Colorectal cancer-associated anaerobic bacteria proliferate in tumor spheroids and alter the microenvironment

Recent reports show that colorectal tumors contain microbiota that are distinct from those that reside in a ‘normal’ colon environment, and that these microbiota can contribute to cancer progression. Fusobacterium nucleatum is the most commonly observed species in the colorectal tumor microenvironment and reportedly influences disease progression through numerous mechanisms. However, a detailed understanding of the role of this organism in cancer progression is limited, in part due to challenges in maintaining F. nucleatum viability under standard aerobic cell culture conditions. Herein we describe the development of a 3-dimensional (3D) tumor spheroid model that can harbor and promote the growth of anaerobic bacteria. Bacteria-tumor cell interactions and metabolic crosstalk were extensively studied by measuring the kinetics of bacterial growth, cell morphology and lysis, cancer-related gene expression, and metabolomics. We observed that viable F. nucleatum assembles biofilm-like structures in the tumor spheroid microenvironment, whereas heat-killed F. nucleatum is internalized and sequestered in the cancer cells. Lastly, we use the model to co-culture 28 Fusobacterium clinical isolates and demonstrate that the model successfully supports co-culture with diverse fusobacterial species. This bacteria-spheroid co-culture model enables mechanistic investigation of the role of anaerobic bacteria in the tumor microenvironment.


Results
Microbial viability in a 3D tumor spheroid co-culture model is both species-and spheroid size-dependent. Previous observations of tumor spheroids have described characteristic oxygen and nutrient gradients that mimic those in solid tumors. Since F. nucleatum is consistently found to be enriched in the colorectal TME, we hypothesized that co-culturing anaerobic F. nucleatum with tumor spheroids may provide a niche for maintaining F. nucleatum viability outside the anaerobic chamber (Fig. 1A). To test this hypothesis, we developed a BSCC model. Two commonly used laboratory strains of F. nucleatum, ATCC 23726 and ATCC 25586, were co-cultured with varying sizes of tumor spheroids from the human colorectal adenocarcinoma cell line HT-29. Following incubation under aerobic conditions for 24 h or 48 h, the BSCCs were transferred into bacterial broth in an anaerobic environment to determine if bacteria remained viable. After 24 h, F. nucleatum 23726 was recovered from 1/3 wells with no tumor spheroids present, and from 3/3 wells with either a 5,000 cell or 40,000 cell tumor spheroid present (Fig. 1B). Similarly, F. nucleatum 25586 was recovered from 1/3 wells without a tumor spheroid, from 0/3 wells with 5,000 cell tumor spheroids, and from 3/3 wells when co-cultured with 40,000 cell tumor spheroids (Fig. 1B). When the same experiment was carried out over 48 h, neither strain was recovered in a viable condition from wells without tumor spheroids, or from wells containing 5,000 cell tumor spheroids. However, we were able to recover both strains from 3/3 wells when starting with 40,000 cell tumor spheroids (Fig. 1B). Similar results were observed for intermediate sized tumor spheroids (i.e. 10,000 or 20,000 cells) and for another human colon cancer cell line, HCT 116 ( Supplementary Fig. 1). When testing the BSCC model with a different anaerobic species, Faecalibacterium prausnitzii, which is not commonly associated with CRC 7,34 , no viable bacteria could be recovered at any timepoint for any tumor spheroid size ( Supplementary  Fig. 1), suggesting that this niche environment cannot universally support the growth and viability of all anaerobic microbes. Based on these results, we determined that BSCCs can harbor viable F. nucleatum for at least 48 h and that 40,000 cell BSCCs consistently yielded viable F. nucleatum.
We used live cell imaging to determine if any morphological changes could be observed in BSCCs. At 24 h after inoculation, a halo of biomass (based on bright-field imaging) began to emerge from BSCCs with F. nucleatum (Fig. 1C). The biomass halo was not observed in BSCCs with either heat-killed (HK) F. nucleatum strains, or when smaller tumor spheroid sizes (<40,000 cells; Fig. 1C, Supplementary Fig. 1) were used. When a GFP-labeled HT-29 cell line was used, the biomass halo did not show a fluorescent signal, suggesting that the halo was not composed of tumor spheroid cells, and was instead likely to be a bacteria emerging from the inside or underneath the BSCC structure (Fig. 1C).
