Secondary Logo

Journal Logo


Combinatorial cell surface display system in Escherichia coli for noninvasive colorectal cancer detection

Hsiao, Chiayia,b,c; Fang, Qingweib,c,d; Zheng, Shuyub,c,d; He, Peixiangb,c; Zhang, Kairanb,c,d; Hong, Yiruib,c; Ran, Yuncongb,c; Sun, Shiyub,c; Liu, Zhixiangb,c; Liang, Zhihanb,c; Zhong, Bozitaob,c; Wu, Weiwenb,c; Xia, Ruoyub,c; Bai, Shihaob,c; Yao, Kuanb,c; Wang, Yushub,c; He, Linb,c; Ma, Ganga,b,c,*

Author Information
doi: 10.1097/JBR.0000000000000052



Colorectal cancer (CRC) has become a predominant cancer, as a result of increasing risk factors such as an aging population and unfavorable modern dietary habits.[1] Although several new treatments have been developed, their effects on cure rates and long-term survival are still limited. One reason for this is that CRC is generally confirmed by physical examination, such as colonoscopy, only when patients exhibit symptoms such as blood in stool, change in bowel habits, and abdominal pain. However, such symptoms do not present until the near end-stage of the disease, thereby complicating efforts to lower mortality and improve cure rates. Thus, there is an urgent need for an early screening method for CRC.

The surface of microorganisms is a fascinating interface that can function in many different ways. Among them, the function of transporting transmembrane proteins to a cognate position has been widely used to transport different peptides onto to the cell surface for further applications. At present, this cell surface-display system has been developed for various applications such as novel reaction processes by enzyme-displaying cells,[2,3] immune system investigation,[4] and live vaccination development.[5] In addition, cell surface-display systems can be used for tumor screening and targeting by expressing certain functional peptides.[6,7] In the field of tumor targeting, various types of surface-displayed peptides have been developed to adhere to different target binding sites on numerous types of cancer.[6–8] However, as a result of the specificity of the intestinal environment, researchers have seldom attempted to target CRC with a cell surface-display system or other detection agents. Disaccharides occur on approximately 90% of human carcinomas.[9] The Thomsen–Friedenreich (TF) antigen (galactose β1–3 N-acetylgalactosamine-α), which is abundant on the mucosa of hyperplastic and neoplastic colon cancer,[10] is often used as a prognostic indicator or marker of metastasized cells because of its involvement in carcinoma cell homotypic aggregation.[9] The possibility of using the TF antigen as a biomarker for CRC has been demonstrated by previous research,[11] which not only showed its detection specificity as a biomarker, but also that the distribution of small detective agents, such as nanobeacons, is quantitatively sufficient for detection.[12]

Delayed diagnosis results not only from inconspicuous symptoms, but also from the obscurity of pathological appearance displayed by endoscopy and other screening techniques. Nanobeacon[12] and other diagnostic methods[13] have been developed, each of which has shown different levels of difficulty when applied in particular environments, such as the intestine. Importantly, although these methods serve as colonoscopy indicators, they require invasive procedures that limit improvements in early diagnosis. In addition, potential safety issues involving remaining foreign substances in vivo still require careful investigation.

Gas vesicles, which are the only subcellular structure in prokaryotes filled with gas,[14] play an important role in various fields of microbiology and have been newly developed as biomedical imaging reporters. Indeed, gas vesicles have been used across a variety of fields, ranging from explaining how certain aquatic microorganisms form surface waterblooms[15] to providing a method for measuring cell turgor pressure.[16–18] Given that the original vesicles produced by Bacillus megaterium cannot be detected by ultrasound, replacement of the gas vesicle protein A (gvpA) gene in the cluster by the homologous gene from cyanobacterium Anabaena flos-aquae can produce high echogenicity.[19,20] This modified cluster was named acoustic reporter gene (ARG, referred to as ARG1) because of its heterologous expression ability and ultrasonic detection. Truncation of the gas vesicle scaffolding protein (GvpC) in ARG1 generates an advanced ARG cluster with a low critical collapse pressure (referred to as ARG2). Gas vesicles with a size of approximately 200 nm[21] not only produce robust ultrasound contrast across a range of frequencies, but also exhibit harmonic scattering that enables enhanced detection against background signal in vivo.[20] Gas vesicles also contain thresholds for pressure-induced collapse, thereby facilitating multiplexed imaging capabilities.[19] Combined with research on the echogenic properties of ARG-expressing vesicles in mammalian hosts,[21] considerable possibilities and prospects for use of gas vesicles as a marker in internal detection could be foreseen.

In this study, we constructed an engineered Escherichia coli (E.coli) strain that displays the specific peptide P30-1 from the phage display library[22] to recognize TF antigen during the early stage of CRC and produce gas vesicles for ultrasonic inspection. We verified the affinity of cell surface-displayed peptides to target the binding sites and adhesive properties of the engineered E.coli using various methods, including flow cytometry, immunohistochemical analysis, and microfluidic chips, on both cell and tissue scales. Ultrasonic examination was also performed to verify the echogenicity of gas vesicles produced and ultrasonic specificity of the strain toward target tissue.

Materials and methods

Chemicals and reagents

FastDigest restriction enzymes and T4 DNA ligase were purchased from Thermo Fisher Scientific (Waltham, MA, USA). KOD-Plus-Neo (DNA polymerase) was purchased from Toyobo (Osaka, Japan). All other chemicals were of analytical grade and commercially available.

Bacterial strains and plasmids

All strains and plasmids used in this study are listed in Additional Table 1, E.coli DH5α was used as the cloning host for DNA manipulation experiments. E.coli BL21(DE3) was used for protein expression and adhesive property examinations. Both DH5α and BL21(DE3) competent cells were purchased from Vazyme Biotech (Nanjing, China). Plasmids pET28a-T7-ARG1 and pET28a-T7-ARG2 were gifts from professor Guoqiang Chen at Tsinghua University (Beijing, China) and professor Fei Sun at Hong Kong University of Science and Technology (Hong Kong, China). All primers used in this study were synthesized by Genewiz (Suzhou, China). Genetic manipulation and construction were performed according to standard protocols.[23]

Bacteria and cell culture

Luria-Bertani (LB) liquid medium (10 g/L tryptone, 5 g/L yeast extract, and 10 g/L NaCl) or agar plates with appropriate antibiotics were used for genetic manipulations, plasmid constructions, and protein expression levels.

