. 2a). Similar to A5, IJA655 induced relative fluorescence in the XBP1s-GFP cell line at 15 μM and 50 μM (Fig. 2a). Both A5 and IJA655 increased the amount of XBP1s mRNA (Fig. 2b). However, A5 reduced unspliced XBP1u levels by 50% whereas Tm and IJA655 slightly increased Xbp1u by about two-fold (Additional file 2, Fig. S3). This result indicates that A5 and IJA655 might have a different effect on transcriptional regulators of XBP. In addition, A5 maximally upregulated HSPA5 and DDIT3, similar to the magnitude of induction by Tm (Fig. 2b) while IJA655 induced these two genes to a lesser extent. Expression of IRE1α and PERK, encoding two upstream regulators of UPR, are also activated to a lesser extent by IJA655 compared to A5 (Additional file 2, Fig. S3).
To test if dipeptide aldehyde upregulates UPR protein expression levels, we analyzed HT-29 cell extracts by Western blotting (Fig. 2c). 1 μg/mL Tm strongly induced XBP1s, CHOP and ATF4 compared to the DMSO control. We also measured these protein levels in HT-29 cells when exposed to different lower concentrations of Tm (Additional file 2, Fig. S4a). XBP1s and ATF4 induction showed a gradual response to Tm with an EC50 of 0.2 μg/mL. The CHOP protein, however, was only detectable when cells were treated with more than 0.5 μg/mL Tm. The low sensitivity of CHOP induction to Tm was not due to the limit of detection of Western blotting, because CHOP in the lysates prepared from 0.5 μg/mL Tm treated HT-29 cells remained detectable when diluted 1:8 (Additional file 2, Fig. S4b). The induction of XBP1s by 0.5 μg/mL Tm is estimated to be 40-fold by comparing the overexposed western blot of undiluted DMSO control cellular lysate and diluted lysates of Tm treated cellular lysates (Additional file 2, Fig. S4b). A5 induced XBP1 to a lesser extent (~ 4-fold less in comparison to Tm treatment), with no CHOP and ATF4 induction (Fig. 2c). However, p-eIF2α and IRE1α were induced by A5 at similar levels to Tm treatment. We hypothesize that mild ER stress caused by A5 could complicate the dynamics and steady-state concentrations of UPR proteins. Taken together, A5 induces ER stress in intestinal epithelial cells.
Dipeptide aldehydes are potent protease inhibitors [34, 35]. We rationalize that A5 induces ER stress by causing misfolded protein accumulation. Therefore, we asked whether inhibiting the activity of a particular class of protease inhibitors leads to UPR activation. To this end, we assayed a panel of 24 protease inhibitors (Additional file 3, Table S2) from three different families (metalloprotease inhibitors, serine/cysteine protease inhibitors, and proteasome inhibitors) for their ability to induce fluorescence in the XBP1s-GFP reporter system (Additional file 2, Fig. S5). Most of the metalloprotease inhibitors we tested did not induce GFP when compared to DMSO, with the exception of TNP-470 and Aladotril. Most serine/cysteine protease inhibitors induced XBP1s-GFP between 1.5- to 3-fold, except for TPCK (Tos-Phe-CH2Cl) and Butabindide (TPPII inhibitor, [36]). Notably, APC-2848 (aka. Mu-Phe-HPh-VSPh), an irreversible cysteine protease inhibitor with a dipeptide vinylsulfone motif, induced fluorescence to a level similar to IJA655. Finally, 3 out of 4 proteasome inhibitors tested induced GFP. These results suggest dipeptide aldehydes potentially activate UPR by inhibiting these proteases.
Indole-derivatives are potent ER stress inducers
In addition to dipeptide aldehydes, we also identified bisindole derivatives as a different class of activators of ER stress (Fig. 1d, e). A7 is a trisindoline isolated from organic extraction of an Escherichia fergusonii (DSM13698) culture grown in MP6 medium (see the “Methods” section). Trisindoline has also been isolated from several other bacterial phyla, including Vibrios, Rubrivivax, and Rhodococcus [37,38,39,40]. Trisindoline and other bisindole derivatives are presumably formed by condensation of indole-derivatives (indoxyl or isatin) during the bacterial life cycle, or non-enzymatically. Intriguingly, we have detected several other indole derivatives from our collection of commensal bacterial cultures. Specifically, three such molecules extracted from Porphyromonas uenonis also induced fluorescence in the XBP1s-GFP reporter (Fig. 3a). Similarly, all molecules triggered the ER-stress response at the transcript level, with AXN321 inducing the highest level of mRNA among the four indole-derivatives (Fig. 3b, 50-fold and 10-fold respectively for XBP1s and HSPA5). Consistently, AXN321 induced XBP1s and ATF4 moderately, similar to the induction by 0.2 μg/mL. No induction of CHOP was detected at the given concentrations (Fig. 3c). Our findings identified bisindole derivatives as a new class of small molecules with ER stress induction activity.