Bacterial growth is associated with increased cytotoxicity to tumor spheroids. We sought to quantify the F. nucleatum bacterial load in BSCCs over time. We isolated total DNA from BSCCs and used qPCR amplification of the 16S rRNA gene to quantify F. nucleatum abundance, as compared to CFU counting, which would have relied on efficiently dismantling BSCCs without affecting F. nucleatum viability. We observed a rapid decrease in 16S rRNA gene abundance that reached a minimum around 12 h in both strains of F. nucleatum, indicating bacterial death likely due to oxygen exposure in the aerobic environment ( Fig. 2A). However, around 24 h, we observed a logarithmic increase in 16S rRNA gene abundance ( Fig. 2A), suggesting logarithmic growth of both F. nucleatum strains in the BSCC model. When cultured in the same media without the tumor spheroids, both F. nucleatum strains showed moderate growth in an anaerobic environment, and no growth in an aerobic environment ( Supplementary Fig. 2).
We tracked the fluorescent signal of GFP-labeled BSCCs over time and observed that the GFP signal became more diffuse (Fig. 1C) at later time points when incubated with viable F. nucleatum. We also noticed that upon pipetting, BSCCs were more fragile, indicating that F. nucleatum could be cytotoxic to the tumor spheroids. To characterize this effect over time, we employed a fluorescent dye exclusion viability assay. Through the first ~24 h, virtually no HT29 cytotoxicity was detected (Fig. 2B). However, between 24-48 h, during the logarithmic growth phase, a steep increase in cytotoxicity was detected in BSCCs containing either one of the two viable F. nucleatum strains (Fig. 2B). These results suggest that uncontrolled F. nucleatum growth could have cytotoxic effects toward colorectal cancer cells.
Viable F. nucleatum assemble extracellular biofilm-like aggregates in tumor spheroid microenvironment whereas HK bacteria are internalized. To gain additional insights into the arrangement of F. nucleatum in the tumor spheroid microenvironment, we used fluorescent immunocytochemical staining and confocal laser scanning microscopy (CLSM) on BSCCs with viable F. nucleatum, HK F. nucleatum, or vehicle control at 12 h, 24 h, and 36 h timepoints. At low magnification (5×), BSCCs displayed a hollow core, perhaps resulting from a zone of necrotic cells, previously reported to occur in larger spheroids (>500 μm) 30,31,35 . Aggregates of bacteria with relatively even distribution throughout the BSCCs were observed at 12 h, when they were co-cultured with viable or HK F. nucleatum 25586. However, viable F. nucleatum aggregates appeared to be smaller and more evenly distributed (Fig. 3A). At 24 h, viable F. nucleatum was more diffuse, with a few visible  www.nature.com/scientificreports www.nature.com/scientificreports/ aggregates and a large F. nucleatum-based biomass emerging from near the BSCC center (Fig. 3A). In contrast to BSCCs with viable bacteria, BSCCs with HK F. nucleatum displayed dozens of visible aggregates at 24 h (Fig. 3A). This suggested that viable and HK F. nucleatum were differentially distributed in the tumor spheroid microenvironment. A distinct center mass of cancer cells was observed in BSCCs with viable F. nucleatum that was not apparent in vehicle-or HK-treated BSCCs (Fig. 3A). At 36 h, the viable F. nucleatum-based biomass was observed to be protruding from the BSCC, with significant HT29 cytoskeletal rearrangement, whereas HK F. nucleatum-treated BSCCs looked largely unchanged from the 24 h timepoint (Fig. 3A).