Plasmids encoding ice nucleation protein (INP)-P30-1 and enhanced green fluorescent protein (EGFP) were co-transformed into E.coli BL21 (DE3) cells, which were grown in LB medium with 50 μg/mL streptomycin and 50 μg/mL ampicillin for 15 hours at 37°C. Next, positive clones were identified and inoculated into large-scale cultures in LB medium containing 50 μg/mL streptomycin and 50 μg/mL ampicillin at a ratio of 1:10. E.coli BL21(DE3) were grown at 37°C to an optical density at 600 nm (OD600) = 0.5, and then incubated with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) at 37°C for another 5 hours to induce INP-P30-1 expression. Plasmids encoding ARGs were transformed into BL21(DE3) cells as described above, but with 50 μg/mL kanamycin for 15 hours at 37°C. Subsequently, positive clones were treated with 1 mM IPTG at 30°C for another 44 hours to induce protein expression.

Mouse model

For induction of CRC in mice, male C57BL/6 mice (6–8 weeks old with an average weight of 23.3 g, Shanghai SLAC Laboratory Animal Co., Ltd., License No. SCXK (Hu) 2017–0005) were stabilized, weighed, and intraperitoneally administered 10 mg/kg azoxymethane (AOM, Merck, Darmstadt, Germany) solution, a potent carcinogen for inducing CRC. Subsequently, mice were allowed to rest for a week. With the AOM injection completed, dextran sulfate sodium (DSS, Merck), an agent with direct toxic effects on the colonic epithelium, was administered to mice in their drinking water over three cycles lasting a total of 9 weeks to create a chronic inflammatory state. During the first week of the three-week cycle, DSS solution (2.0% to 2.5%) was given as drinking water to subjects. DSS solution was mixed right before it was given to the subject to ensure the best results. After 1 week of consuming DSS, the drinking water for the next 2 weeks was switched to normal clear water. The cycle was repeated twice more while carefully monitoring weight changes in the subject. For control mice, normal saline was used as a placebo along with the induction process. Mouse model induction involved 11 mice treated with the method mentioned above as experimental group, and 3 mice for control. All animal studies were approved by the Institutional Animal Care and Use Committee of Shanghai Jiao Tong University, China (approval No. 201801015) on February 23, 2018 and were in accordance with the National Research Council Guide for Care and Use of Laboratory Animals.

TF antigen quantity examination – cell line suspension

After using trypsin to digest the colorectal adenocarcinoma cell line HT29 (provided by the Fifth People's Hospital of Shanghai) and human epithelium (HACAT, stored in Professor Gang Ma's lab) cells, phosphate-buffered saline (PBS) was used to resuspend cells at a density of 1 × 106 cells/mL. Approximately 1 mL of cell suspension was transferred to a 1.5-mL centrifuge tube, meanwhile 4 μL of fluorescein isothiocyanate-labeled peanut agglutin (PNA-FITC) was added for incubation at 37°C for 30 minutes away from light. Next, samples were evaluated using flow cytometry (FACSCalibur; BD Biosciences, San Jose, CA, USA). FlowJo software (Tree Star, Ashland, OR, USA) was used to gate and separate positive fluorescence events. To obtain accurate cell counts, we isolated living cells and evaluated gated signals on the forward scatter–side scatter density plot to exclude any remaining background.

TF antigen quantity examination – tissue cell suspension

Colorectal tissue was removed from model mice, digested in 0.25% trypsin at 37°C for 15 to 20 minutes, centrifuged for 10 minutes at 1000 × g, and the supernatant was removed. After resuspending samples in PBS, the mixture was filtered through a 5-μm filter and cell density was determined. Next, PNA-FITC (1 μg/mL) was added to samples for incubation at 37°C for 30 minutes, followed by centrifugation for 10 minutes at 1000 × g and resuspension in PBS; this process was repeated three times. Finally, samples were evaluated using flow cytometry and analyzed by FlowJo software.

TF antigen quantity examination – tissue sample

Rectal or colon sections were longitudinally cut open to form a rectangular piece of tissue, which was placed on cardboard and soaked in 4% paraformaldehyde (PFA) overnight. Afterwards, the soaking solution was switched to 30% sucrose solution for incubation overnight. Tissue was subsequently embedded in Optical Cutting Temperature Compound (OCT; 25608-930; Tissue-Tek, Torrance, CA, USA) and immediately frozen on dry ice. Mounted tissues were sectioned using a cryostat (CM3050S; Leica, Wetzlar, Germany) at −20°C to −23°C according to standard procedures.[24] Ten-micron thick sections were placed on positively charged slides (Labserv 32550, Thermo Scientific, Waltham, MA, USA) and baked on a hot plate at 37°C for 8 minutes.

To examine TF antigen abundance, sections were washed with PBS containing 0.1% Tween-20 (PBST) three times (10 minutes each) and incubated in blocking buffer (PBS containing 5% FBS and 0.1% Triton X-100) for 1 hour. After removing the blocking buffer, samples were washed again three times with PBST. Finally, 50 μg/mL PNA-FITC was added to samples for incubation at 4°C overnight. Before observing under a fluorescence microscope (Axio Imager M2, Zeiss, Oberkochen, Germany), samples were washed with PBST again three times. Finally, samples were sealed with mounting medium containing 4′,6-diamidino-2-phenylindole. For pathological examination, fixed tissue sections were stained with hematoxylin and eosin (H&E) and observed under an optical microscope (Leica).

