Colourectal consists of four pillars: Interaction, Detection, Signalling and Biosafety. These together enable early in vivo detection of colorectal cancer. On this page, an overview is given of all the results we accomplished to create Colourectal, our in vivo diagnostic tool for colorectal cancer. In-depth reports of the separate pillars can be found by clicking on the tabs at the top of the page. Additionally, for a more substantial overview of all the scientific highlights, go to our proof of concept page!

The first pillar of Colourectal, our living diagnostic, is the binding of potential cancer cells. This allows Colourectal to get close to the cancer and stay there, giving it a higher chance of detecting our biomarkers . Some bacteria have natural adhesins, Type 1 fimbriae, which are structural pili. These fimbriae, in Escherichia coli, are able to bind to α-D-mannose groups in glycoproteins on host cells [1]. Carcinoembryonic antigen cell adhesion molecules (CEACAMs) are one such glycoprotein. This glycoprotein is normally expressed in healthy colon tissue [2]. However, in early colonic adenomas and hyperplastic polyps, which can both grow into cancer, CEACAM6 expression is increased [2]. This makes it a good target for early detection of cancerous growths. Type 1 fimbriae are expressed by the fim genes and are constructed by assembling different subunits, FimA, C, D, F, G and H, where FimH is the binding domain at the end of the pilus [3]. A potential issue arises due to the fact that CEACAM6 is also expressed in healthy tissue, making the binding less specific. To increase specificity to colon cancer while keeping the normal binding capacity of FimH, we intended to fuse FimH with the nine amino acids long peptide CPIEDRPMC (RPM ), which binds specifically to integrin α5β1 in invasive colon cancer [4]. S ince this fusion protein may not work or fold properly, we opted to fuse at both the C and N termini, behind the signal peptide sequence, and test both options, resulting in parts (BBa_K4244065) and (BBa_K4244067). The 2015 iGEM team of Harvard made a fusion part with RPM on the N terminus (BBa_K1850011) , however they added both a His- and SpyTag and therefore we decided to design our own constructs.

Implementation of these fusions were done on SEVA plasmid 258. This plasmid contains XylS-Pm regulator promoter system originally from Pseudomonas putida. The promoter works with a titratable inducer m-toluate. This allows us to test different expression levels based on different inducer concentrations. By exploring different concentrations, we can identify expression levels of the fused protein that will replace standard FimH in the assembly of the fimbriae without being a high burden on the cell. To ensure that potential effects on binding capacity are not only due to the overexpression of fimH, we also created a control plasmid with the native fimH only (BBa_K4244069) . Furthermore, the plasmid contains a broad-host-range replication origin, RFS1010 and its replication proteins RepA, B and C and kanamycin resistance gene.

Bacterial cells of E. coli Nissle 1917 (EcN) were be tested on binding ability, using three different binding assays. Here, they would be combined with the immobilized proteins CEACAM6 and integrin α5β1. Further, they would also be combined with immobilized Caco-2 cells.

Cells containing either C- or N-fusion to FimH should have a higher binding capability on integrin α5β1 compared to wild-type EcN if RPM can reach its target. Hopefully, they are still able to bind to CEACAM6. In addition, the wild type of EcN would also be used as a control to determine normal adhesion levels and would be compared to wild-type E. coli K-12.

Unfortunately, we were not able to construct the fusion and the native fimH plasmids yet. Therefore, we were unable to perform the binding assays in time for the wiki freeze. In the upcoming weeks, we will continue to troubleshoot the integration of the C- and N-fusion in addition to the FimH overexpression circuit.

Since we designed our EcN to attach to colorectal cancer cells, it is important to investigate the interaction between our living diagnostic and the cancer cells. A commonly used model for colorectal cancer molecular studies is provided by the Cancer coli-2 (Caco-2) cells, that were originally isolated from a (human) colorectal adenocarcinoma [5].

Several members of the E. coli family are known to possess the pks genomic island that encodes for colibactin, a toxin that can cause a genotoxic effect in eukaryotic cells [6]. Colibactin is activated during direct contact between the living bacteria and mammalian cells [7]. However, there is conflicting evidence on whether EcN has this genotoxic potential [8]-[ 10]. Fortunately, the toxic effect can be shut down, while keeping the probiotic properties that EcN has, by substituting an amino acid in the clbP gene, which is responsible for the colibactin activation [9]. Alternatively, the whole clbP gene could be knocked out. In the future we could consider using one of the two mutants to silence the colibactin gene. During this project we only used the wild-type EcN, and its heat inactivated form, to assess the interaction between EcN and Caco-2 cells as genotoxicity is a slow process and it will most likely not affect our experiments.

The potential toxic effect that EcN might have on the cancer cells was assessed with a cytotoxicity assay . This assay allowed us to determine the effect of potential cytotoxic compounds produced by EcN on the proliferation of the Caco-2 cells. From this experiment, as shown in Figure 1, it appeared there was no relation between the Caco-2 cell proliferation and the concentration of EcN. This means that an increased EcN dose does not affect the metabolic activity of human cells.

The increase that is seen between 24 and 48 hours indicates a change in metabolic activity in the human cells. This does not necessarily mean that the cancer cells are growing, but it could suggest an increase in cellular processes. To obtain a conclusion on the change in physiological response over time we should perform a cytotoxicity assay on healthy, non-cancerous cells.

Additionally, one of our chromoproteins (anm2CP) was also assessed for its potential toxic effect on human cells. We incubated the human cells with different concentrations of chromoprotein, ranging from 0.01 to 2 mg/ml for 24 hours. The chromoproteins did not affect the Caco-2 cells at concentrations of 0.1 mg/ml or lower, as shown in Figure 2. In contrast, from chromoprotein concentrations of 0.5 mg/ml and higher the proliferation decreased significantly.

This experiment gives an indication of how chromoproteins affect the human cell proliferation. However, it is not a good representation for our real-life application. These results need to be confirmed in a more realistic model for the human colon. The modelled concentration could ultimately be used to validate if this is enough to colour the stool.

