Engineering

For our tattoo to change contrast in response to a biomarker, we chose to rely on the pigment melanin (Guo et al., 2023). Since we want to control the localisation of melanin, as well as reduce it's toxicity for the cells, we designed synthetic melanosomes for a spatially confined melanogenesis by tyrosinases. These synthetic melanosomes are made up of melanin-producing tyrosinases packed in bacterial nanocages - so called encapsulins (Allen et al., 2024). To form our nanocages for the later pigment production, we first designed an encapsulin system derived from the Myxococcus xanthus (Mx) encapsulin (Sigmund et al., 2018). Then, we designed variants using the encapsulin of Quasibacillus thermotolerans (Qt) as well as of Thermotoga maritima (Tm). These "naked" encapsulins served as the basic building block of our encapsulin system and provided us with a foundation for developing regulated and functionalized encapsulin variants.

In addition to using encapsulins of different bacterial strains, we also designed multiple strategies for tyrosinase loading. Our cargo-signal loading approach is made up of two plasmids, one expressing the encapsulin, and the other expressing the tyrosinase derived from Bacillus megaterium (BmTyr), coupled with a cargo signal, to achieve the self assembly of the encapsulins, followed by loading the nanocages with the tyrosinase.

Another approach is the direct fusion loading, where we fused the BmTyr directly to the N-terminus of our encapsulins, so that selfassembly of our encapsulins would result in already tyrosinase-loaded nanocages, and therefore bypassing the need for separate cargo loading.

Lastly, we also designed pro-encapsulins, which allow for controlled nanocage assembly. First, assembly is blocked by steric hindrance domains made up of Lanthanide-Binding Tag (LBT-15) repeats. Once the LBT-15 domains are cleaved off with a TEV protease, nanocage assembly is induced.

We consulted with Prof. Stefanie Frank (University College London), who is a recognized expert in both encapsulin nanocages and cell-free expression systems, and gave us valuable input for our design and experimental strategy. She advised us to use high-performance liquid chromatography (HPLC) with size-exclusion chromatography (SEC) to validate assembly of our nanocages, since assembled encapsulins should yield a distinct HPLC profile, with signals expected at 280 nm for monomeric proteins and 320 nm for assembled nanocages. Therefore, we added a strep-tag II to our encapsulin constructs, which allowed purification over a strep-column.

We build a series of encapsulin constructs, all expressed under both a CMV and a T7 promoter.

The CMV promoter enables robust expression in mammalian, and therefore HEK293T, cells, while the T7 promoter allows for potential in vitro expression in cell-free systems. In the end, we prioritized in vivo expression to more closely reflect our envisioned biosensing application, as well as to follow the advice from Prof. Stefanie Frank, who warned us that cell-free expression often leads to self-assembly of empty nanocages. The core of all of our encapsulin constructs consists of the "naked" encapsulins from Myxococcus xanthus, Quasibacillus thermotolerans or Thermotoga maritima, followed by the genetically encoded bilirubin (BR)-inducible fluorescence protein UnaG (eUnaG) as a fluorescent tag, and a strep-tag II. They also feature a HIV-1 Rev nuclear export signal (HIV-1 Rev NES), to ensure cytoplasmic localization after cell division.
For the pro-encapsulins, we used three repeating LBT-15 domains, separated by GS linkers, upstream of the encapsulin seqences, that prevent the encapsulins from self assembling. Between these domains and the encapsulin sequences we placed a TEV cleavage sequence, to achieve inducible nanocage assembly upon proteolytic cleavage. This design directly links encapsulin assembly to the detection of a biomarker by our modular extracellular sensor architecture (MESA) receptor. Upon biomarker binding, MESA dimerizes and reconstitutes a split TEV protease, which can then cleave the pro-encapsulins, inducing assembly of nanocages only when the biomarker is present.
For the tyrosinase-encapsulin fusion constructs, we replaced the LBT-15 domains with the sequence of the BmTyr to decrease melanin production outside of the nanocages, and therefore reducing toxicity , as well as bypassing the need for correct loading of the nanocages. In parallel, we build the cargo-based constructs as an alternative strategy for nanocage loading, containing an encapsulin-specific cargo signal (MxSig, QtSig, or TmSig (@Sigmund_Massner_Erdmann_Stelzl_Rolbieski_Desai_Bricault_Wörner_Snijder_Geerlof_et_al_2018 , @Sigmund_Pettinger_Kube_Schneider_Schifferer_Schneider_Efremova_Pujol-Martí_Aichler_Walch_etal_2019 , McHugh et al., 2014)) upstream of BmTyr, as well as a destabilization domain (DD-C) (Sigmund et al., 2018). The DD-C domain ensures decreased toxicity due to the tyrosinase activity, by making the tyrosinase unstable in the cytoplasm unless it's correctly encapsulated. This allowed us to test tyrosinase loading without decreasing cell viability. "Naked" encapsulins were assembled using restriction enzymes followed by ligation, while all other constructs - including pro-encapsulins, fusions, and cargo-based variants - were cloned via Gibson Assembly.

