Introduction

Bacterial therapy has re-emerged as a promising modality for cancer treatment, building on the pioneering works of William Coley in the late 19^th century(Carlson et al. 2020). As a living therapeutic agent, bacteria offer several advantages over traditional cancer treatments, including (1) active tumor targeting (Yamamoto et al. 2016; Kim et al. 2023), (2) tumor colonization (Westphal et al. 2008; Weibel et al. 2008), (3) immune system stimulation (Qiu et al. 2020; Yang et al. 2023), and (4) great engineerability for customized functionalities (Gurbatri et al. 2020; Hu et al. 2022; Canale et al. 2021). The tumor microenvironment (TME) provides unique cues, such as hypoxia, low pH values, and immune suppressors, which facilitate the selective targeting and colonization of certain bacteria, including E. coli (Ryan et al. 2009; Flentie et al. 2012). However, the restricted resource in the TME may also pose challenges for bacterial survival and growth, potentially limiting their anti-tumor efficacy. While the molecular physiology of E. coli has been extensively studied under well-controlled laboratory conditions (Han and Lee 2006; Mateus et al. 2020), such as rich medium or minimal medium, it is still unclear how E. coli adapts to the nutrition-limited and immune-responsive environment in tumors. An in-depth understanding of the intricate relationship between the TME and the adaptation of colonized bacteria may provide hints to unlock the full potential of bacterial therapy against cancers.

It is well-established that the innate and adaptive immunity are sequentially activated upon bacterial infection in humans. However, even before the innate immune response commences, a critical defense mechanism known as “nutritional immunity” serves as the first line of protection to hinder the bacterial infection. Nutritional immunity is employed by the host organism through restricting the availability of essential nutrients to the invading pathogens (Weinberg 1975; Murdoch and Skaar 2022). For examples, humans have evolved specialized proteins to chelate several trace minerals, such as iron, zinc, and manganese, keeping their free-form concentrations at very low level within the body (Andrews and Schmidt 2007; Kambe et al. 2015; Roth et al. 2013). In response, pathogens have counter-evolved mechanisms to evade the nutritional immunity. For instance, a variety of siderophores are developed by bacteria in order to acquire the ferric ions from the host (Kramer et al. 2020). Intriguingly, humans have further retaliated by evolving the antimicrobial proteins, such as lipocalin 2 (LCN2), that sequester the siderophores (Singh et al. 2015; Bachman et al. 2009). These co-evolutionary events highlight the importance of nutritional immunity in combating bacteria (Golonka et al. 2019). Previous studies on bacterial cancer therapy have often overlooked the role of nutritional immunity, potentially resulting in suboptimal therapeutic efficacy.

In this work, we aimed to understand how bacteria modulate their physiological states at the molecular level in response to the tumor microenvironment. We employed liquid chromatography-tandem mass spectrometry (LC-MS/MS) for a quantitative comparison of the proteome between E. coli cultured in nutrient-rich medium and E. coli colonized in tumors. We found that E. coli colonized in tumors dramatically increases the protein expressions involving in the enterobactin biosynthesis and the iron ion homeostasis. This finding suggested that E. coli was stressed in an iron-deficient environment. Driven by this discovery, we hypothesized that enhancing the iron acquisition ability of E. coli might potentially improve its anti-tumor activity. We engineered E. coli as a potent iron scavenger by introducing LCN2 blockades, such as cyclic-di-GMP or glycosylated enterobactin. The engineered bacteria successfully evaded the nutritional immunity and achieved superior anti-tumor activity with frequent complete remission in mouse models.