One of the CRC biomarkers we are using in this project is Matrix metalloproteinase 9 (MMP-9). Therefore, it is important to investigate whether EcN is affecting the expression of MMP-9. Fábrega et al. (2017) found that EcN can decrease the MMP-9 expression in colonic mice [11] We incubated the Caco-2 cells with heat inactivated EcN for t = 0, 4, 8, 12, and 24 hours and collected the supernatant from these cells, as MMP-9 is a secreted protein. The Western Blot in Figure 3 shows a clear band around 65 kDa. The expected size was 92 kDa for the inactive protein and 84 kDa for the active MMP-9 [12] However, MMP-9 can be converted in vivo to a smaller active form of 64 kDa [13]. From this and from the fact that the blot is very clean it is probable that the band around 64 kDa represents MMP-9. In future experiments, a positive control, such as purified MMP-9, could be included.

Over time, the cells secrete more MMP-9. The blots were stained with Ponceau staining to ensure that the samples were loaded with an equal concentration of proteins. When comparing the untreated cells with the Caco-2 cells that were exposed to heat inactivated bacteria at t = 12 and t = 24 hours, we can clearly see that the bands show the same intensity. This suggests that the secretion of MMP-9 is not affected by the presence of EcN.

To investigate the effect EcN had on the expression profile of the colorectal cancer cells we used bulk RNA sequencing followed by differential expression analysis. We compared Caco-2 cells in normal conditions with Caco-2 cells that have been in contact with heat inactivated EcN. Several time points, t = 4, 8, 12, 24 hours were compared to assess the differential expression over time. In total 16518 genes were analysed, from which almost 600 genes showed altered temporal expression. We used a regression strategy to identify genes with significant temporal expression via maSigPro. The genes were divided into six different clusters , shown in Figure 4A. However, only cluster three was associated with a specific biological function, namely cell division (Figure 4B). The increased expression of genes related to the cell cycle might be correlated to the increase in proliferation we observed in Figure 1.

Additionally, the colorectal cancer cells could also affect the expression profile of EcN. Therefore, we performed transcriptomics on EcN that had been in contact with Caco-2 cells. This allowed us to see if EcN presents genes that respond differently to the presence of Caco-2 cells. EcN was incubated on Caco-2 cells for 30 and 120 minutes and the expression profiles were compared to an untreated control. We could use this data to evaluate activated genes and potentially identify promoters that are affected by colorectal cancer. In the future, we could use this information to improve colorectal sensing. Due to the turnover time of the RNA sequencing we were unfortunately not able to obtain the results before the wiki freeze.

Our current FimH constructs were sent for sequencing but do not align with expected results. To understand what went wrong, the whole plasmids will be sent for sequencing.

From the cytotoxicity assay it appeared that the Caco-2 cells’ proliferation is not affected by a change in concentration of the heat inactivated EcN. However, there might be a time-dependent proliferative effect on the Caco-2 cells. The chromoproteins showed no toxic effect at a concentration up to 0.1 mg/ml. The next step would be to use a representative model for the colon to determine the cytotoxic effect of the different factors inside the human body. The effect of EcN on the MMP-9 secretion was analysed by a Western Blot, and no effect was found. Except for the change in expression in cell cycle related genes, the RNA sequencing results showed no significant effect on the gene expression of the Caco-2 cells after exposure to the heat inactivated bacteria.

To make our living diagnostic capable of detecting cancer, we first must understand some of the underlying concepts of cancer cells. Cancer is a disease where cells proliferate uncontrollably due to changes in signalling and metabolism [1]. Because of this, cancer cells produce more L-lactate than healthy cells [2]. This increase in lactate production is also found in colorectal cancer (CRC) cells. The microenvironment of healthy colon cells contains between 1.5 – 3 mM of L-lactate, while tumorous colon microenvironments contain between 10 – 25 mM of L-lactate [3]. By sensing lactate with our living diagnostic, we want to make it sensitive to CRC.

Some gut bacteria, such as Escherichia coli (E. coli), can use lactate as a substrate for growth. In E. coli, the operon that is responsible for this is called the lldPRD operon. This operon contains a transcription factor, LldR, which can bind to two different operator regions called O1 and O2. These operator regions are located up- and downstream of the operon’s promoter, PlldPRD, see Figure 1A. By binding to these regions, LldR represses the operon, see Figure 1B. However, if lactate is bound to LldR, it changes conformation, resulting in activation of gene expression, see Figure 1C [4]. This system has already been used in previous research, including previous iGEM competitions, to develop lactate biosensors [3,5,6] [7–9].

By employing this same strategy, we decided to develop a synthetic biosensor in Escherichia coli Nissle 1917 (EcN) with the aim to sense L-lactate concentrations found in the microenvironment of colon tumours. In its current state, LldR and PlldPRD are unfit for this purpose due to two reasons. The first is that the promoter is repressed by glucose, which is found in the colon [10,11]. And secondly, the lldPRD operon is regulated by a system native to E. coli which regulates the expression of several operons under aerobic and anaerobic conditions [12]. To address these challenges, Zúñiga et al. developed a lactate sensitive promoter called ALPaGA (A Lactate Promoter Operating in Glucose and Anoxia) which is not repressed by the above-mentioned conditions [5], making it suitable for sensing cancer in the colon environment.

At first, we constructed a lactate-sensing genetic circuit. In this circuit a reporter gene, the super folder green fluorescent protein (sfGFP) (BBa_I746916), was placed after the lactate inducible promoter ALPaGA (BBa_K4244000), while LldR (BBa_K822000) was constitutively expressed. A genetic construct like this should only show fluorescence when lactate is present. An overview of the circuit can be found in Figure 2.