Bacterial adhesion property examination – chamber slide

After HT29 cells reached 70% confluence, bacteria (INP-P30-1 + EGFP and EGFP control strains) were diluted to 10% of the original density, 150 μL/OD600 of which was added to chamber slides for incubation at 37°C in a shaker for 30 minutes. The solution in each well was removed, and wells were washed with 0.5 mL of normal saline three times to eliminate unbound bacteria. Next, chamber slides were treated with 4% PFA for 15 minutes, washed with PBS three times, and sealed with mounting medium. Finally, slides were observed under a fluorescence microscope to examine the affinity of engineered bacteria.

Bacterial adhesion property examination – cell line suspension

HT29 cells were digested, resuspended in PBS, and separated into nine centrifuge tubes (four for the experimental group, four for the control group, and one for the negative control group), each containing 150 μL of cell suspension. On the other hand, INP-P30-1 + EGFP E.coli and EGFP E.coli, both of the which express fluorescent proteins, were incubated to OD600 = 0.5. Approximately 150 μL of each strain was added to four parallel centrifuge tubes (INP-P30-1 + EGFP E.coli for the experimental group, EGFP E.coli for the control group), along with a PBS negative control. All samples were incubated on a shaker at 37°C for 30 minutes in the dark. The 10 samples (9 previously described plus a positive control with bacteria only) were treated with 3 μL of 7-aminoactinomycin D (7-AAD) dye. Finally, samples were evaluated by flow cytometry. Microfluidic chip-based assay was further used to detect the affinity of INP-P30-1 + EGFP strain on HT29 cells and the detail design is presented in Additional Figure 1,

Bacterial adhesion property examination – tissue sample

Colon or rectal tissue was longitudinally cut open and sequentially soaked in 4% PFA and sucrose solution (40%) overnight. Concurrently, engineered bacteria were incubated and induced until a final OD600 = 1, and evident GFP expression was confirmed under a fluorescence microscope. Next, tissue sections were separately placed into one of the two kinds of bacterial media for incubation at 37°C for 30 minutes. After incubation, tissues were gently washed three times with PBS, embedded in OCT, and frozen on dry ice to prepare sections, as described above. Frozen tissue was sliced into 10-μm thick sections and observed under a fluorescence microscope (Axio Imager M2) to evaluate bacterial aggregation.

Ultrasound imaging in vitro

Six-well plates were chosen as the container for imaging. Each well was filled with three layers of agarose gel (Additional Fig. 2, The bottom layer was 0.2% (w/v) agarose gel to avoid ultrasound signal interference from the underside of the plate. Cells at 2 × the final concentration were mixed at a 1:1 ratio with molten 0.4% agarose (w/v, at 50°C) and immediately loaded into the well to form the middle layer. The top layer was 0.2% (w/v) agarose gel to make imaging of the middle layer more convenient, and served as blank control. Increasing agarose concentrations of up to 0.5% were also acceptable, although a slight weakness in collapse results was observed. To avoid rapid attenuation of ultrasonic waves in air, we used PBS or double-distilled H2O to cover the imaging samples, thereby allowing the transducer to dip into the solution. Unnecessary bubbles must not be produced during preparation.

For imaging, a VEVO LAZR-X imaging system and MX250 transducer (Fujifilm VisualSonics, Tokyo, Japan) were used in B mode with VA Phantom application setting. Specific parameters were as follows: 21-MHz frequency, 2% (min, 1 minute) power for imaging, 100% (max, 5 minutes) power for collapse, a maximum of 50 frames, and 16-dB gain. Focus was adjusted to the depth of the layer where engineered E.coli were present. The field of view was adjusted to zoom in on the region of interest and achieve effective noise repression. Image processing code is available on GitHub (

According to Eqs. (1) and (2), when the transmitting power was increased from 2% to 100%, the sound pressure was estimated to increase by sevenfold, as follows:

where I is sound intensity, P is sound power, S is the area of the region of interest whose plane is perpendicular to the direction of sound wave propagation, pm is maximum sound pressure, and Z is the specific acoustic impedance of the medium. A sevenfold change in sound pressure during the collapse process provided an operational margin for future work to more precisely determine the threshold condition of ARG collapse and exclude interference associated with background and noise signals.

Ultrasound imaging ex vivo

Under the protocol approved by the Institutional Animal Care and Use Committee of Shanghai Jiao Tong University, AOM/DSS-induced and control mice were sacrificed for harvesting of colon tissues under aseptic operation.

For ex vivo ultrasound imaging of ARG1 E.coli, mixtures of E.coli (ARG1 or control) and 0.4% molten agarose were made at a ratio of 1:1. 1 mL of the mixture was applied to PBS-washed normal colon tissue using insulin needles. Subsequently, colons were fixed with 0.5% agarose phantom for ultrasound imaging. No randomization or blinding was necessary in this study. Images from transverse and longitudinal views were captured according to the condition described above for ultrasound imaging.

For ex vivo ultrasound imaging of INP-P30-1 + ARG2 E.coli (Additional Fig. 3,, INP-P30-1 + ARG2 E.coli was injected into both CRC and control colons. After incubating for 30 minutes at 37°C, colons were washed with PBS 3 times to eliminate unspecific binding, and filled with 0.2% agarose gel to stabilize. Colons were then fixed in 0.5% agarose phantom for ultrasound imaging. No randomization or blinding was necessary in this study.

Magnetic resonance imaging (MRI) imaging in vitro

Bacterial cultures were diluted to different concentrations with double-distilled H2O, mixed at a 1:1 ratio with molten 1% agarose (w/v, at 50°C), and immediately added to a 2-mL centrifuge tube (Additional Fig. 4, MRI imaging was performed on a 3 Tesla uMR790 platform (United Imaging Healthcare, Shanghai, China). A multi-slice multi-echo sequence was used to acquire images. Specific parameters were as follows: 3-T magnetic field strength, 180° flip angle, 3000-ms repetition time, echo train length of 16, 20-ms echo spacing, 20-ms first echo time, 5-mm slice thickness, average of 1, and 180-pixel bandwidth.