Results

Tumor is an iron-deficient microenvironment for bacterial colonization

While many bacteria are known to colonize and proliferate in the TME, it remains elusive how bacteria adapt to such a nutrition-limited environment as compared to a nutrition-rich one. To this end, we performed quantitative LC-MS/MS experiments to compare the proteome of E. coli colonized in the murine tumors or cultured in the rich medium (LB broth) (Fig. S1). Not surprisingly, there are many more proteins enriched in the rich medium than in the tumor condition. The Venn diagram and the volcano blot revealed hundreds of protein IDs enriched in the rich medium condition (Fig. 1a, 1b, Table S1). The Gene Ontology(GO)-term analysis revealed that many of these proteins are associated with the machineries of biosynthetic processes, cell division, and energy production (Fig. 1c), reflecting that the E. coli was under a highly proliferative state in the rich medium. Interestingly, there were also 71 proteins that were preferentially expressed in the tumor condition over the rich medium condition. The GO-term analysis and the hierarchical clustering revealed that many of these proteins are involved in the processes of enterobactin synthesis and iron ion homeostasis (Fig. 1d, 1e, and Fig. S2). Fig. 1f showed that the individual proteins in these two processes were markedly up-regulated in the tumor condition. These proteins are known to be tightly controlled by the iron sensing system in E. coli, and only become up-regulated under the stress of iron deficiency (Seo et al. 2014). It is worth mentioning that while enterobactin facilitates the uptake of ferric ions into bacteria, the host immune cells can counteract by secreting a protein called LCN2, which possesses a specialized pocket to bind and sequester enterobactin (Fischbach et al. 2006). To investigate this possibility, we employed LC-MS/MS and analyzed the LCN2 expression in the tumors with or without E. coli colonization. Indeed, we observed a significant up-regulation of LCN2 expression in the tumors with E. coli inoculation (Fig. 1g). Collectively, our data suggest that tumor is an iron-deficient environment for E. coli colonization.

Figure 1.

Cyclic-di-GMP-producing E. coli synergizes with iron chelators for cancer therapy

Based on the proteomic findings, we aimed to focus our therapeutic strategies on modulating the iron competition between bacteria and cancer cells in the tumor. First, we hypothesize that an iron chelator that lowers the effective pool concentration of iron may provide a selection pressure, which disfavors the growth of cancer cells over bacteria. We tested three iron chelators, deferoxamine, ciclopirox, and VLX600, which have been approved for clinical trials or medical applications (Lang et al. 2019; Qi et al. 2020; Fryknas et al. 2016). All three chelators showed high cytotoxicity toward the cancer cells, suggesting the essential role of irons for the survival of mammalian cells (Fig 2a). On the other hand, E. coli can better tolerate these iron chelators at relatively high concentrations. Among them, VLX600 was the most potent drug against the cancer cells (IC50 = 0.33 μM) and provided the largest therapeutic window (280-fold difference) between the cancer cells and E. coli. In addition to the iron chelator, we also attempted to find approaches for counteracting the enterobactin sequestering function of LCN2. It has been reported that cyclic di-GMP (CDG) can block the binding between LCN2 and enterobactin, therefore restoring the functionality of enterobactin (Li et al. 2015). We engineered the E. coli by introducing a plasmid that carries the gene of diguanylate cyclase (DGC), an enzyme responsible for catalyzing the biosynthesis of cyclic di-GMP (CDG) (Lv et al. 2019). We showed that the E. coli transformed with the DGC plasmid (hereafter referred to DGC-E. coli) actively synthesized CDG and secreted it into the supernatant as detected by the LC/MS-MS mass spectrometry (Fig. 2b). It is worth noting that CDG is also a potent ligand for the STING pathway, which may stimulate the anti-tumor immunity (Diner et al. 2013; Krasteva and Sondermann 2017). When applying the DGC-E. coli supernatant to the macrophages, the macrophages enhanced the IFN-β secretion, suggesting the activation of the STING pathway by CDG (Fig. 2c).

Figure 2.

We evaluated the anti-tumor activity of VLX600 and DGC-E. coli in a syngeneic mouse model. The MC38 tumor-bearing mice received VLX600 and/or DGC-E. coli as depicted in Fig. 2d. The VLX600 monotherapy only marginally suppressed the tumor growth as compared to the PBS control (Fig. 2e). The DGC-E. coli monotherapy, although inhibited the tumor progression to a certain degree, did not show superior efficacy than the wild-type E. coli without the DGC plasmid transformation. Strikingly, the combination of DGC-E. coli and VLX600 showed significantly improved efficacy as compared to the individual mono-therapies. The tumor sizes were greatly suppressed, and two out of the four mice achieved complete remission (Fig. 2e and 2f). Of note, the combination of the wild-type E. coli and VLX600 did not result in such an improved activity, suggesting the syngeneic interaction of CDG and VLX600. Finally, when the mice cured by the combinatorial therapy were rechallenged by the same cancer cell line, no tumor developments were observed (Fig. S3), suggesting the establishment of a durable anti-cancer immunological memory in these mice.