The circuit from Figure 2 was tested by transforming it into EcN. The strain was grown for 16 hours in microoxic conditions, in minimal medium (M9) containing glucose and induced with different concentrations of L-lactate. After 16 hours, the fluorescence was measured and plotted against the L-lactate concentration, see Figure 3. The goal of this experiment was to see if we could use this genetic circuit to differentiate healthy levels from cancerous levels of L-lactate, highlighted as the green and red vertical bars in Figure 3, respectively. In its current form, the genetic circuit is sensitive to L-lactate since there is an increase in fluorescence with increased L-lactate levels. However, the system is currently unable to distinguish healthy from cancerous levels of L-lactate, because in both cases significant fluorescence is observed. For this reason, we set out to find a way to change the operational range of the dose-response by moving it to the right. By doing so, we want to eliminate the promoter responding to healthy concentrations of lactate found in the colon.

There are many tools to regulate gene expression in bacteria, of which two examples are Clustered Regularly Interspaced Palindromic Repeats interference (CRISPRi) and antisense RNA (asRNA) [13–15]. We hypothesized that we could modify the operational range of the dose response of ALPaGA by combining these two tools. To test this hypothesis, we designed a CRISPRi-asRNA genetic circuit. This circuit is organized as follow, a constitutively expressed CRISPRi cassette represses sfGFP. The cassette consists of a deactivated CRISPR associated (dCas) protein and a single-guide RNA (sgRNA) complementary to sfGFP. Additionally, an asRNA complementary to the sgRNA would be made inducible by lactate. This can be obtained by placing the ALPaGA promoter in front of the asRNA and constitutively expressing LldR. Designing a genetic circuit in this way means that the reporter gene is repressed if lactate is absent but expressed when lactate is present, see figure 4 for an overview. The initial hypothesis was that by changing the binding affinity between the asRNA-sgRNA complex, the activation efficiency would change, resulting in the operational range shifting to the left or to the right.

Unfortunately, due to time constraints, we were unable to test the CRISPRi-asRNA genetic circuit in the lab before the iGEM Grand Jamboree deadline. However, we were able to model this construct in silico, the details of this can be found in the page Lactate Model. The model predicted that only the sensitivity, meaning the slope of the dose response curve, would change when adding the CRISPRi-asRNA genetic circuit. The added CRISPRi and asRNA genetic components were not predicted to affect the operational range of the dose response curve, see Figure 5 for the modelled dose response curves. Because of this, we set out to find different ways to change the operational range of the dose response curve.

One of the things we learned from the Lactate Model is that we could shift the operational range of the dose response by changing $k\_f_{LldRcomplex}$. This parameter describes the rate of activation of LldR. We hypothesized that, in biological terms, this rate is influenced by the intracellular L-lactate concentration. This means that, if we could control the internal concentration of L-lactate, we could control the operational range of the dose response. We wanted to test this hypothesis by generating a strain that cannot metabolise L-lactate. From literature we found that by knocking out lldD and ykgF we should be able to generate an E. coli strain that cannot catabolise L-lactate [16]. Furthermore, we found that EcN specifically contains an additional lactate dehydrogenase, nldH, which we also planned on knocking out [17]. Even though the knockouts were successfully generated before the iGEM deadline, there was not enough time to test our hypothesis [16].

Upon detection of colorectal cancer, our living diagnostic will secrete a coloured protein from the chromoprotein family, colouring the user’s stool. Chromoproteins (CP) are a class of small, coloured proteins that come in a variety of shades and are responsible for colouration in corals or sea anemones (Uppsala_Chromoproteins) . Unlike fluorescent proteins (FP), CPs absorb light to give off colour in ambient light, allowing for instrument-free detection. This feature makes them very useful as reporter proteins in non-laboratory conditions [1].

CPs are encoded by a single gene, making them relatively easy to clone and engineer. Thanks to the work of the iGEM 2011 Uppsala team, a collection of chromoproteins optimized for Escherichia coli, was made available to use. In 2017, the SHSBNU China team mixed different CPs with curry to imitate stool samples. Based on their results, pink and blue chromoproteins produced the most significant change in stool colouration. This, combined with the fact that these colours are the least likely to be found in food, made us opt for blue and pink CPs. The blue CP will be used as the indicator of colorectal cancer, whereas the pink one will be used as a positive control to assess if the living diagnostic is still active once ingested. In order to avoid confusion with blood stains in the stool, secretion of the pink CP will be controlled by the user through an inducible system. Thus, multimeric proteins amilCP (BBa_K592009) and spisPink (BBa_K1033932) , were selected as good candidates for our project. In addition, two monomeric CPs were also tested: the pink anm2CP (BBa_K2387001) and blue Ultramarine (UM) (BBa_K4244001) [2]. (Refer to figure/clip of monomeric and multimeric (amilCP and UM))

The first step of our project consisted of producing the CPs, as can be seen in Figure 1, out of the four CPs tested only spisPink did not exhibit any colouration. We decided to continue with only anm2CP for the pink variant and keep UM and amilCP for the blue variant.

CPs have not been observed to be naturally secreted [3,4] and no research has been found that focuses on the CP secretion in E. coli. We assume that CPs generally do not spontaneously diffuse outside the bacterial cell, therefore a secretion system needs to be integrated into our design. By adding a signal peptide (SP) sequence to the C- or N-termini of a protein, the protein can be secreted via its respective system [5]. In E. coli, type I and II secretory pathways are the most prevalent for recombinant protein translocation out of the six reported pathways for Gram-negative bacteria [6]. Type I secretion entails a single-step transfer of specific proteins to the extracellular space through both inner and outer membranes with the sequence remaining attached to the protein [7,8]. On the other hand, type II secretion entails two-step secretion mechanisms where proteins are directed to the periplasmic region where they accumulate before entering the pseudopilus apparatus (Figure 2, orange complex) that transports cargo across the outer membrane [9,10]. The general secretory (Sec) or twin-arginine translocation (TAT) cellular machinery enables type II translocation in E. coli [11]. The Sec pathway export unfolded protein whereas the TAT system exports fully folded proteins into the periplasmic region [12] in both systems the SP sequence gets cleaved off the protein after translocation.

Another system that can be used for the secretion of CPs is the curli secretion pathway (Type VIII) [13]. This pathway is natively used in E. coli to secrete protein units of curli fibres (CsgA) which self-assemble into extracellular curli fibres during biofilm formation [14,15]. CsgA is secreted into the periplasm with the help of the SEC peptide and across the outer membrane with the help of the N22 peptide [16]. These two components are essential for the secretion of proteins through this pathway, and previous research has shown that they can be harnessed for the secretion of heterologous proteins too [17].