Statistical analysis

To test statistical significance, we used two-sided heteroscedastic t-tests with a significance level of type I error set at P = 0.05 for rejecting the null hypothesis. Sample sizes for all experiments, including animal experiments, were based on preliminary experiments to be adequate for statistical analysis. All statistical analyses were performed using R software (


Construction of peptide cell surface-display system in E.coli

As a transmembrane protein, INP[15,16] was chosen as our vector protein to translocate P30-1 peptide, which can recognize TF antigen on the outer cell membrane. After IPTG induction, the engineered E.coli strain INP-P30-1 co-expressed the corresponding INP-peptide fusion protein and EGFP reporter (INP-P30-1 + EGFP strain), allowing further characterization. Bacteria only expressing EGFP were used as a control group.

Verification of TF antigen abundance in HT29 and CRC tissue

To validate the specificity of TF antigen in CRC cells, we first measured amounts of TF antigen in HT29 cells by flow cytometry. HT29 (Fig. 1A) and normal HACAT (Fig. 1B) cells were suspended and incubated with the fluorescent antibody PNA-FITC, which specifically recognizes TF antigen. The results show that the fluorescent antibody exhibited a significantly increased tendency to bind HT29 cells (71.93 ± 0.68% positive rate) compared with HACAT cells (0.135 ± 0.135%, P < 0.0001, Fig. 1C and D). This result indicated a higher abundance of TF antigen in HT29 cells compared with normal epithelial cells. TF antigen quantity was further examined in primary cells of mouse colorectal tissue by flow cytometry (Fig. 2A). Similar to cell line experiments, FITC levels of diseased tissue were much higher than the control group (Fig. 2B), with a PNA-FITC positive ratio of 29.3 ± 6.2% in colorectal adenocarcinomas and 1.03 ± 0.36% in the control group (Fig. 2B and D). These results indicate a high amount of TF antigen was present on the cell surface of primary CRC tissues. Moreover, TF antigen levels indicated by FITC intensity did not display a monotonic pattern during tumor development.

Figure 1
Figure 1:
Thomsen–Friedenreich (TF) antigen quantification in the colorectal adenocarcinoma cell line HT29 and human epithelium (HACAT) cells by flow cytometry. (A) Flow cytometry analysis of HACAT cells co-incubated with fluorescein isothiocyanate-labeled peanut agglutin (PNA-FITC) before quantification and gated out from forward scatter (FSC)-side scatter (SSC) plots when most free Escherichia coli was removed. Next, we used 7-aminoactinomycin D (7-AAD) dye to evaluate cell viability. Extremely few positive signals were obtained for HACAT samples. (B) Flow cytometry analysis of HT29 cell samples. Compared with HACAT cells in A, a larger quantity of FITC-positive cells were gated out in HT29 cells. (C) Merge of plots shown in A and B. The histogram attached to the plot indicates signal distribution according to PNA-FITC signal intensity. (D) Statistical results of comparison between HACAT and HT29 cells. Results from each group were derived from five biological replicates. Error bars indicate standard deviation, ****P < 0.0001 (two-sided heteroscedastic t-test).
Figure 2
Figure 2:
Quantification of Thomsen-Friedenreich (TF) antigens in colorectal cancer (CRC) colon by flow cytometry and immunohistochemistry. (A) Flow cytometry analysis of CRC and control colon primary cells. Fluorescein isothiocyanate-labeled peanut agglutin (PNA-FITC) signal positively represented the TF antigen level. A shift of cluster (red circle vs blue circle) was observed between CRC and control groups. (B) Comparison of percentages of PNA-FITC-positive cells in control and CRC groups. (C) Decreased TF antigen expression during tumor progression. Ratios of PNA-FITC-positive cells to whole live cells was calculated based on flow cytometry results. (D) TF antigen detection in mice colon tissue sections by IHC. Representative images (left) and magnified inset images (right). (E) Numerical comparison of PNA-FITC-positive cells in control and CRC tissue groups. Data, expressed as the mean ± SD, were analyzed by two-sided heteroscedastic t-test. *P < 0.05, ***P < 0.001. (F) Increased PNA-FITC-positive signal on tumor sites compared with off-tumor sites in CRC tissue. DAPI = 4′,6-diamidino-2-phenylindole, , FSC-H = forward scatter height.

Although TF antigen levels increased during the early stages, they rapidly declined during the later stages of our mouse model (Fig. 2C). This finding is consistent with a previous report examining altered TF antigen levels during colon cancer progression,[25] although a conflicting report also exists.[26] The notion that TF antigen is more extensively produced in primary tumors compared with metastatic tumors is supported by another study that compared TF antigen levels between primary and metastatic mouse CRC cell lines.[27] Decreased TF antigen during the advanced stages of tumor development involves hypersialylation,[27] since diminished differences in PNA-binding capacity could be observed when CRC cell line was treated with sialidase. Taken together, dynamic alterations of TF antigen levels in our model agreed with our expectations; although, the metastatic potential of different stages in our CRC model have yet to be determined.

Additionally, PNA-FITC was applied to mouse colorectal tissue specimens. The results not only demonstrated that diseased tissues contained more TF antigen than healthy tissues (Fig. 2D and E), but also showed distinguished differences in amounts of antigen between on- and off-tumor sites (Fig. 2F). Collectively, our results demonstrated the specificity of TF antigen in diseased tissues and it potential as a target binding site for colorectal tumor diagnostics. Indeed, examinations of TF antigen abundance and specificity verified our qualification of the TF antigen as a CRC target-binding site.

Peptide cell surface-display system identification in HT29 cells and CRC tissues

Having verified the abundance of TF antigen (target binding sites) in HT29 cells and CRC tissues, we next evaluated our engineered peptide cell-display system (INP-P30-1) using immunocytochemistry of HT29 cells (chamber slide and cell suspension) and mouse model specimens. For chamber slide experiments, we first tested sets of incubation temperatures and times for mixing the INP-P30-1 + EGFP strain with HT29 cells. Optimal conditions were determined to be 37°C for 30 minutes (Fig. 3A). Maintaining this condition was extremely crucial because incubation for longer than 30 minutes increased nonspecific adhesion. Subsequently, INP-P30-1 + EGFP and EGFP strains were simultaneously characterized under the same conditions. The results showed that under conditions with roughly equivalent cell amounts, both INP-P30-1 strains outnumbered the control group in terms of increased adhesion to HT29 cells (Fig. 3B and C).