BSCCs were also observed under high magnification (63×). At 12 h, both the viable and HK F. nucleatum appeared to be in direct contact with the cancer cells (Fig. 3B), and to be actively engaged with the cytoskeleton, as evidenced by colocalization with tubulin staining (Fig. 3B). By 24 h, a bacterial biomass had accumulated in the viable F. nucleatum co-culture, forming a structure resembling a biofilm, while the HK aggregates remained similar in size ( Fig. 3B). At 24 h, we also observed that the tubulin structure of BSCCs had been compromised when compared to vehicle control or to the 12 h viable F. nucleatum condition (Fig. 3B). On the other hand, for the 24 h HK condition, we observed an intact cytoskeleton with cellular internalization of the bacteria taking place (Fig. 3B). By 36 h, viable F. nucleatum displayed filamentous growth when compared to the same condition at 12 h. The observations from fluorescence CLSM further support our earlier conclusion that viable F. nucleatum forms biofilm-like biomass and progressively damages the structural integrity of tumor spheroids at 24 h and beyond, which was not observed in HK F. nucleatum or in vehicle conditions. Viable F. nucleatum differentially regulates cancer-related gene expression under proliferating conditions. F. nucleatum has been consistently associated with CRCs and detected in tumor tissues [6][7][8][9]14,19,[36][37][38] , and high levels of Fusobacteria are reported to be associated with a worsening prognosis in cancer patients 39 . Therefore, in an effort to understand how viable F. nucleatum affects gene expression changes in cancer cells, we used high-throughput qPCR to measure the expression of >500 cancer-related genes. BSCCs were co-cultured with viable or HK F. nucleatum 23726, viable F. nucleatum 25586, or vehicle control (medium alone) for 24 h before RNA was isolated and subjected to qPCR. Pairwise analyses of all three conditions versus vehicle control resulted in a combined total of 137 significantly differential expressed genes (DEGs; p < 0.05, unpaired T-test  Supplementary Fig. 3). DEGs were visualized in heat maps after hierarchical cluster analysis using Average Linkage (Pearson distance measurement, Fig. 4A) for individual replicates. The replicate measurements for viable F. nucleatum clustered together and separately from HK F. nucleatum and vehicle controls (Fig. 4A).
Canonical pathway enrichment of DEGs suggested that 177 different pathways were significantly enriched (p < 0.001, Fisher's exact test) between all three conditions; of which 101 were common between the BSCCs with the two viable strains and 10 in common between viable and HK conditions (Fig. 4B). Several of the most highly enriched pathways in BSCCs with viable F. nucleatum, but not in BSCCs with HK F. nucleatum, included molecular mechanisms of cancer, colorectal cancer metastasis signaling, and IL-8 signaling (Fig. 4C). We measured IL-8 www.nature.com/scientificreports www.nature.com/scientificreports/ protein levels at 24 h and observed ~2-3 fold greater IL-8 in BSCCs with viable F. nucleatum compared to vehicle or the respective HK strain ( Supplementary Fig. 4). Canonical pathways that were most highly enriched in BSCCs with both viable F. nucleatum and HK F. nucleatum, include PTEN signaling, STAT3 pathway, and virus entry via www.nature.com/scientificreports www.nature.com/scientificreports/ endocytic pathways (Fig. 4D). These pathways connect with findings in the literature and are further detailed in the discussion section.
When functional annotations of related DEGs were analyzed, there were increased numbers of disease functions attributed to co-culturing with viable F. nucleatum compared to HK F. nucleatum 23726 ( Supplementary  Fig. 4). Whereas the predicted activation state of "apoptosis in colorectal cancer cell lines" was negative for HK F. nucleatum 23726 treatment ( Supplementary Fig. 4), it was positive in BSCCs with viable F. nucleatum ( Supplementary Fig. 4). Altogether, these results support increased and differential biological activity at the transcriptional level when F. nucleatum is viable/proliferating in the tumor spheroid microenvironment.
Viable F. nucleatum affects metabolite levels in tumor spheroid microenvironment. We used a combination of targeted and untargeted metabolomics to determine if viable F. nucleatum alters metabolic processes in the tumor spheroid microenvironment. Targeted triple quadrupole LC-MS analysis of 155 central carbon metabolites (Supplementary Table 5) of BSCCs with viable F. nucleatum, HK F. nucleatum, or vehicle were measured at 0 h, 18 h, 24 h, and 42 h after inoculation. Cell-associated metabolite levels of BSCCs with viable or HK F. nucleatum were compared to vehicle-treated control BSCCs at each timepoint. Significantly different metabolites (p < 0.05, one way ANOVA) between any combination of treatments or timepoints were visualized in a heat map after hierarchical cluster analysis using Average Linkage (Pearson distance measurement, Fig. 5A) analysis. The BSCCs with viable F. nucleatum had metabolite patterns that distinguished them from BSCCs with heat-killed F. nucleatum or vehicle, most notably at the 42 h timepoint (Fig. 5A, Supplementary Fig. 5). Previous reports have described amino acids as the preferred substrate for energy generation in F. nucleatum 40,41 . Our results revealed that the levels of six different amino acids (histidine, tryptophan, glutamine, serine, methionine, threonine) decreased over time in BSCCs with viable F. nucleatum. Also, no significant accumulation of any amino acids was observed in these BSCCs (Fig. 5A, Supplementary Fig. 5).