Salmochelin-secreting E. coli significantly impeded tumor growth

Encouraged by the results shown above, we sought to engineer E. coli into a more potent iron scavenger and avoid the usage of iron chelators due to the potential systemic toxicity. Because CDG binds LCN2 with a relatively weak affinity (K_d = 1.6 μM), it can only partially sequester LCN2 even using a very high concentration (Fig. S4). Alternatively, bacteria have evolved another strategy, known as glycosylated enterobactin or salmochelin, to escape from the LCN2 blockade (Fischbach et al. 2006). Some pathogenic E. coli strains carry a gene cluster called IroA, which consists 5 genes to perform enterobactin glycosylation and its downstream uptake and processing. The sugars on the enterobactin create steric hindrance to the LCN2 pocket, therefore completely abolishing the LCN2 binding. To investigate this effect, we cloned the IroA cluster into a plasmid and transformed it into a non-IroA-carrying E. coli strain BL21(DE3). We incubated the E. coli with varying concentrations of LCN2 and measured their viability by colony formation assay. Fig. 3a showed that the E. coli without the IroA plasmid transformation (referred as WT-E. coli) was sensitive to LCN2, whereas the E. coli transformed with the IroA plasmid (referred as IroA-E. coli) was significantly more resistant to LCN2. The IroA-E. coli also showed stronger potency in acquisition of the ferric ions than the WT-E. coli, presumably due to the expression IroN, a member of the IroA cluster in charge of transporting both non-glycosylated and glycosylated enterobactin (Fig. 3b). In line with this observation, we found that the enterobactin (including the glycosylated form) extracted from the IroA-E. coli was more cytotoxic to the cancer cells than that extracted from the WT E. coli (Fig. 3c). In the mouse tumor experiments, the IroA-E. coli was significantly more efficacious than the WT-E. coli. Six out of ten mice treated by IroA-E. coli achieved complete remission, while none of the mice treated by the WT-E. coli experienced a cure (Fig. 3d-3f). Overall, we showed that the IroA cluster equips E. coli with an effective iron-scavenging capability, exerting a potent anti-tumor effect in the LCN2-rich tumor microenvironment.

Figure 3.

IroA-E. coli is less iron-deficient than WT-E. coli in the tumor microenvironment

Because the salmochelin secreted by IroA-E. coli is resistant to the sequestration of LCN2, we speculated that IroA-E. coli could have ameliorated the iron deficiency problem in the TME. To verify this speculation, we quantitatively compared the proteome of WT-E. coli and IroA-E. coli colonized in the tumors (Fig. 4a). The GO-term analysis revealed that the many proteins enriched in WT-E. coli belong to the enterobactin biosynthesis process and the iron homeostasis. Fig. 4b and Fig. 4c showed the fold changes of the individual proteins in these two terms. Strikingly, all of these proteins were expressed much higher in WT-E. coli than in IroA-E. coli. These results suggested that IroA-E. coli was less stressed by the iron deficient environment in the TME, presumably due to its superior iron scavenging ability from the glycosylated enterobactin. Besides the proteome of E. coli, we also examined the proteome changes in the host cells. We found that two iron-related proteins in mice, transferrin and transferring receptor, were elevated in the IroA-E. coli treatment as compared to the WT-E. coli treatment (Fig. 4d). These two proteins are known to up-regulate when the cells sense a lack of iron in the environment (Ponka and Lok 1999; Theil 1990; Ponka et al. 2015). Our finding suggests a competitive scenario between the host cells and E. coli where the potent acquisition of ferric ion by IroA-E. coli posed an iron-deficient stress to the host cells.

Figure 4.