The signal peptides tested in our project are shown in Table 1. We decided to test the signal peptide for secretion using a medium strength constitutive rubisco promoter, rbcL (BBa_K4244057) with its terminator (BBa_K4244058) and super folder green fluorescent proteins (sfGFP) (BBa_K3143689). rbcL was chosen to favour secretion in E. coli avoiding high levels of expression. In addition, chloroplast promoters are eubacterial in origin and hence compatible with gene expression in bacteria [18]. sfGFP is more suitable for secretion assays as it matures faster than CPs, and the fluorescence can be quantified more easily. Even though, this will gives us an indication that secretion is possible in our strain the experiment will have to be repeated with chromoproteins as the amino acid sequence and tertiary structure affect the secretion efficiency of the signal peptide[10].

However, as observed in Figure 3, no fluorescence was detected in the supernatant over 8 hours of growth, signalling that secretion of sfGFP was not achieved with any of the constructs under a constitutive promoter. The experiment was re-performed using a rhamnose inducible promoter as we hypothesized that the high expression level caused by the constitutive promoter caused inclusion bodies [19] clogging transporter. The plasmid were transformed in Escherichia coli Nissle 1917 (EcN) ΔrhaB ΔrhaT to allow for better titration [20]. An endpoint measurement was done after 5 hours of growth induced with rhamnose (500 μM) (Figure 4), as we observed from Figure 3 that protein production reached a plateau after 5 hours. Different rhamnose concentrations (0, 25, 50, 100, 250 μM) were tested leading to the same conclusion that no fluorescence was detected in the supernatant (data not shown).

A clear observation is that less fluorescence is emitted by the fused sfGFP than by GFP without SP. This comes from the SP affecting the growth of EcN. Regarding secretion activity, the results are in contrast with previous literature [21–23]. One reason could come from the type II secretion system that translocates the protein in the periplasmic space where they accumulate and are not secreted into the extracellular medium [21]. One way to solve that would be to knockout lpp a protein that helps with the outer membrane sturdiness. This can render the outer membrane permeable enough to allow extracellular protein production as demonstrated by Shin & Chen, 2008 where they secreted twice the amount of active xylanase compared to the wild type [24]. Regarding the type I system, Linton et al. (2012) demonstrated uvGFP translocation using HlyA by performing a Western blot of concentrated supernatant from a 100 ml induced culture [21]. Our expression system is only different by the inducible promoter so our system might not be sensitive enough to detect this little amount. Another hypothesis might come from the amino acid sequence not being to fold properly in the extracellular space. A Western blot on these fractions could help us determine if sfGFP is indeed in the supernatant but an inactive state as the extracellular space might not be appropriate for proper sfGFP folding. However, due to time, we were not able to perform this experiment. Additionally, the same secretion experiment could be performed using a different strain of E. coli. Even though the signal peptides, except PelB, tested are native to E. coli species the sequences were not amplified from the genome of EcN but taken from the iGEM registry. This could play a role in the non-secretion of sfGFP as it was not directly amplified from our strain. Therefore, introducing these constructs in another strain (BL21 or K-12) could help us determine if the secretion is strain specific or not. Alternatively, we could also amplify a known signal peptide from EcN and fuse it to our gene of interest. Finally, additional screening can also be performed with other SP if none of the solutions above gives positive results.

While the chromoproteins will be secreted when high lactate levels are detected, using a second cancer biomarker will ensure there is less risk of false positives. We have coupled the colour of our chromoproteins to interaction with a second cancer biomarker. For our living diagnostic, this second marker is Matrix Metalloprotease 9 (MMP-9). MMP-9 is a protease which has been shown to play a vital role in cancer cell invasion and is one of the most widely investigated MMPs [25]. MMP-9 is an extracellular protein that can degrade extracellular matrix proteins through proteolytic cleavage. This protease recognises and cleaves a specific amino acid sequence. The consensus sequence of MMP-9 is Pro-X-X-Hy-(Ser/Thr), with X being any amino acid and Hy being a hydrophobic amino acid [26].

We set out to make use of this cleaving property by engineering our chromoproteins in such a way that they are only coloured upon cleavage by MMP-9. For this design, we took inspiration from the already reported flipGFP system [27]. This reporter system is based off a splitGFP variant, in which sfGFP is divided into two proteins that need to interact with each other to restore the beta-barrel structure of GFP and thus fluorescence. In the flipGFP variant, GFP was split in two parts: β-sheets 1-9 and β-sheets 10-11. The latter two sheets are usually anti-parallel, but in the flipGFP system the addition of dimerising linkers E5 and K5 forces them in a parallel configuration, disrupting fluorescence. In the original flipGFP system, cleavage of a specific protease could abolish dimerisation of E5 and K5, thus restoring the anti-parallel configuration of β-10-11 and fluorescence of the FP, as demonstrated in Figure 5.

We sought to mimic this system in our chromoproteins, as chromoproteins share a similar structure with the green fluorescent protein (GFP) with a distinctive β-barrel composed of eleven β-sheets surrounding an α-helix core that restrains the chromophore group [28] as can be seen in Figure 6. Our MMP-9-sensitive flipCPs were designed alongside an MMP-9-sensitive flipGFP. In our design, the K5 domain was preceded by cleavage sequence specific for either MMP-9 or the TEV protease, which serves as a positive control [29].

To get an indication of whether we could expect similar disruption for the flipCP to flipGFP we performed in silico predictive modelling using Alphafold [30]. The predicted structure for flipGFP is shown in Figure 7, whereas the results for the flipCPs can be found at their respective pages ( flipUM , flip-anm2CP , flip-spisPink , flip-amilCP ). Based on the results of our predictive modelling, we could discern the optimal insertion sites for the E5 and K5 regions with respect to the β-sheets. Our predictive modelling results showed us that the CPs β-barrel structure could indeed be disrupted, and that upon removal of the K5 region normal chromoprotein structure would be restored. Moreover, it helped us determining the optimal insertion sites for the E5 and K5 helices. This served as an important starting point for designing and building the actual flipCPs.