Untargeted, global metabolomics analysis of the same samples and timepoints revealed 1,707 unique m/z-RT pairs across all samples and conditions. Principal component analysis (PCA) plots were generated for each time point (Fig. 5B). At 0 h and 18 h, considerable overlap between BSCCs with viable F. nucleatum, HK F. nucleatum, www.nature.com/scientificreports www.nature.com/scientificreports/ and vehicle was observed, suggesting that at relatively early timepoints the metabolite levels for these conditions did not vary significantly (Fig. 5B). However, at 24 h, the PCA plots show that the metabolite levels for BSCCs cultured with viable F. nucleatum, begin to separate from that of BSCCs with HK F. nucleatum or vehicle-treatment. At 42 h, the different groups of viable F. nucleatum and HK-/vehicle-treated BSCCs clearly separate (Fig. 5B). This suggests that throughout the course of the co-culture experiment, viable F. nucleatum affects the tumor spheroid microenvironment and shapes the global metabolome. These results also demonstrate that the BSCC model is capable of capturing metabolic responses of tumor spheroids to the presence of viable and proliferating CRC-relevant bacteria.
In a previous report, patient-derived colorectal tumors that were biofilm-positive had higher levels of polyamines compared to biofilm-negative, with N 1 ,N 12 -diacetylspermine being the most significant 42 . Using the www.nature.com/scientificreports www.nature.com/scientificreports/ data generated from our high resolution, accurate mass spectrometer, we identified N 1 ,N 12 -diacetylspermine and N 1 -acetylspermidine, confirmed by injection of authentic standards. The abundances for these metabolites slowly increased over time in BSCCs with HK F. nucleatum or vehicle-treatment, but drastically increased in BSCCs with viable F. nucleatum between 24 h and 42 h (Fig. 5C). This suggests that metabolic alterations detected in patient-derived tumors can be recapitulated in the BSCC model.
In addition to detecting dozens of other metabolites, we also identified acetyl-CoA (Fig. 5C). Acetyl-CoA is utilized by spermidine/spermine N 1 -acetyltransferase (SSAT) in the production of acetylated polyamines. Acetyl-CoA can also be used for fermentation by F. nucleatum, yielding butyrate, a preferred energy source for colonocytes that may have varying roles in CRC development 40,43 . Diverse clinical isolates of Fusobacterium remain viable in BSCCs and affect BSCC morphology. While F. nucleatum 23726 and F. nucleatum 25586 are model strains that are often used for in vitro CRC research 14,15,17,44 , we sought to determine if a greater diversity of Fusobacteria can persist in the BSCC model 45 . Therefore, we co-cultured HT29 tumor spheroids with an additional 28 clinical isolates of Fusobacterium, mostly consisting of F. nucleatum, but also comprised of Fusobacterium gonidiaformans, Fusobacterium necrophorum (2 strains), Fusobacterium periodonticum (2 strains), and Fusobacterium ulcerans. After 48 h of co-culture, BSCCs were imaged and then transferred into bacterial broth in an anaerobic environment to determine if bacteria remained viable. In addition to F. nucleatum 23726 and 25586, 21 of the clinical isolates were consistently viable (Fig. 6). Each of the five Fusobacterium species tested were represented within this group. As seen with F. nucleatum 23726 and 25586, several of these strains displayed the notable biofilm-like growth ( Fig. 6; left to right, EAVG_019 through EAVG_003). This phenotype appeared to be limited to F. nucleatum and F. necrophorum strains in the BSCC model. Other strains induced irregular BSCC morphological changes, including bacterial growth outward of the BSCC ( Fig. 6; EAVG_010, through EAVG_028), but not the same extent of biofilm-like growth as the aforementioned 8 strains. Other BSCCs with viable Fusobacteria showed no obvious morphological changes through 48 h ( Fig. 6; EAVO_002 through EAVG_015). Several strains had intermediate viability (or varying frequency of successful recovery), all of which had no obvious effect on BSCC morphology ( Fig. 6; EAVG_025 through CC2_6JVN3). We were unable to recover only one strain, F. periodonticum EAVG_011, from the BSCC model (Fig. 6).