Iron-scavenging IroA-E. coli significantly enhanced CD8+ T cell-mediated antitumor activity

Given that the IroA-E. coli treatment achieved long-term remission, we sought to investigate whether it also triggered the immunological memory against cancers. We re-challenged the mice cured by IroA-E. coli with the same cancer cell line (MC38) and found no tumors were able to develop (Fig. 5a), indicating a long-term anti-cancer memory has been established in these mice. Furthermore, we analyzed the tumor-infiltrating lymphocytes (TILs) in the mice treated by PBS, WT-E. coli, IroA-E. coli, respectively. We found that the TILs, especially the CD8+ T cells, were significantly elevated in the IroA-E. coli group (Fig. 5b). Given these observations, we speculated that the adaptive immune system might have contributed to the anti-tumor activity of the IroA-E. coli treatment to certain degrees. To verify this speculation, we performed the CD8+ T-cell depletion experiments in mice. We found that, indeed, the anti-tumor activity of IroA-E. coli was partially weakened in the mice treated with the anti-CD8 depletion antibody, and no long-term remission was achieved in these mice (Fig. 5c). Of note, the tumor progression was still significantly impeded under the CD8 depletion condition, indicating the inherent anti-tumor activity of IroA-E. coli still existed. Our findings suggest that the iron scavenging from IroA-E. coli may result in cancer cell deaths and tumor antigen releases, which in turn activate the adaptive immune cells for further cancer cell eradication.

Figure 5.

The systemic delivery of IroA-E. coli and oxaliplatin significantly suppressed tumor growth

It has been reported that E. coli possesses great tumor targeting ability following the intravenous injection in mice. Our previous study has also shown that the combination of E. coli and oxaliplatin synergistically suppresses the tumor growth. We therefore attempted to apply IroA-E coli and oxaliplatin for cancer treatment using a systemic delivery approach as depicted in Fig 6a. All mice maintained over 90% of their weights and remained active and healthy during the course of the treatments (Fig. 6b). Unlike the intratumoral injection, the intravenous injection of E. coli, either the wild-type or the IroA transformant, did not show substantial anti-tumor efficacy (Fig. 6c-6e). Also, the monotherapy of oxaliplatin only slightly inhibited the tumor growth. The combination of oxaliplatin with the wild-type E. coli delayed the tumor progression but did not achieve complete remission. Remarkably, the combination of oxaliplatin and IroA-E. coli significantly suppressed the tumor growth and achieved complete remission in 1 out of 4 mice. Overall, our data revealed that the systemic delivery of IroA-E. coli was synergistic with the oxaliplatin chemotherapy in the mouse tumor models.

Figure 6.

Discussion

There are accumulating studies that apply bacteria as drug delivery vehicles for cancer therapy with various payloads, including toxins, cytokines, immune checkpoint inhibitors, etc. However, in order to optimize the therapeutic outcomes of these engineering endeavors, it is also very important to understand the bacterial adaptation in the TME. The growth conditions differ vastly between the in vitro cultivation and the intratumoral environments. The rich medium provides ample nutrients for the optimal bacterial growth and payload production, whereas tumor is a nutrient-deprived, acidic, and hypoxic environment, which may stress the bacteria and restraint the engineered functionality. Also, unlike the in vitro cultivation, bacteria face the competition from the cancer cells as well as the surveillance from the host’s immune system. This work provides the first proteomic data comparing E. coli cultured in the rich medium and colonized in the mouse tumor. The results clearly show that the TME is an iron-deficient environment for E. coli. Moreover, upon the bacterial inoculation, the host cells up-regulate LCN2 to further block the iron uptake of E. coli by enterobactin. This observation inspired us to develop E. coli that overcame the nutritional immunity from the host. This discovery-driven design, especially the IroA implementation, proves to be highly effective in enhancing the anti-tumor activity of E. coli. It is of interest to investigate that if the iron uptake ability is also critical for other types of bacteria when applied in cancer therapy.

In addition to the enterobactin biosynthetic process and the iron ion homeostasis, our proteomic data also revealed other insights into the key nutrients whose scarcity within the TME may impede the bacterial growth. For examples, the proteins involving “the de novo synthesis of IMP” are highly enriched in the E. coli grown in tumors compared to the rich medium condition. Inosine phosphate (IMP) is the precursor of purine. When the environment is deficient of purine, the bacteria need to synthesize them using de novo pathways (Rolfes and Zalkin 1988; Cho et al. 2011; Meng et al. 1990). A previous report by Samant et al. has shown that the genes associated with the de novo purine synthesis are the most critical factors for the growth of E. coli or other gram-negative bacteria in human serum (Samant et al. 2008). Their data indicate that, similar to iron, humans also control the purine at a very low level in blood as a strategy of nutritional immunity in order to restraint the proliferation of bacteria. In light of these findings, one potential avenue could involve engineering E. coli to bolster its ability for de novo nucleotide biosynthesis, therefore facilitating better adaptation to the TME. Given that bacteria and cancer cells vie for growths within the TME, an enhanced bacterial adaptation to the TME may potentially improve the anti-tumor activity.