Based on the predicted CPs structures, construction was started for the four chromoprotein variants alongside GFP. However, initial tests with the four different CPs encountered difficulties due to inconsistent expression. E. coli DH5a was initially chosen as test platform, but the low CP yields led to waiting times of days before colour could be discerned. This is consistent with previous reports showing that consistent expression of CPs can prove challenging [3]. Therefore, we decided to switch to the protein production strain E. coli BL21 (DE3) as our new strain for testing the CP constructs.

Due to time investment in learning how to reliably work with CPs, we decided to create a flipGFP variant which was sensitive to MMP-9, and to focus on only one of the four CPs to engineer our flipCFP. Since anm2CP and Ultramarine produced clear colour and Ultramarine was used before in creation of a splitUltramarine [28], we selected Ultramarine as our CP of choice. For this, splitGFP and splitUltramarine were first generated by inserting a terminator-promoter-RBS block between the 9th and 10th β-sheet ( BBa_K4244043 , BBa_K4244044 ) of both proteins. Subsequently, the E5 insert was added to see if the colour would be retained in the absence of the K5 region. Unfortunately, no fluorescence or blue colour was observed for the modified proteins. This was unexpected especially for the splitGFP construct, as splitGFP is a well-established construct in literature. Further testing is required to discern whether the problem lies in the protein modification or the protein expression.

Since Colourectal involves the administration of a genetically modified organism to humans, it is important to build in biosafety and biocontainment mechanisms. A multi-layered strategy encompassing different genetic safeguards makes of our living diagnostic a safe tool for humans and the environment. In the first place, Escherichia coli Nissle 1917 (EcN) should not be able to colonise parts of the body other than the colon, as this would represent an important health concern. To this end, our living diagnostic will be engineered for surviving only in the presence of mucin, a glycoprotein found in the lining of the colon. Second, our living diagnostic should not be able to survive outside of the human body to avoid its spread into the environment. To achieve this, a kill switch, which is activated at temperatures below that of the human body, has been implemented. In the third place, another inducible kill switch has been integrated, this time triggered by the fourth pill of the Colourectal self-test, which is administered to the patient in case of an adverse response, or once the diagnosis process finishes.

In order to mitigate risks associated to the spread of the living diagnostic from the colon to other parts of the body, a mucin auxotrophy has been implemented. Our molecular system is based on a chimeric two-component system (TCS) activated by mucin, the main family of proteins found in the mucus. Mucins are continuously produced by the cells lining the gastrointestinal tract [1,2] and therefore, always present in the colon. The TCS controls the expression of the guanylate kinase gene gmk, essential for cell for its role in nucleotide metabolism [3]. When mucin is present (permissive conditions), the essential gene is expressed allowing cell survival. By contrary, in absence of mucin (non-permissive conditions), gmk is not expressed, leading to cell death.

The TCS consists of a chimeric sensor formed by Dismed2, the extracellular domain of RetS, which is in turn a sensor protein from Pseudomonas aeruginosa known to sense mucin [4], and the transmitter domain of EnvZ, part of the popular E. coli-derived two-component system EnvZ-OmpR [5,6]. The EnvZ-OmpR TCS functions in such a way that EnvZ, when activated, phosphorylates the transcription factor OmpR, which subsequently activates the OmpC promoter. As indicated, in our system OmpC regulates the transcription of the essential gene gmk.

The chimeric sensor, named Dismed2-EnvZ was constructed by cloning the two domains under the control of the constitutive promoter (BBa_J23100) and strong RBS (BBa_J34801). The construct (BBa_K4244009) is inserted in a medium copy number plasmid to avoid any possible toxic effect in the host. After successful cloning of Dismed2-EnvZ as well as the OmpC promoter with the reporter gene (GFP) (BBa_K4244011), these constructs have been transformed into an E. coli from the Keio collection lacking the native gene for envZ [7] to test the mucin dependency. The results of our plate reader experiments with a fluorescent reporter show that increasing mucin concentrations result in high activation levels of gene expression (Figure 2).

The results also suggest that the chimeric receptor shows leakiness, a natural phenomenon given that the EnvZ-OmpC is a native osmotic pressure responsive system in E. coli Nissle 1917. This is shown by the levels of fluorescence in the cells only containing the OmpC-GFP (control -), and the cells containing the complete chimeric TCS but without the presence of the ligand (mucin). Therefore, the main future goal of our system would be the reduction of this leakiness. As a potential solution, we propose the deletion of the two other components of the EnvZ-OmpR system: ompF and ompC. Adding to this, next step would be the cloning gmk under OmpC and knock-out this gene from EcN genome to perform cell death assay with following conditions: mucin and no mucin.

When our living diagnostic leaves the user’s body through the faeces, it could potentially spread into the wild representing a risk for the environment. To prevent this, a temperature-dependent kill switch has been implemented as biocontainment strategy allowing cell survival only at the permissive condition of 37 °C, the body temperature.

Two temperature kill switches with temperature-dependent tunable promoters, which had previously been successfully demonstrated in literature [8–10] were tested in EcN as part of our project. In the work of Piraner and colleagues, the temperature kill switch was constructed using the transcriptional auto-repressor TIpA (BBa_K2500004) from the virulence plasmid of Salmonella typhimurium [8]. This protein contains a coiled C-terminal domain that undergoes uncoiling between 37 °C and 45 °C and an N-terminal DNA-binding domain that acquires a dimeric state with lower temperatures and blocks transcription. If tIpA and its promoter are upstream the antitoxin of a toxin-antitoxin system (Ba_K4244019) (figure 3, left), they will cause a decrease of antitoxin expression when temperature drops, leading to the microorganism’s death [10]. The second kill switch that was tested, was obtained from the work of Stirling and colleagues, and contains the temperature-sensitive regulatory region of the cold shock protein A (PcspA) [9]. The pcspA (BBa_K4244021) is a constitutive promoter that has a high rate of transcription but contains a long 5′ UTR of 159 bp that, at 37 °C, acquires an unstable secondary structure and leads to its degradation. On the other hand, at lower temperatures it forms a stable configuration, which allows translation. In this case, a toxin (ccdB) under the cspA promoter (Figure 3, right) is expressed but counteracted by the antitoxin ccdA (BBa_K4244023), which in turn is expressed by being placed after the constitutive LacUV5 (BBa_M36801) [10]. As indicated, at lower temperature the cspA stabilizes and ccdB overexpresses ccdA, which leads to cell death.