To determine how the tumor spheroid might influence Fusobacteria growth morphology, Fusobacteria were similarly seeded into the same media and microplates (excluding tumor spheroids) and incubated for 48 h in anaerobic conditions. From this, Fusobacteria formed diverse aggregates and microcolonies that showed no obvious relation to their growth characteristics in the BSCC model ( Supplementary Fig. 5). This observation demonstrates that these bacteria are phenotypically different when grown in the BSCC model versus when grown under similar conditions without tumor spheroids in an anaerobic chamber.

Discussion
In vitro models are important tools that can be applied towards gaining mechanistic understanding, rapid hypothesis testing, and screening for tool compounds or therapeutic candidates. With the advent of new sequencing technologies, our appreciation for the role of the microbiome in various disease scenarios is rapidly expanding; however, previous attempts at building in vitro co-culture models to recapitulate these interactions have not evolved at the same pace. While the enrichment of Fusobacteria in the TME of CRC patients is a common finding [6][7][8][9]19,38,39 , and viable Fusobacteria can be cultured from the tumor 10 , there is currently a gap in our understanding of how the viability of these bacteria affect the TME. 3D cell culture technologies have the potential to bridge these knowledge gaps. 3D gut organoids are physiologically relevant systems that can be used to characterize interactions of various microbiota with a healthy, differentiated epithelium 27,28 , but may be less relevant to interactions of intra-tumor bacteria. On the other hand, 3D tumor spheroids, which were used in this study, serve as a representative in vitro model of the TME [29][30][31] . However, BSCCs are less likely to be useful in generally studying the vast microbiota interactions of the native intestinal lumen environment.
To characterize this model, we first demonstrated that F. nucleatum 23726 and 25586 consistently remained viable in 40,000 cell spheroids through 48 h. This F. nucleatum viability is likely in part due to the hypoxic niche that is characteristic of large spheroids 35 . Upon observing significant biomass increase (sometimes visible to the naked eye) concurrent with tumor spheroid fragility, we characterized the kinetics of F. nucleatum growth and tumor spheroid cytotoxicity. Using fluorescence confocal microscopy, we observed biofilm-like aggregate formation and altered tumor spheroid morphology as the BSCC progressed, which revealed that at 24 and 36 h viable F. nucleatum are primarily seen in extracellular aggregates in the tumor spheroid microenvironment, whereas HK F. nucleatum are internalized into the cytoplasm of the HT29 cells. This suggests that proliferating F. nucleatum either escape the cancer cells after internalization, avoid endocytosis by forming and residing in biofilm, or perhaps a combination of both. Indeed, evidence of bacterial biofilm in colon cancer has been previously reported 10,42,46,47 , further supporting that this is an important characteristic of bacteria in this niche. Future identification of biofilm components of F. nucleatum (or other bacteria) that are present in the TME could lead to new discoveries of how these bacteria interact with the tumor.
To build upon these qualitative findings from fluorescence microscopy, we took approaches toward molecular quantification of the interactions between F. nucleatum and the tumor spheroids, via high-throughput qPCR (and later on using high resolution metabolomics). We measured cancer-related gene expression in response to proliferating F. nucleatum 23726 and 25586 before substantial cytotoxicity was observed. The transcriptional response of HT29 cells to both viable F. nucleatum strains was very similar; perhaps unsurprisingly, as these strains are genetically quite similar 45 . Colorectal cancer metastasis signaling, IL-8 signaling, and molecular mechanisms of cancer were highly enriched in response to both viable strains. The latter (molecular mechanisms of cancer) may be unsurprising as the gene panel was based on cancer related pathways, but it is worth noting that this was not enriched in BSCCs with HK F. nucleatum. Increased IL-8 gene expression has been observed in patients with high Fusobacterium abundance 16 , and increased IL-8 protein levels were measured in BSCCs with viable F. nucleatum in this study. Interestingly, F. nucleatum has been associated with metastasis in multiple reports 6,18,19,44 and even cultured from metastatic lesions 10 . Several of the most highly enriched canonical pathways from HK 23726-treated BSCCs (e.g. PTEN, STAT3, viral entry via endocytosis) were also significantly enriched (p < 0.001, Fisher's exact test) in BSCCs with viable strains, and F. nucleatum has been previously connected to these pathways. F. nucleatum has been reported to down-regulate PTEN expression in epithelial cancer cells and induce STAT3 expression in macrophages, both via TLR4-dependent mechanisms 14,48 . F. nucleatum has also been shown to internalize via clathrin-mediated endocytosis upon binding E-cadherin with its FadA adhesin 13 . This suggests that the BSCC with viable F. nucleatum captures some of the signaling pathways that occur with HK F. nucleatum, likely a result of bacterial surface components. A limitation of our transcriptomic analysis was that the cancer-related gene panel only provides a focused subset of the transcriptome and it is likely that unbiased transcriptomic analyses (i.e. RNA-sequencing) would reveal additional pathways affected. While we are employing this co-culture model to better understand the effects of viable F. nucleatum, we must also point out that the bacterial density as a result of proliferation can also be a confounder in the experiment. At 24 h, both strains are entering logarithmic growth phase ( Fig. 2A) and therefore may be at a different density than the HK F. nucleatum.