Overall, our research revealed that the tug of war for iron plays a critical role when applying bacteria for cancer therapy. To aid the bacteria in this war, we have adopted several approaches, including the engineering of E. coli to secret cyclic-di-GMP and salmochelin. These methods have demonstrated effectiveness in treating murine tumors, resulting in a significant portion of complete remission. Notably, the cured mice have also established durable anti-tumor immunological memory. Our iron-scavenging strategy opens new avenues by overcoming the nutritional immunity hurdles in the realm of bacteria-based cancer therapy.

Materials and methods

Cancer cell strain and cultivation

The MC38 murine colon cancer cells were cultured in DMEM supplemented with 10%FBS, 1% penicillin-streptomycin solution (100U/ml), 1mM sodium pyruvate, 10mM HEPES, and 1% MEM Non-Essential Amino Acids Solution (100X). RAW264.7 macrophage cells were cultured in DMEM supplemented with 10%FBS, and 1% penicillin-streptomycin solution (100U/ml). All cells lines were maintained in a humidified incubator at 37□ and 5% CO₂.

Mouse experiments

All animal experiments were conducted under specific pathogen-free conditions according to the guidelines approved by the Animal Care and Usage Committee of Academia Sinica. Mice were housed at temperature of 19-23□ with a 12-hour light-dark cycle and a humidity of 50-60%. A maximum of five mice were housed in a single individually ventilated cage with soft wood for nesting. Tumor dimensions were measured using a caliper, and tumor volume was calculated using the following formula: 0.52 x ((tumor length + tumor width)/2)³. Mice aged 6-10 weeks were subcutaneously injected with 5 × 10□ MC38 murine colon cancer cells suspended in 100 μL PBS at the right flank. The Escherichia coli strain BL21(DE3) was cultured in LB medium at 37□ overnight. The overnight cultures were diluted 100-fold in fresh LB medium and incubated at 37□ for ∼3 hours until the log phage. Prior to the intratumoral injection, the bacteria density was determined by measuring the OD600 (1 OD = 4 × 10⁸ CFU/ml). For the proteomic experiment, the tumor-bearing mice were intratumorally injected with 4 × 10⁸ BL21(DE3) in 50 μL PBS when the tumor volumes reached approximately 150 mm³. The tumors were harvested the following day for the LC-MS/MS analysis. For the intratumor-injection-based therapeutic experiment, the tumor-bearing mice were treated when the tumor volume reached ∼150 mm³. The tumor-bearing mice were intratumorally injected with 4 × 10⁸ BL21(DE3) on day 0 and day 9. VLX 600 (4.5 mg/kg) was administered via intravenous injection every three days for a total of four injections. For the intravenous-injection-based therapeutic experiment, the tumor-bearing mice were intravenously injected with 1 × 10⁸ BL21(DE3) in 100 μL PBS every three days for four times. The Oxaliplatin (5 mg/kg) was administered intraperitoneally every three days for a total of five injections. For the CD8⁺ T cell depletion experiment, the mice received intraperitoneal injections of 100 μg anti-mouse CD8α antibodies (BioXCell, Cat# BE0061) on days -6, -2, 2, 6, and 10, along with the intratumoral injections of 4 × 10[ BL21(DE3) on days 0 and 9.

Proteomics sample preparation

For the tumor-colonized bacteria, the tumors were excised, and homogenized, and processed to extract intratumoral bacteria cells. Red blood cells were removed using RBC lysis buffer. The mouse cells were removed through low-speed centrifugation at 1,200 g for 2 min three times. The E coli was collected by centrifugation at 4,500 g for 20 min. The samples were lysed using 4% SDS, 100mM Tris-HCL (pH 9), and 1x protease inhibitor cocktail set III. The cell lysates were heated at 95[ for 5 minutes and sonicated for 15 min using a Bioruptor Plus (Diagenode). The supernatant was collected after centrifugation at 18,000 g for 30 min at 4□ .