Figure 3
Figure 3:
Characterization of ice nucleation protein P30-1 plus enhanced green fluorescent protein Escherichia coli (INP-P30-1 + EGFP E.coli) adherence capacity to chamber slides. (A) Procedure for preparing chamber slides. (B) Observation of binding between E.coli and the colorectal adenocarcinoma cell line HT29 under a fluorescence microscope. Green dots represent E.coli adhered to HT29 cells. Scale bars: 50 μm. (C) Number of EGFP dots observed on slides of INP-P30-1 + EGFP and EGFP groups. Data represent mean values from three biological replicates, while error bars indicate standard deviation. Data were analyzed by two-sided heteroscedastic t-test. DAPI = 4′,6-diamidino-2-phenylindole, EGFP = enhanced green fluorescent protein, INP = ice nucleation protein, PFA = paraformaldehyde.

For the cell suspension assay, HT29 cell suspensions were incubated with the INP-P30-1 + EGFP strain for specified amounts of time. Dead cells were discriminated from living cells by treatment with 7-AAD dye to decrease noise (Fig. 4A). To identify positive signal indicating successful and precise adherence to HT29 cells, we screened out free bacteria before examining fluorescence intensities. Signal that showed a similar scale for cells and strong fluorescence intensity was concluded as positive after excluding free bacteria (Fig. 4B–D). The results showed an average positive rate of 46.4% for the INP-P30-1 + EGFP strain, which was significantly increased compared with the control group (11.9%) on the basis of four parallel repeats (Fig. 4E). In addition to flow cytometry of chamber slides and cell suspensions, the affinity of INP-P30-1 + EGFP for HT29 cells was confirmed by a microfluidic chip-based assay (Additional Fig. 1,, thereby providing a new application for this device in examining the adhesive properties of particles on a microscale.

Figure 4
Figure 4:
Characterization of ice nucleation protein P30-1 plus enhanced green fluorescent protein Escherichia coli (INP-P30-1 + EGFP E.coli) adherence capacity toward colorectal adenocarcinoma cell line (HT29) suspensions by flow cytometry. (A) Sample preparation process. HT29 cells were co-incubated with INP-P30-1 + EGFP E.coli and then quantified by flow cytometry. (B) Flow cytometry analysis of HT29 cells binding with INP-P30-1 + EGFP E.coli. (C) Flow cytometry analysis of HT29 cells co-incubated with EGFP E.coli. (D) Comparison between INP-P30-1 + EGFP and control groups. The histogram attached to the merged plot indicates signal distribution according to green fluorescent protein (EGFP) signal intensity. (E) Statistical results indicate EGFP-positive rates of 29.1% for INP-P30-1 + EGFP E.coli and 2.81% for the control, each group has 3 biological replicates. Data, expressed as the mean ± SD, were analyzed by two-sided heteroscedastic t-test. ***P < 0.001. FSC-H = forward scatter height.

Supported by cogent cell-based experimental results, we next examined our peptide-display system in mouse model specimens, as shown in Figure 5A. The results indicated that the on-tumor sites of CRC tissue, which exhibited abnormal hyperplasia in H&E-stained slices and high signals for TF antigen (Fig. 5B), bound increased amounts of the INP-P-30-1 + EGFP strain (47 ± 3%) compared with off-tumor sites (11.5 ± 1.5%, P < 0.05; Fig. 5C). Comparison of numbers of bacteria observed in CRC and normal colon issue (control) also revealed an increased INP-P30-1 + EGFP E.coli population in the CRC apical colon, thereby illustrating the high-binding specificity between P30-1 peptide and TF antigen (Fig. 5D). Certain signals (highlighted in yellow triangles) observed in the basolateral side of the colon were considered to be background because bacteria were loaded onto the intact colon to prevent in-depth invasion of bacteria into the colon (Fig. 5D). Therefore, the apical side of the colon was considered as the region of interest to calculate numbers of attached E.coli (Fig. 5D). Although the rough surface of cancerous tissue slightly increased affinity in the control group, adhesive specificity remained considerably high in the INP-P30-1 + EGFP group (Fig. 5E). Overall numbers of E.coli were significantly higher in INP-P-30-1-treated CRC compared with INP-P-30-1-treated normal colon tissue (P < 0.0001). Meanwhile, the E.coli population was higher in INP-P-30-1-treated CRC compared with similar EGFP-treated tissues (P < 0.01), suggesting an important role for INP-P30-1 in differentiating binding ability. In summary, the efficiency of our cell surface-display system for TF antigen targeting was certified at cellular and mouse tissue levels, wherein high specificity and affinity of the INP-P30-1 + EGFP strain toward target binding sites were proven.

Figure 5
Figure 5:
Verification of adherence capacity of ice nucleation protein (INP)-P30-1 plus enhanced green fluorescent protein Escherichia coli (INP-P30-1 + EGFP E.coli) toward colorectal cancer (CRC) tissue. (A) Scheme of applying bacteria to dissected tissue. (B) A fourfold increase in bound INP-P30-1 + EGFP E.coli on tumor compared with off-tumor tissues. Correspondingly, pathological morphology was confirmed by hematoxylin and eosin staining. (C) Statistical description of INP-P30-1 + EGFP E.coli binding difference between on- and off-tumor sites. (D) A significant increase in bound E.coli was observed when comparing INP-P30-1 + EGFP E.coli-treated CRC slices with EGFP E.coli-treated CRC slices, and INP-P30-1 + EGFP E.coli-treated control slices with EGFP E.coli-treated control slices. White arrows indicate attached E.coli. Yellow triangles indicate background signal. White-dotted lines indicate regions of interest. (E) Statistical description and comparison of four groups mentioned in D. Cases 01 and 02 are biological repeats with the same treatment. Data, expressed as the mean ± SD, were analyzed by two-sided heteroscedastic t-test. *P < 0.05, **P < 0.01, ****P < 0.0001. EGFP = enhanced green fluorescent protein, OCT = optical cutting temperature.