Both systems containing the temperature-sensitive kill switches have been tested with a reporter gene in EcN. This was done to test if at different temperatures the results in our microorganism matched those of literature: increase of fluorescence with decreasing temperature for the pTlpA system and decrease of fluorescence with decrease of temperature in the pcspA system. Since both systems showed the correct behaviour with the fluorescent reporter, this gene was replaced with either a toxin or an antitoxin according to the two systems’ rationales in order to perform temperature sensitive survival assays. On the one hand, the pTlpA system did not show the expected results in our host even after our troubleshooting thanks to our partnership with NanoBuddy, due to growth could also be observed under non-permissive conditions. On the other hand, the pcspA system showed a dramatic decrease of CFU at temperatures that were lower than 37 °C as shown in Figure 4.

Since the pcspA system was shown to work both with a reporter gene and with the toxin-antitoxin kill switch, the next step would be its integration in the EcN genome. This would allow both to have an antibiotic resistance-free system that is not susceptible to plasmid loss, and to decrease the risk of emergence of mutants that could survive also under non-permissive conditions. Having the construct in the chromosome of EcN increases the genetic stability of the kill switch, decreasing the possibility of escapers.

What if our living diagnostic causes any discomfort to the person who ingests it? What if colorectal cancer has already been detected and it is no longer necessary to report its presence through coloured stool? And how do we make sure that any trace of the Colourectal microorganism abandons the user’s body after the self-test process?

The third biosafety mechanism of our project has been installed to address all these questions by means of another inducible kill switch with two clear states: ON in the presence of inducer, and OFF in its absence. Three components were identified as the most important for the performance of such a kill switch: an inducible promoter/regulator system, an effective killing mechanism, and a way to ensure zero basal levels of killing under the permissive conditions.

A great variety of inducible expression systems driven by different molecules can be found both in the literature and in the iGEM repository (e.g., pBad/araC, BBa_I0500). However, since we aimed to introduce this kill switch circuit in our living diagnostic, which will perform its function in the human colon environment, the search for a suitable system became tricky. The inducer molecule of choice could not be part of the colon, be produced by tumour cells, or be a food component. Moreover, it could not be toxic to humans, the environment, or the colon microbiome.

All these requirements drastically reduced the number of possibilities, but the inducible expression system NahR/Psal, activated by acetylsalicylic acid (aspirin), initially seemed to be a possible solution. Aspirin is not present in food or in the colon, and it is considered safe as it is used for medical purposes. Since it is generally administered as a painkiller, we thought it would be possible to use it as the inducer molecule for the kill switch and advice users of Colourectal to avoid intake of aspirin during the self-test process. The part for this inducible expression system (BBa_J61051) was obtained from the iGEM plates to start experimentally testing its activity. However, gastroenterologist Markus Gwingger strongly advised us to not use aspirin as our inducing molecule, since aspirin is often taken as blood thinner by our target group.

Therefore, our design needed to be changed and a different inducible expression system had to be tested. Because we developed a three-level rhamnose inducible system in our project, we decided to include this kill switch in the third step of this induction system. As a proof of concept, we used a rhamnose-inducible system to test the killing mechanism.

The method of choice to cause cell death was DNA degradation, controlled by our inducible expression system and elicited by a type-IC Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) device, which belongs to the Class I CRISPR-Cas systems. Unlike the more frequently used Class II systems, that only consist of 1 effector protein (e.g., Cas9 and Cas12a), the effector complex of Class I encompasses a CRISPR RNA (crRNA) and a number of different Cas proteins, and commonly receives the name of Cascade. In type-IC, the Cascade complex has 3 subunits – Cas5, Cas7 and Cas8 – that are responsible for recognising the target DNA and recruiting Cas3, an enzyme with single-strand DNA helicase-nuclease activity able to cleave and degrade target DNA in a processive way (Figure 5) [11,12]. For the Cascade complex to recognise the target sequence, a short 2-4 nucleotide sequence known as Protospacer Adjacent Motif (PAM) must be found upstream the sequence. In the case of Type-IC CRISPR-Cas, the most commonly used PAM sequence is 5'-TTC-3' [11].

Type-I CRISPR-Cas systems have proven to be very effective mechanisms to achieve cell death in previous studies. For instance, an article by Caliando and Voigt (2014) demonstrates the utilisation of a highly effective type-IE system for inducible cell killing. However, we decided to use the type-IC CRISPR system from P. aeruginosa since, although its mechanism is not that well-known, its mechanism is very similar and has a more compact size (3 proteins instead of 5 in the Cascade complex), which would reduce the burden in the cell [11]. The advantage of this system compared to the more frequently used Cas9 or Cas12a is that the helicase-nuclease activity of Cas3 allows for large genomic deletions, that are assumed to be harder for the cell to repair than just double stranded breaks, and hence more deadly [12].

A good spacer targeting EcN’s genome had to be designed to achieve a system with a high killing efficiency. With that purpose, we analysed different features in EcN’s genome and discovered the presence of Repetitive Extragenic Palindromic (REP) elements. REP elements are very conserved sequences of around 35-40 bp long that are found in around 500 positions of the E. coli genome, always in intergenic regions. Their role in the cell is not well known, but they are believed to be involved in translation regulation [14]. With that information, we designed a single spacer to target multiple points in the genome simultaneously (5’-TTGCCGGATGCGGCGTAAACGCCTTATCCGGCCT-3’), expecting to achieve a highly genotoxic response upon induction. The REP element we used as a target was repeated 20 times in the EcN’s genome with a 100 % accuracy, and in three of those occasions it was preceded by the necessary PAM sequence (TTC), suggesting that our spacer would target the genome at least at three different loci.