We also characterized the metabolite content of BSCCs at various timepoints using high resolution metabolomics to gain an understanding of how viable F. nucleatum may influence metabolite levels and availability in the TME. From these experiments, we made two observations in this BSCC model that parallel what has been previously reported in non-coculture systems. First, we saw a depletion of six amino acids: histidine, glutamine, serine, methionine, threonine, and tryptophan. The former five of these have been reported as energy substrates in F. nucleatum 40,41 . The latter, tryptophan, has been shown to induce biofilm formation in F. nucleatum 25586 (which was seen in the BSCC), potentially through the production of the extracellular signaling molecule indole 49 . We also observed an increase in the polyamines N 1 -acetylspermidine and N 1 ,N 12 -diacetylspermine. Polyamines are important mediators in colon carcinogenesis, and polyamine metabolism has been a target of preclinical and clinical chemoprevention studies going back decades 50,51 . Interestingly, Johnson et al. observed increased levels of polyamines in patient-derived colorectal tumors that were biofilm-positive (in comparison to biofilm-negative tumors), and hypothesized that these metabolites were being contributed by the biofilm community 42 . While our observations in the BSCC model did not distinguish which species was utilizing or providing these metabolites (e.g. host vs. microbe), it establishes a platform for addressing these questions.
In order to more closely mimic the TME, tumor spheroid models have been developed to co-culture cancer cells with endothelial cells 52,53 , fibroblasts 53,54 , stellate cells 55 , or various immune cells [56][57][58][59][60] ; and three-way co-cultures are emerging 61,62 . While evidence for the influence of intratumor bacteria on cancer progression builds, this BSCC with colorectal cancer-associated intratumor bacteria adds a novel model to study this disease. While this current BSCC model only incorporates bacteria and cancer cells, these different models are Scientific RepoRtS | (2020) 10:5321 | https://doi.org/10.1038/s41598-020-62139-z www.nature.com/scientificreports www.nature.com/scientificreports/ continuously being integrated offering potential for future characterization of how intratumor bacteria affect other cell types in the TME beyond cancer cells. Although the CRC field has the most established evidence of a tumor microbiome, other tumor microbiomes are emerging such as in pancreatic cancers 63,64 , indicating this tumor-microbe relationship is likely not limited to CRC. Lastly, as F. nucleatum is not the only bacterium reported in the CRC TME, this model offers potential for studying other bacterial species and consortia in a reductionist system for more complete understanding of interspecies and interkingdom ecologies in the context of colorectal and other cancers.
Tumor spheroids were formed by resuspending cells in fresh medium without antibiotics and aliquoting 200 μl per well of a 96-well ultra-low attachment spheroid microplate (Corning, Cat No. 4520). Tumor spheroids were incubated for 48 h at 37 °C, 5% CO 2 , before inoculating with bacteria.
Bacterial strains used in this study are listed in Supplementary Table 1. All Fusobacterium strains were streaked from frozen stock on brain-heart infusion (BHI) agar (AS-6226, Anaerobe Systems, Morgan Hill, CA, USA) in an anerobic chamber (Coy Laboratory Products, Grass Lake, MI, USA) and incubated at 37 °C in the chamber. Faecalibacterium prausnitzii was streaked out onto yeast casitone fatty acids agar with carbohydrates (YCFAC, AS-675, Anaerobe Systems) in an anaerobic chamber and incubated at 37 °C. Liquid cultures of all bacterial strains were made by inoculating a single colony into 5 ml BHI broth (AS-872, Anaerobe Systems). Liquid cultures were incubated anaerobically for 48 h at 37 °C before use, yielding bacterial densities of approximately 18 × 10 6 CFUs/ml for ATCC 23726 and 4 × 10 6 CFUs/ml for ATCC 25586. To initiate BSCCs, liquid cultures were pelleted down, supernatant was discarded, pellets were resuspended in the same volume of BHI broth, and 4 μL of suspension were transferred to each well of the tumor spheroid microplate (MOIs of approximately 1.8:1 and 0.4:1 for ATCC 23726 and ATCC 25586, respectively). BSCCs were then incubated at 37 °C, 5% CO 2 (standard humidified incubator) from this point forward.