Noninvasive detection of INP-P30-1 + ARG strain through medical imaging

For ultrasound molecular imaging of CRC, we devised INP-P30-1 + ARG1 and INP-P30-1 + ARG2 strains combining the P30-1 surface-display system and two ARG types as a TF antigen-targeting ultrasound molecular probe. The detectability of ARG strains was first examined by B-mode ultrasound imaging both in vitro and ex vivo. Results of in vitro experiments (for experiment setup, see Additional Fig. 2, exhibited a significant ultrasonic signal enhancement (approximately fourfold higher) of ARG1 and ARG2 E.coli compared with those in LB broth and normal E.coli control groups (Fig. 6A and B), indicating that ARG E.coli can be effectively distinguished from normal E.coli in vitro by signal strength. In addition, ex vivo experiments showed that the engineered E.coli produced a stable contrast compared with controls (Fig. 6C and D), thereby further verifying the detectability of ARG strains in colon ultrasound imaging.

Figure 6
Figure 6:
Ultrasound imaging of ARG Escherichia coli (E.coli) in vitro and ex vivo. (A) In vitro ultrasound imaging of ARG1, ARG2, Luria-Bertani (LB) broth control, and blank E.coli control. A parallel ultrasound beam was focused on the middle layer where samples were located. (B) Gray values of the four ultrasound images shown in A. (C) Ex vivo ultrasound imaging of ARG1 E.coli and control E.coli in normal colon from both transverse and longitudinal views. The colon was stabilized in 0.5% agarose gel. (D) Gray values of the region of interest in the four images shown in C. (E) Ultrasound imaging of ARG2 E.coli with different treatments. Left: intact; middle: collapse for 5 minutes; right: collapse for 10 minutes. (F) Gray values of ultrasound images shown in E. Data, expressed as the mean ± SD, were analyzed by two-sided heteroscedastic t-test. *P < 0.05. ****P < 0.0001. ARG = acoustic reporter gene.

To improve the ultrasonic detection sensitivity of our engineered bacteria, we first explored the commonly adopted microbubble “destruction-replenishment”[28] or collapse[21] strategy in vitro. Theoretically, we can remove unspecific signals of the background and circulating bubbles by subtracting the post-destruction ultrasound image from the pre-destruction image, thereby leaving only signals of the molecularly targeted microbubbles for inference.[28] The in vitro gas vesicle collapse experiment was performed deliberately, and the results shows that when ultrasound pressure was increased up to sevenfold the basic pressure, signal strength of the ARG2 strain decreased to approximately 30%, thus confirming the practicability of the collapse strategy to improve ARG2 detection sensitivity in vitro (Fig. 6E and F).

In addition to ultrasound imaging, whole-cell E.coli-based ARG contrast can potentially be noninvasively detected using MRI. Quantitative T2 mapping showed that along with increased cell density, the T2 value of ARG2 E.coli decreased more rapidly than EGFP E.coli in vitro (Fig. 7). This result indicated that the proposed whole-cell-based imaging probe has promising applications as a dual-modality imaging probe for combined ultrasound and MRI.

Figure 7
Figure 7:
Magnetic resonance imaging of ARG2 Escherichia coli (E.coli) in vitro. (A) T2 mapping images of ARG2 and EGFP at a series of cell densities. Original cell densities of 3.43 × 108 and 4.38 × 108 cells/mL for ARG2 and EGFP, respectively, were diluted to 0.1 × , 0.2 × , 0.3 × , 0.4 × , and 0.5 × original cell densities for multi-echo spin echo MRI. Quantitative T2 mapping images were calculated from multi-echo spin echo images with a time of echo ranging from 20 ms to 320 ms. (B) Mean T2 value of ARG2 and EGFP at various cell densities. Mean T2 value was calculated for each of three biological replicates. Error bars indicate standard deviation of the three T2 mean values. ARG = acoustic reporter gene, EGFP = enhanced green fluorescent protein.


Early screening is of great importance in inhibiting CRC progress. The earlier the disease is diagnosed, the more quickly proper treatment can be exerted. For the past few years, early screening of CRC has been promoted and performed worldwide. However, given the huge imbalance in levels of medical care worldwide, as well as the expense and potential pain associated with existing screening methods, early diagnosis of colon cancer remains difficult. Application of synthetic biology techniques has shown considerable potential and novel perspectives for the diagnosis and therapy of numerous severe diseases. We attempted to improve current early diagnostic strategies for potential patients with CRC by developing an engineered, ultrasonic detectable TF antigen-targeting system in E.coli.

On the basis of successful ultrasound imaging and gas vesicle collapse of ARG strains, we validated the affinity of the TF antigen-targeting INP-P30-1 + ARG2 strain for CRC using ultrasonic detection ex vivo. However, as the result of noise produced during the imaging process and error caused by slight motions of the colon, our experiment was hindered by a non-negligible degree of false positives (Additional Fig. 3,, indicating that potential clinical applications of the INP-P30-1 + ARG strain require additional supporting evidence. Using the collapse strategy in clinical situations can be difficult. First, the width of gas vesicles has an inverse relationship with their critical collapse pressure.[14,29,30] Given that ARG gas vesicles are on a nanoscale and, thus, much smaller than clinically used microscale contrast bubbles, they may require additional harsh collapse conditions. Second, although Bourdeau et al[21] calibrated the collapse pressure of ARG2 and ARG1 to 2.7 and 4.7 MPa, respectively (Halo and Ana gas vesicles only require 650 kPa for destruction[19]), methods to perform ARG gas vesicle collapse with general imaging platforms remain unclear.