Approximately 50 μL supernatant was mixed with 200 μL methanol, 50 μL chloroform, and 150 μL double-distilled water. The aqueous phase was removed after sitting the sample at room temperature for 10 minutes. Subsequently, the sample was mixed with another 150 μL methanol. The pellet was collected, dried for 20 minutes, and resuspended in 8 M urea and 50 mM triethylammonium bicarbonate buffer. The samples were reduced with 10 mM DTT, alkylated with 50 mM IAA, and digested using LysC and trypsin. Following acidification, the supernatant was loaded onto the SDB-XC StageTips (Rappsilber et al. 2007) and eluted by 80% ACN containing 0.1%TFA. The sample was lyophilized and stored at -20□ before further LC-MS/MS analysis. For the bacteria grown in the rich medium, the E. coli strain BL21(DE3) was cultured in LB medium at 37□ overnight. The overnight culture was diluted 100-fold in fresh LB medium and incubated at 37□ until the log phage (OD=0.6). The bacteria were harvested by centrifugation and treated similarly to the bacteria harvested from tumors.

LC-MS/MS experiments

The sample was loaded onto the trap column (2 cm × 75 μm i.d., Symmetry C18), and then separated on a nanoACQUITY UPLC System (Waters, USA) equipped with a 25 cm × 75 μm i.d. BEH130 C18 column (Waters, USA) using a 5-35% buffer B (buffer A: 0.1% formic acid; buffer B: 0.1% formic acid in acetonitrile) gradient as the separation phase and a flow rate of 300 nl/min. The total running time was 120 min. The mass spectrometric data were collected on a high-resolution Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) operating in the data-dependent mode. Full MS resolution was set to 60,000 at 200 m/z and the mass range was set to 350−1600. dd-MS2 resolution was set to 15,000 at 200 m/z. Isolation width was set to 1.3 m/z. Normalized collision energy was set to 28%. The LC-MS/MS data were matched with the human SwissProt database using the Mascot search engine v.2.6.1 (Matrix Science, UK) with the following parameters: the mass tolerance of precursor peptide was set to 10 ppm, and the tolerance for MS/MS fragments was 0.02 Da.

Proteomics data processing and statistical analysis

Raw MS data were processed using MaxQuant version 2.0.1. Database search was performed with the Andromeda search engine through the Uniprot database(Cox et al. 2011). Both protein and peptide level were filtered by a 1% false discovery rate (FDR). The variable modification setting included oxidation(M) and Acetyl (Protein N-term), and the fixed modification setting included carbamidomethyl(C). The “match between runs” was set as 1 minute, and the MaxQuant LFQ algorithm was employed for normalization. The statistical analysis was performed using Perseus version 1.6.15.0 and Prism version 8.0.2(Tyanova et al. 2016). The proteinGroups output table from MaxQuant was utilized for proteomics analysis. The potential contaminant, reverse, and only-identified-by-site were filtered out. The LFQ-intensity was log2-transformed and filtered for validity. NaN values were imputed, and bacterial cell LFQ intensity was normalized using z-score. The LFQ intensity of bacterial cells was normalized using z-score (n-average/standard deviation). A t-test with an FDR of 0.05 and S0 of 1 was performed to extract significantly different proteins. These proteins were uploaded to the DAVID database for biological interpretation, and the results were visualized in Prism. The raw data of LFQ intensity for significantly different proteins were averaged to calculate the difference in protein expression level.

Enterobactin extraction

BL21(DE3) was cultured in the LB medium at 37□ overnight. The overnight culture was 100-fold diluted to the M9 medium supplemented with 0.2% casamino acids, 0.2% glucose, 1 mM MgSO₄, and 1 mg/mL vitamin B1 and grown for 20 hours. The bacteria were removed by centrifugation, and the supernatant was sterilized using a 0.22 μm filter. The cell-free supernatant was acidified to pH=2 using 10N HCl. An equal volume of ethyl acetate was added to the acidified supernatant and mixed using Intelli Mixer ERM-2L. The organic fraction was collected after 30 min incubation at room temperature and dried using a miVac centrifugal concentrator. All samples were resuspended in DMSO and stored at -20□ for further experiments.