A plasmid, referred as pCas3cRh, containing the CRISPR array with the designed spacer and the cas genes (cas3, cas6, cas8 and cas7) under the inducible expression system RhaRS/PrhaB was employed to test the kill switch efficiency in EcN. We performed this test in M9 medium with 50 mM of glucose and different rhamnose concentrations (0, 0.1, 0.5, 1, 3 and 5 mM) (Figure 6).

As it can be observed, our inducer managed to hinder cell growth in a gradual manner when applied at different concentrations. By inducing our kill switch with high concentrations of rhamnose (3 and 5 mM), we managed to significantly impair the growth of EcN, which was effectively inhibited for 15 h. The emergence of escapers with the ability to circumvent the CRISPR-Cas mechanism [12] most likely resulted in the growth observed after that period of time.

In our project there are three systems that should be induced by an external metabolite. However, inducers that are safe for human consumption, not present in the colon and that are compatible with Escherichia coli Nissle 1917 (EcN) are limited. Therefore, we have built a pipeline to control three separate genetic circuits at three different levels of the same inducer metabolite, the “three-step inducible system”. Here, we created the genetic circuit to enable this, and we tested several inducers on the suitability for this system.

The first step in this system is the introduction of a positive control. To see whether our living diagnostic colonised the colon, a positive control is needed. In our project we already made use of colouring the stool blue. Therefore, we introduced a genetic circuit where we can induce the production of a pink chromoprotein to see if the bacteria are present during the diagnostic test. Further, we created an inducible kill-switch that is based on a Clustered Regularly Interspaced Palindromic Repeats (CRISPR)-associated (Cas) system. This circuit will be used to make sure that none of our modified bacteria are able colonise the colon after the test is done and to create an emergency kill-switch to stop the test if needed. This circuit was introduced on the third step of the three-step inducible system, because this should only be triggered after the living diagnostic completed the test. Lastly, our initial idea contained a localisation circuit, a system that made use of the fact that the living diagnostic cells accumulate around tumour cells. These cells would be made visible on a scan, by inducing the production of a visualisation protein. This would enable a medical professional to confirm the presence of a tumour and localise it. However, by talking to stakeholders, we concluded that a localisation system would not work with currently available techniques. However, if a suitable visualisation technique were to be developed, this system could increase the sensing specificity of our living diagnostic. The second step in the three-step inducible system will be reserved for this purpose.

The three-step inducible system was intended to be used for the steps mentioned above. However, we made it as modular as possible so it can be coupled to different genetic circuits and introduced into other chassis.

The system is based on the binding efficiency of deactivated Streptococcus pyogenes Cas9 (dSpyCas9). dSpyCas9 is based on SpyCas9, which is a protein that is part of the self-defence mechanism of bacteria against bacteriophages in nature. When a bacterium survives an infection by a bacteriophage, it will save part of the attacker’s genetic sequence in a library called CRISPR. From these small sequences guides are created, which are used by the Cas protein to recognise the genetic sequence of the bacteriophage in a future infection. It does this by checking the DNA constantly for PAM regions, a sequence characteristic for bacteriophage DNA. The protein then matches the sequence next to the PAM to the guide and if a match is made, cuts the bacteriophage DNA [1]. In genome editing these guides can be used to target specific sequences [2]. dSpyCas9 is a non-cutting version of SpyCas9 and will not cut the specific target sequence, but only bind there [3]. This way, the dSpyCas9 protein obstructs other proteins that would normally bind at that location, for example replicases or nucleases. The binding strength of dSpyCas9 can be influenced by changing the sequence of the guide. The guide contains a spacer, a sequence that binds specifically to the target sequence. In this project three spacers called B1, t5 and P1 [4,5] were found for which the decline in binding efficiency, due to decreased length of the spacer, was known (Table 1).

A three-step inducible system was created using these spacers. This is based on the hypothesis that more dSpyCas9 proteins are needed to repress a gene with a weaker binding spacer than with a stronger binding spacer [3]. To then create three steps, the amount of dSpyCas9 must be controlled [6]. This was achieved by controlling the amount of dSpyCas9 with an inducer molecule and an inverter in between. For this inverter, the cI lambda repressor (BBa_C0051) was used. This is a protein that binds to the operator regions before the cI regulated pR promoter (BBa_R0051) [7]. Due to this inverter, the amount of dSpyCas9 will decrease when the concentration of inducer goes up (Figure 1). Additionally, the cI protein was fused to a degradation tag, stopping the accumulation of cI.

This system was tested by fusing luciferase to dSpyCas9 (BBa_K4244031) and measuring the construct at several time points at different concentrations of an inducer molecule. We hypothesised that at a higher inducer concentration, the amount of cI would increase and the amount of dSpyCas9 would go down. For this system we need a titratable inducer system, where more inducer leads to higher expression. We chose to use the inducible expression system RhaRS/PrhaB (BBa_K4244027), because for this inducer there are two knockouts known in E. coli that increase the titratability [8]. These knockouts disable the cells to transport and catabolise rhamnose. Therefore, all rhamnose that diffuses into the cells is used for the promoter system, the strain obtained from this is E. coli ∆rhaB ∆rhaT. These two knockouts were performed in EcN and the EcN ∆rhaB’∆rhaT was obtained. This strain was tested by growing them in M9 media with different concentrations of rhamnose obtained from the paper [8]. Higher concentrations were also tested. Here, only the cells without the knockout should grow. In this experiment it was observed that up to a concentration of 25000 μM of rhamnose, the knockout cells grow substantially less than normal EcN cells. From this result it can be concluded that the knockouts from the paper work in our strain and should result in a more titratable induction.

After this positive result, we performed the following experiments in the knockout strain. First, the luciferase experiment was performed with this knockout EcN strain [8]. Here, the amount of luciferase decrease, with increasing amounts of rhamnose is shown (Figure 3).