Live cell imaging. BSCCs were imaged using two separate live cell imaging platforms: Incucyte S3 Live-cell Analysis System and Biotek Cytation 5 Cell Imaging Multi-mode Reader, using standard brightfield, phase contrast, and fluorescence protocols. cell lysis assay. The CellTox ™ Green (Promega, G8741) cytoxicity assay was used according to manufacturer's protocol with slight modification. CellTox ™ Green Dye (1000×) was diluted 10-fold in fresh media (without antibiotics). BSCCs were initiated as described above. Vehicle control was 4 μL of sterile BHI added to tumor spheroid. The lysis solution accompanying the kit was added at 4 μL per well. At the time of adding F. nucleatum/ controls, 20 μL of CellTox Green Dye/cell media solution was added to each BSCC. Plates were placed in Incucyte S3 Live-cell Analysis System and whole-well imaged for both phase contrast and green fluorescence (acquisition time 300 ms) every 6 h. Green fluorescence signal in the images were analyzed using Incucyte software. Fixed threshold segmentation was applied (600 GCU threshold) and data is displayed as total green object integrated intensity (GCU × μm 2 /well).

16S quantification of bacterial growth. DNA was isolated from each BSCC using the Qiagen QIAamp
PowerFecal DNA Kit (12830-50) following manufacturer's instructions. Bacterial DNA abundance was assessed by qPCR amplification of the V1-V2 region of the 16S rRNA gene using the Taqman Fast Advanced qPCR master mix (Thermo Fisher 4444963) as previously described 65 . qPCR reactions were carried out in triplicate (20 μL each). Degenerate bacterial 16S rDNA-specific primers and probe (Sigma) with the following sequences were used: forward primer, 5′AGAGTTTGATCCTGGCTCAG3′; reverse primer, 5′-CTGCTGCCTYCCGTA-3′; probe: 5′-/56-FAM/TAA + CA + CATG + CA + AGT + CGA/3BHQ_1/3′, "+" indicates the position of an LNA base. A standard curve was prepared using a near full length amplicon of Escherichia coli 16S rRNA gene inserted into a Topo vector for normalization. Thermocycling was carried out using a QuantStudio 12 K Flex (Applied Biosystems) as follows: initiation at 95 °C for 5 min followed by 40 cycles of 94 °C × 30 s, 50 °C × 30 s, and 72 °C × 30 s.

IL-8 quantification.
After 24 h of co-culture, supernatants were collected and pooled for each BSCC. Each pooled supernatant was filtered using a 0.2 μm spin filter (Costar) in microcentrifuge tubes. Filtered supernatants were used to measure IL-8 concentrations using a Simple Plex Cartridge kit from Protein Simple with their Ella Simple Plex instrument. Briefly, 50 μL of the supernatants was added to the sample wells in addition to the high and low controls included in the kit. The Simple Plex Runner and Explorer softwares were used to setup the assay and analyze the results, respectively. confocal laser scanning microscopy. Each BSCC condition was pooled in a microcentrifuge tube, allowed to settle to the bottom, and media was removed. BSCCs were fixed with 4% paraformaldehyde (v/v) in PBS and permeabilized using a 1% (v/v) Triton X (Invitrogen 28314) solution in tris-buffered saline (TBS) followed by a blocking step using a solution of 5% (v/v) BSA in TBS. BSCCs were stained using Hoechst (Life Technologies) to stain for cell nuclei. For tubulin visualization, a mouse anti-tubulin antibody (Sigma T6199) with an Alexa Fluor 488 goat anti-mouse secondary antibody (Invitrogen A11029) was used. F. nucleatum 25586 was Scientific RepoRtS | (2020) 10:5321 | https://doi.org/10.1038/s41598-020-62139-z www.nature.com/scientificreports www.nature.com/scientificreports/ stained with a rabbit anti-F.nucleatum 25586 polyclonal antibody mix (Diatheva ANT0084) which was detected with an Alexa Fluor 647-conjugated donkey anti-rabbit IgG secondary antibody (Invitrogen A31573). Labeling of BSCCs with F. nucleatum 23726 was also attempted, however, the polyclonal anti-F. nucleatum antibodies did not label this strain well in our hands. CLSM was performed with a Zeiss LSM 800 Axio Observer.Z1/7 using the Zen 2.5 software with the following settings: excitation at 353 nm, emission at 410-470 nm for Hoechst; excitation at 488 nm, emission at 485-565 nm for tubulin; and excitation at 631 nm, emission at 644-700 nm to detect F. nucleatum.