A functional peptide-display system carried by bacteria provides new possibilities for early CRC diagnosis and treatment. Researchers use biomarkers, such as TF antigen, to decrease the delay of diagnosis. Nanobeacon and other diagnostic methods have been developed, but each has shown boundaries when used in special environments, such as intestinal conditions. However, by constructing a cell surface-display system and transformed it into bacteria, intestinal conditions no longer inhibited its use, but instead provided a precisely accurate environment for it to work. Given that hundreds of microorganism species comprise the microbial community in the intestinal ecology, the existence of an engineered microorganism is considerably safer in the intestine than in other in vivo environments, such as blood, where microorganisms trigger an immune response. Moreover, as gene editing is easy to perform in bacteria, modifying equipped bacteria to possess certain useful properties has considerable possibilities. The biosafety of our device can also be further ensured by introducing a light- or chemical-triggered suicide pathway in bacteria to prevent leakage or harm to the environment or gut microbiota. As an essential quality for antigen recognition, binding specificity also indicates the practicability of such an agent for various applications. Comparison of adhesive properties between on- and off-tumor sites clearly demonstrated the high-binding specificity of the engineered strain to target foci (Fig. 6C). In addition to ultrasonic imaging, various possible applications for cell surface-display systems, such as precise drug synthesis and delivery to disease foci, are possible.

Compared with traditional ultrasound and MRI molecular contrast probes with typical targeting designs, linking, and signaling components, the proposed whole-cell-based imaging probe exhibited several advantages. First, the whole-cell-based probe design provides high integrity and overcomes the shortcomings of traditional chemical conjugation. Given that chemically conjugating targeting and signaling components with a linker is generally difficult, our surface-display system design provides an easier solution to attach the targeting peptide to the probe with a simple standard plasmid transformation protocol. Second, with cell membrane protection, the ARG reporter gas vesicle signal will increase in stability. Whole-body micro-positron emission tomography imaging studies have shown that the half-life of traditional molecularly-targeted microbubbles in blood was only 3.5 minutes, and 95% of microbubbles were cleared from the blood stream within 30 minutes.[31] The short life of traditional microbubble-based ultrasound imaging probes limits their applications for long-term observation scenarios, such as navigating surgeries to excise tumors, which generally lasts for several hours. In contrast, cells harboring ARG gas vesicles produced detectable signal that was stable for at least 14 hours in vitro.[21] Given that many probiotic E.coli naturally inhabit the colon, engineered E.coli will live long enough in intestinal conditions for consecutive detection. Third, the manufacture of the proposed probe was efficient due to the microbe's enormous self-replication capacity, which will undoubtedly help reduce the high cost of translating molecular imaging probes for clinical use as it is one of the main hindrances for their development. In future work, nonlinear ultrasound imaging methods harnessing harmonic signals, a more elaborate collapse protocol, image post-processing techniques that include image registration to counteract the effect of colon motion, and an ultrasound-MRI united imaging protocol will be further explored to enhance the signal-to-noise ratio and reduce false positive rates. We will also consider cascading repetitive peptide sequences to enhance the binding ability of the surface-display system for the probe.

In conclusion, by verifying the echogenicity of gas vesicles in mammalian subjects while the vesicles remained inside the microorganism host, this study revealed another way of using gas vesicles for internal detection. In addition to the modification of vesicle nanostructures for additional function,[20] creating a bacterial host that not only presents echogenic properties of gas vesicles, but retains considerable genetic flexibility for additional functional modification. By combining our cell surface-display system with ARG, we developed a promising method of detecting CRC and reduced the difficulty of early diagnosis for this disease. Moreover, such characteristics are not the only modifications that can be generated in the host strain. According to specific needs for disease foci localization and noninvasive detection, various engineering techniques can be performed to widen or enhance strain function.


We thank Prof. Chan Ding from Shanghai Veterinary Research Institute and Prof. Shijun Bao from Gansu Agricultural University for providing INP plasmid, Prof. Guoqiang Chen from Tsinghua University and professor Fei Sun in Hong Kong University of Science and Technology for providing plasmids containing ARG genes, Prof. Xiang Chen and Chao Han from Department of Micro/Nano Electronics at Shanghai Jiao Tong University for micro-fluid chip experiment, Dr. Xia Zhang from School of Life Sciences and Biotechnology at Shanghai Jiao Tong University for providing EGFP plasmid, Dr. Qian Yu from Bio-X Institutes at Shanghai Jiao Tong University for guiding CRC induction process, Prof. Yiping Du from Institute of Medical Imaging Technology at Shanghai Jiao Tong University for the support of MRI, Prof. Huajun She and Yufei Zhang for discussions on MRI experiments, Dr. Gengming Niu from the Fifth People's Hospital of Shanghai, Fudan University for providing the HT29 cell line, and Prof. Zhiyong Li from School of Life Sciences and Biotechnology at Shanghai Jiao Tong University for comments about writing.

Author contributions

GM and KZ conceived the study. GM, CH, QF, and SZ composed the manuscript. PH, ZLiu, ZLiang and YH performed the cloning and transformation experiments. YR, SS, RX and SB conducted cell experiments. CH and SZ performed mouse experiments. QF performed ultrasound-related experiments and MRI. YR, SZ, WW, BZ and KY prepared the graphs. LH, YW and GM guided the project and supported the experimental designs and materials. All authors read and provided critical input for the manuscript.

Financial support

This work was supported by the National Natural Science Foundation of China (No. 31671504, 31970775, and 19Z103150074) and the Cross Research Fund of Biomedical Engineering of Shanghai Jiao Tong University, China (No. YG2016MS04).

Institutional review board statement

All animal studies were approved by the Institutional Animal Care and Use Committee of Shanghai Jiao Tong University, China (approval No. 201801015) on February 23, 2018.

Conflicts of interest

The authors declare that they have no conflict of interest.