Soraphen A prevents tunicamycin induced UPR response
One potent inhibitor of Tm-induced UPR identified in our screen was SoraA, a cyclic polyketide produced by Sorangium cellulosum containing a biosynthetic gene cluster (MIBiG BGC000147). Notably, the XBP1 inhibitor Myc II is also a macrocyclic polyketide (structure shown in Fig. 4a). We compared how these two polyketides modulate Tm induction of UPR pathways (Fig. 4b). When the cells were pretreated with 5 μM Myc II followed by Tm, both XBP1u and XBP1s are below basal levels (by 20-fold and 150-fold respectively, compared to the DMSO control, Fig. 4b). Myc II also prevented Tm induction of XBP1s protein (Fig. 4c). SoraA reduced XBP1s and XBP1s activation by 10-fold and 5-fold, respectively, relative to DMSO treatment without affecting the transcript level of XBP1u (Fig. 4b, c). In addition to reducing the XBP1s/XBP1u ratio, SoraA also decreased the Tm-induced activation of IRE1α transcript and proteins (Additional file 2, Fig. S6a and Fig. S6b). These data suggest that SoraA functions predominantly by inhibiting IRE1α levels, whereas Myc II inhibits IRE1α splicing and XBP1 transcription. Despite these differences, both Myc II and SoraA eliminated ATF4 and CHOP protein activation (Fig. 4c), and PERK, HSPA5, and DDIT3 transcript activation induced by Tm (Additional file 2, Fig. S6a).
We examined whether these polyketides could reverse the UPR protein induction when triggered by Tm before inhibitor treatment (Additional file 2, Fig. S6b). Myc II eliminated Tm-induced activation of XBP1s, CHOP and ATF4 when Tm was added 1 h before treatment. SoraA decreased, but did not abolish, XBP1s and CHOP activation by Tm treatment. ATF4 induction could not be reversed by SoraA treatment. These results indicate that SoraA is less able to inhibit UPR response when given after Tm-mediated induction of ER stress has already occurred, suggesting a different mode of action from Myc II. SoraA was reported to inhibit acetyl Co-A carboxylase (ACC) and interfere with fatty acid elongation [41]. Therefore, we characterized a set of ACC inhibitors for their activity to antagonize the effect of Tm-induced UPR response using the XBP1s-GFP assay. Indeed, three out of four AAC inhibitors we tested, with the exception of 5-(Tetradecyloxy)-2-furoic acid (TOFA), reduced Tm-mediated XBP1s-GFP activation (Fig. 4d). Consistently, these AAC inhibitor molecules also reduced HSPA5 by Tm induction when added as pretreatment (Fig. 4e). These results suggest certain ACC inhibitors could antagonize the UPR response, demonstrating crosstalk between lipid metabolism and ER stress.
ER stress modulating metabolites induce apoptosis and compromise barrier function
As mentioned above, prolonged ER stress activation and UPR response can result in apoptosis. We tested these selected metabolites for their effects on HT-29 apoptosis after 2-day compound treatment using the PO-PRO-1/ 7AAD staining kit and FACS analysis. As controls, addition of 1–5 μg/mL Tm induced apoptosis by up to 3-fold (Fig. 5a). Treatment with the ER stress activators A5 and A7 induced apoptosis by 2.5-fold and 1.5-fold, respectively, whereas the ER stress inhibitors Myc II or SoraA had no effect (Fig. 5b). Pretreatment with Myc II or SoraA did not reduce Tm induced apoptosis (Fig. 5c). We hypothesized that induction of cellular apoptosis will disrupt epithelial barrier function, and tested transepithelial electrical resistance (TEER) in HT-29 cells formed in Transwell inserts. Indeed, addition of 2–5 μg/mL Tm decreases the TEER by 30–40% after day 1 and by 40–80% after day 3, respectively (Fig. 5d and Additional file 2, Fig. S7a). Addition of 1 μg/mL is insufficient to reduce TEER significantly, presumably due to the replenishing cellular growth. A7 decreased TEER after two days of incubation, while neither A5 nor SoraA significantly changed TEER (Fig. 5e). Similar results were recapitulated in Caco-2 cells as shown in Additional file 2, Figs. S7b and S7c.