High-throughput Rt-qpcR. BSCCs were initiated as described above with the following treatments: BHI (vehicle), F. nucleatum 23726, F. nucleatum 25586, and HK F. nucleatum 23726. After 24 h of co-culture, groups of four BSCCs were pooled in triplicate, except for HK F. nucleatum 23726, which had duplicate pools. RNA isolation was carried out for each sample set following the RNeasy Mini Kit protocol (Qiagen 74104). DNase (Qiagen 79254) treatment was performed on the purification column following manufacturer's instructions. Isolated RNA was used as template for transcriptional profiling of genes involved in key pathways of cancer using the TaqMan OpenArray Human Cancer Panel (Applied Biosystems 4475391), which consists of 648 TaqMan qPCR assays arranged on an OpenArray plate. The TaqMan ® OpenArray ® Pathway Panels low sample input protocol (10-50 ng/μL) was followed. Gene-specific reverse transcription was performed using SuperScript ™ VILO ™ cDNA Synthesis Kit (Thermo Fisher 11754050) and TaqMan PreAmp primer pools A and B (Thermo Fisher 4485255). Pre-amplification of cDNA was performed by combining cDNA, primer pools, and TaqMan PreAmp mastermix (Thermo Fisher 4391128), and subsequent thermal cycling. Pool A and Pool B for each sample was combined 1:1 and then diluted 1:20 into nuclease-free water. Mixed pre-amplification products were mixed 1:1 with 2x TaqMan OpenArray real-time mastermix (Thermo Fisher 4462164) and loaded onto the OpenArray plate using the OpenArray AccuFill System (Thermo Fisher 4457243). Loaded OpenArray plates were placed in QuantStudio ™ 12 K Flex Real-time PCR system for qPCR thermal cycling. Data was analyzed in ExpressionSuite software, where dCt and relative quantification (compared against vehicle control) were calculated using B2M, UBC, and YWHAZ as endogenous controls. Unpaired t-tests for each treatment vs. control were calculated using GraphPad Prism. Heatmaps were made using Heatmapper 66 and Venn diagrams were made using BioVenn 67 . pathway Analysis. Differentially expressed genes (p < 0.05, unpaired T-test) for each treatment group were uploaded into Qiagen Ingenuity Pathway Analysis (IPA) software for further analysis and interpretation. Default analysis settings were used except for species ("human" was used) and tissues and cell lines ("colon cancer cell lines" was used). Disease/function networks shown are p < 0.0001 (Fisher's exact test) except for BSCC with F. nucleatum 25586 where networks shown are p < 0.0001 and have calculated Z-scores (33 total disease/functions were p < 0.0001 for this treatment). For each sample, four BSCCs in media were pipetted onto a PTFE membrane filter (0.45 μm pore size, 47 mm diameter; EMD Millipore JHWP04700) under vacuum using glass base apparatus (Fisher XX1014702). Media was allowed to pass through filter, and BSCCs remained on top of filter. Filter paper containing cells were placed in a 60 mm diameter petri dish (cells facing down) containing 1.5 mL ice cold extraction solvent (acetonitrile: methanol: water; 40:40:20) and allowed to extract for 20 minutes at −20 °C (placed on top of ice/dry ice). After 20 minutes, extraction solvent was transferred to a clean 1.5 mL microcentrifuge tube, and tubes were centrifuged for 10 minutes at 4 °C at 15,000 × g. Supernatant (1000 µL) was transferred to a clean 1.5 mL microcentrifuge tube, dried under nitrogen, and re-dissolved in 60 µL water:methanol (80:20) for LC/MS analysis.