In the second part of the system the three spacers mentioned above play a role. The P1 spacer was used to test whether the binding efficiency of the spacers influences the system. Three plasmids were designed, each containing a monomeric red fluorescent protein (mRFP) (BBa_K4244035) and a P1 spacer with a different length. Every spacer targeted a specific sequence introduced between mRFP and its promoter. The goal was to produce mRFP at three different concentrations of an inducer metabolite, depending on the length of the spacer. As mentioned above, the length of the spacer influences the binding efficiency and therefore we hypothesized that more dSpyCas9 was needed to repress a weak binding guide and less repression of dSpyCas9 is needed to produce this mRFP. Therefore, the longer the spacer, the more repression under the influence of an inducer molecule is needed to produce the protein. To evaluate this system, a plate reader experiment was performed to measure the mRFP production at different rhamnose concentrations. This system was built on plasmids with a medium copy number, and the fluorescent proteins were under the influence of medium promoters and RBS regions. These strains were induced at 0.08 OD and grown over time with different concentrations of rhamnose (Figure 4).

From Figure 4, it can be seen that in the experiment with the sgRNA with the shortest target spacer of 7 nucleotides (nt), there is none or a very small amount of repression of the fluorescent gene. Therefore, the binding efficiency of this spacer is next to nothing and in future experiments with the P1 spacer, a longer spacer should be used to create a threshold system. However, in the experiments with the spacers of 9 and 20 nt very low production of mRFP can be seen, meaning that the dSpyCas9c managed to repress these genes. Additionally, it can be observed that at a rhamnose concentration of 250 μM the fluorescence increases with around 400%. This is the same for the spacer of 9 nt and spacer of 20 nt. At this point the amount of dSpyCas9 is so low, following the repression of the first part of the system, that the fluorescent protein is able to be produced. In our hypothesis the strain containing the 9 nt spacer should produce the fluorescent protein at a lower level already. This is not yet the case, both start producing mRFP between 250 and 500 μM. An explanation for this result is that dSpyCas9c is very stable, due to which a low amount of dSpyCas9 is needed to repress the system. Therefore, the fluorescent protein will only be produced at very low levels of dSpyCas9, this is after 250 μM of rhamnose. Before that, there is such a great amount of dSpyCas9 that the repression with the use of cI has almost no effect.

A future approach to create a system that reacts at different concentrations of rhamnose could be to decrease the overall amount of dSpyCas9. Reduced repression might already lead to an effect. To do this, two approaches were considered. One was to decrease the amount of protein produced, e.g. with a lower copy number plasmid. The other was to decrease the stability of dSpyCas9 by fusing it to a degradation tag. This way, more dSpyCas9 is needed to achieve the same level of repression observed in the current system. Alternatively, the current design could be used by switching the luciferase, fused to dSpyCas9, with a degradation tag. With these two approaches the two peaks from figure 4 should shift away from each other, and this way different steps will be created where the three separate genetic systems can be placed on.

Additionally, in future research the two other spacers, t5 and B1, could be tested to find the most suitable one for each step. The same experiments could then be repeated with a library of randomly generated spacers, targeting a specific sequence between the fluorescent protein and its promoter. These can contain mismatches, and length differences. From these the most suitable ones could be selected to increase the dynamic range of the system and create more distinct concentrations levels where separate steps of the system are turned on.

The three-step system described above should be inducible at different levels instead of being constitutively expressed. One reason for this is that the user of our living diagnostic should be able to activate each genetic circuit at a specific time and these should be expressed using the same metabolite. In this way, different systems can be activated while the user only has to consume more of the same metabolite. Additionally, constant expression of any of the pathways would be too much of a burden to the bacterial cells. The metabolite used for activation should not be toxic to humans at the concentrations that it is used at to activate the pathways. Several inducer systems were assessed, all with different inducers, allowing us to select the most suitable candidate for our purpose. As described before, rhamnose was selected to be one of the inducer metabolites. The other metabolites, each inducing their own promoter system, were 3,4-dihydroxybenzoic acid (DHBA), choline chloride, naringenin, vanillic acid, p-isopropylbenzoate (cumate) and propionate. DHBA is a naturally occurring benzoic acid found in several foods like fruits, nuts and vegetables, and has been found to have anti-carcinogenic properties [9]. Choline chloride is a dietary component found in foods and is important for many cellular processes and is widely used as an additive for animal feed. It has a low toxicity for humans, is not mutagenic and has no toxic effects on development [10]. Naringenin is a flavourless, colourless flavonoid found in a variety of fruits and herbs. It is being researched for several biological effects, including having an anticancer effect by inhibiting tumour growth [11, 12]. Vanillic acid is a constituent of vanilla, commonly used as a flavouring agent. It has been found to improve the condition of patients with both cancer and obesity [13]. P-isopropylbenzoate, also called cumate, is used as a preservative and thus safe for consumption. Choi et al. developed a tightly regulated, titratable gene expression system using this inducer [14]. The same was achieved by developing a promoter system induced by propionate, which is also used as a food preservative [15].

Each of these inducers corresponds to a promoter system that was put into a low copy number plasmid to mimic genomic production [14–16]. The plasmids were designed in such a way that the results could be compared with the results of the three-step system described above. These inducible promoters caused the production of the cI repressor which then represses luciferase production. An increasing metabolite concentration should thus cause a lower luciferase signal.

In Figure 4, the result of one of the inducer experiments with cumate can be seen. EcN cells with the corresponding plasmid were induced with cumate and then measured with a plate reader. In this figure the measurement 3 hours after induction can be seen. At first hand the luminescence seems to decrease when the cumate concentration increases, but no significant conclusions can be drawn from this when looking at the error bars. Other inducer systems had similar results, or did not show any luminescence. This means that either the tested inducer systems were not titratable, or that something in the plasmid design caused them to not be titratable. Either way, the rhamnose system tested before is the best option for now. Additionally, more research should be done on this subject until the best titratable, inducible promoter system can be chosen for our living diagnostic.