1. Fedirko V, Tramacere I, Bagnardi V, et al. Alcohol drinking and colorectal cancer risk: an overall and dose-response meta-analysis of published studies. Ann Oncol 2011; 22:1958–1972.
2. Shigechi H, Koh J, Fujita Y, et al. Direct production of ethanol from raw corn starch via fermentation by use of a novel surface-engineered yeast strain codisplaying glucoamylase and alpha-amylase. Appl Environ Microbiol 2004; 70:5037–5040.
3. Fujita Y, Ito J, Ueda M, et al. Synergistic saccharification, and direct fermentation to ethanol, of amorphous cellulose by use of an engineered yeast strain codisplaying three types of cellulolytic enzyme. Appl Environ Microbiol 2004; 70:1207–1212.
4. Norton PM, Wells JM, Brown HW, et al. Protection against tetanus toxin in mice nasally immunized with recombinant Lactococcus lactis expressing tetanus toxin fragment C. Vaccine 1997; 15:616–619.
5. Ribeiro LA, Azevedo V, Le Loir Y, et al. Production and targeting of the Brucella abortus antigen L7/L12 in Lactococcus lactis: a first step towards food-grade live vaccines against brucellosis. Appl Environ Microbiol 2002; 68:910–916.
6. Piñero-Lambea C, Bodelón G, Fernández-Periáñez R, et al. Programming controlled adhesion of E. coli to target surfaces, cells, and tumors with synthetic adhesins. ACS Synth Biol 2015; 4:463–473.
7. Arap MA, Lahdenranta J, Mintz PJ, et al. Cell surface expression of the stress response chaperone GRP78 enables tumor targeting by circulating ligands. Cancer Cell 2004; 6:275–284.
8. Bereta M, Hayhurst A, Gajda M, et al. Improving tumor targeting and therapeutic potential of Salmonella VNP20009 by displaying cell surface CEA-specific antibodies. Vaccine 2007; 25:4183–4192.
9. Landon LA, Peletskaya EN, Glinsky VV, et al. Combinatorial evolution of high-affinity peptides that bind to the Thomsen-Friedenreich carcinoma antigen. J Protein Chem 2003; 22:193–204.
10. Rosenberg WM, Prince C, Kaklamanis L, et al. Increased expression of CD44v6 and CD44v3 in ulcerative colitis but not colonic Crohn's disease. Lancet 1995; 345:1205–1209.
11. Carter JH, Deddens JA, Pullman JL, et al. Validation of the galactose oxidase-Schiff's reagent sequence for early detection and prognosis in human colorectal adenocarcinoma. Clin Cancer Res 1997; 3:1479–1489.
12. Kumagai H, Pham W, Kataoka M, et al. Multifunctional nanobeacon for imaging Thomsen-Friedenreich antigen-associated colorectal cancer. Int J Cancer 2013; 132:2107–2117.
13. Inadomi JM. Screening for colorectal neoplasia. N Engl J Med 2017; 376:149–156.
14. Walsby AE. Gas vesicles. Microbiol Rev 1994; 58:94–144.
15. Konopka A. Metalimnetic cyanobacteria in hard-water lakes: Buoyancy regulation and physiological state. Limnol Oceanogr 1989; 34:1174–1184.
16. Walsby AE, Fogg GE. The pressure relationships of gas vacuoles. Proc R Soc Lond B Biol Sci 1971; 178:301–326.
17. Walsby AE, Fogg GE. Water relations of gas-vacuolate prokaryotes. Proc R Soc Lond B Biol Sci 1980; 208:73–102.
18. Pinette M, Koch A. Daneo-Moore PA, Higgins LM, et al. Biophysics of ampicillin action on a gas-vacuolated gram-negative rod. American Society for Microbiology, Antibiotic inhibition of bacterial cell surface assembly and function: American Society for Microbiology.. Washington, DC: 1988.
19. Shapiro MG, Goodwill PW, Neogy A, et al. Biogenic gas nanostructures as ultrasonic molecular reporters. Nat Nanotechnol 2014; 9:311–316.
20. Lakshmanan A, Farhadi A, Nety SP, et al. Molecular engineering of acoustic protein nanostructures. ACS Nano 2016; 10:7314–7322.
21. Bourdeau RW, Lee-Gosselin A, Lakshmanan A, et al. Acoustic reporter genes for noninvasive imaging of microorganisms in mammalian hosts. Nature 2018; 553:86–90.
22. Liu R, Li X, Xiao W, et al. Tumor-targeting peptides from combinatorial libraries. Adv Drug Deliv Rev 2017; 110-111:13–37.
23. Sambrook J, Fritsch EF, Maniatis T. Molecular Cloning: A Laboratory Manual. Cold Spring Harbor: Cold Spring Harbor Laboratory Press; 1989.
24. Abed J, Emgård JEM, Zamir G, et al. Fap2 mediates fusobacterium nucleatum colorectal adenocarcinoma enrichment by binding to tumor-expressed Gal-GalNAc. Cell Host Microbe 2016; 20:215–225.
25. Bresalier RS, Ho SB, Schoeppner HL, et al. Enhanced sialylation of mucin-associated carbohydrate structures in human colon cancer metastasis. Gastroenterology 1996; 110:1354–1367.
26. Cao Y, Karsten UR, Liebrich W, et al. Expression of Thomsen-Friedenreich-related antigens in primary and metastatic colorectal carcinomas. A reevaluation. Cancer 1995; 76:1700–1708.
27. Nemoto-Sasaki Y, Mitsuki M, Morimoto-Tomita M, et al. Correlation between the sialylation of cell surface Thomsen-Friedenreich antigen and the metastatic potential of colon carcinoma cells in a mouse model. Glycoconj J 2001; 18:895–906.
28. Abou-Elkacem L, Bachawal SV, Willmann JK. Ultrasound molecular imaging: Moving toward clinical translation. Eur J Radiol 2015; 84:1685–1693.
29. Hayes PK, Walsby AE. The inverse correlation between width and strength of gas vesicles in cyanobacteria. Br Phycol J 1986; 21:191–197.
30. Walsby AE, Bleything A. The dimensions of cyanobacterial gas vesicles in relation to their efficiency in providing buoyancy and withstanding pressure. Microbiology 1988; 134:2635–2645.
31. Willmann JK, Cheng Z, Davis C, et al. Targeted microbubbles for imaging tumor angiogenesis: assessment of whole-body biodistribution with dynamic micro-PET in mice. Radiology 2008; 249:212–219.

colorectal cancer; cell-surface display; ice nucleation protein; ultrasonic detection

Supplemental Digital Content

Copyright © 2019 The Chinese Medical Association. Published by Wolters Kluwer Health, Inc.