Cytotoxicity assay of enterobactin

1 × 10⁴ MC38 cells in a 96-well plate were treated with the enterobactin extracted from WT-E. coli or IroA-E. coli for 48 hours. The cell viability was measured using the Cell Counting Kit-8 (CCK-8) according to manufacturer’s protocol. A cell-free mixture was used as a background reference, and the untreated cells were used as a control.

Characterization of cyclic-di-GMP (CDG) secreted from DGC-E. coli

The DGC plasmid carries the gene of the diguanylate cyclase fragment 82-248 residues from Thermotoga maritima with a single mutation of Arg158Ala. The BL21(DE3) cells transformed with the DGC plasmid were cultured in LB at 37□ overnight. The overnight cultures were 100-fold diluted to fresh LB supplemented with 0.1mM IPTG and kanamycin (50 ug/ml) and cultured at 37□ for 20 hours. The bacteria were pelleted, and the supernatant was collected and filtered using a 0.22 μm strainer. The supernatant was analyzed by LC-MS/MS to identify the CDG.

Interferon-β quantification

5 × 10⁵ RAW264.7 cells in a 24-well plate were treated with the conditioned medium from DGC-E. coli or non-transformed E. coli for 18 hours. Subsequently, the cell culture medium was collected for IFN-β quantification. The IFN-β levels were measured using the Mouse IFN-beta ELISA kit (R&D, #P318019) following the manufacturer’s protocol.

Iron chelating assay

The iron uptake ability of bacteria was determined using chrome azurol S (CAS) assay (Louden et al. 2011). The BL21(DE3) cells with or without IroA transformation were cultured in the LB medium at 37□ overnight. The bacteria were collected by centrifugation, washed twice with PBS. 10⁷ bacteria were dropped on a CAS plate and incubated at 37□ overnight. The diameter of the cleared zone was measured to quantify the iron uptake ability of the bacteria.

Lipocalin 2 resistance assay

The IroA-E. coli or non-transformed E. coli was cultured overnight in the LB medium. Next day, the bacterial culture was diluted to fresh RPMI supplemented with 10% FBS by 100-fold and incubated at 37□ for 5 hours to reach the log phase. 10⁵ bacteria were treated with different concentrations of LCN2 and incubated at 37□ for 20 hours. The live bacteria were quantified by serial tittering on LB agar plates.

Tumor-infiltrating lymphocyte (TIL) analysis

Tumor tissues were cut into small pieces in the digestion buffer (1mL RPMI supplemented with 10% FBS, 0.33 mg collagenase type IV (Sigma, Cat#C5138), and 66 μg DNase I from bovine pancreas (Cyrusbiosicence, Cat#101-9003-98-9)) and transferred into a C tube followed by sample processing using gentleMACS. The cell suspensions were filtered through a 40 μm strainer. The RBC lysis buffer was added to the tumor suspension to remove red blood cells. The cells were then blocked using the CD16/32 Fc blocker for 5 minutes on ice. The T cells were stained with eFluor780 viability dye, PE-CD45, APC-CD4, and FITC-CD8a antibodies. The stained T cells were analyzed by flow cytometer (Attune NxT cytometer), and the data were processed using FlowJo software.

Acknowledgements

We thank Academia Sinica Core Facility and Innovative Instrument Project (AS-CFII-111-212). We thank the Academia Sinica DNA Sequencing Core Facility (AS-CFII-108-115).

Funding

This work was supported by an Academia Sinica Career Development Award (AS-CDA-108-L07) and the Ministry of Science and Technology, Taiwan (110-2113-M-001-064-MY3).

Competing interests

The authors declare no conflict of interest.

Author contributions

S.W.H. and K.Y.M. conceived and designed the experiments. S.W.H., S.K.L., Y.A.Y., and W.J.L. performed the experiments. S.W.H. and K.Y.M. discussed the experimental results and wrote the manuscript. All authors provided clarification, guidance, and revision on the manuscript.