To evaluate the assembly of our encapsulin nanocages, we transfected HEK293T cells with various combinations of our designed constructs, lysed the cells 24 hours after transfection, and purified our proteins over the strep-tag II using a strep-tactin column.

Encapsulin	Plasmid
”Naked” encaspulin	MxEnc-eUnaG-STII-NES
---	QtEnc-eUnaG-STII-NES
---	TmEnc-eUnaG-STII-NES
”Naked” encapsulin with cargo	MxEnc-eUnaG-STII-NES
---	TmEnc-eUnaG-STII-NES
Pro-encapsulins	3xLBT15-TEVCS(M)-MxEnc-eUnaG-STII-NES
---	3xLBT15-TEVCS(M)-QtEnc-eUnaG-STII-NES
---	3xLBT15-TEVCS(M)-TmEnc-eUnaG-STII-NES
Pro-encapsulins with cargo	3xLBT15-TEVCS(M)-MxEnc-eUnaG-STII-NES
---	MxSig-BmTyr-DD-C
---	3xLBT15-TEVCS(M)-TmEnc-eUnaG-STII-NES
Tyrosinase-encapsulin fusions	BmTyr-TEVCS(M)-MxEnc-eUnaG-STII-NES
---	BmTyr-TEVCS(M)-QtEnc-eUnaG-STII-NES
---	BmTyr-TEVCS(M)-TmEnc-eUnaG-STII-NES

For transfection we used a 24-well plate. We transfected three wells individually with each of the "naked" encapsulin variants - Mx, Qt or Tm. Additionally, we co-transfected two wells with the "naked" Mx or Tm encapsulins together with their corresponding cargo constructs to test successful loading. The combination of the Qt encapsulin with its respective cargo was omitted due to unsuccessful cloning of the cargo plasmid containing the QtSig sequence. For the pro-encapsulins, we transfected three wells with each of the pro-encapsulin versions separately - Mx, Qt or Tm -, as well as two wells with a co-transfection of the Mx or Tm cargo constructs with their respective encapsulin construct. To test whether the pro-encapsulins assemble upon TEV cleavage, we planned on comparing samples treated with purified TEV protease post-lysis against untreated controls. Lastly, three wells were transfected with each of the tyrosinase-encapsulin fusion constructs without the addition of cargo plasmids, as they don't require separate loading with a tyrosinase. After purification, protein concentrations were quantified using a nanodrop spectrophotometer at 280 nm to ensure sufficient yield for downstream HPLC analysis.

The measured protein concentrations after purification were approximately tenfold lower than the amount reqquired for HPLC analysis, ranging from 0.00 mg/mL to 0.05 mg/mL. Consequently, we were unable to proceed with HPLC-based evaluation of encapsulin assembly. Several factors could have contributed to this low protein yield. One likely cause could be the strep-tactin column itself, which had been already used multiple times before our experiment, and therefore may have lost binding efficiency due to imcomplete regernation. Another possibility is that the strep-tag II on our encapsulin constructs was not sufficiently exposed or sterically accessible enough, leading to insufficient binding during purification. Lastly, insufficient expression levels of our constructs in the HEK293T cells could also have led to reduced proteins yields. To overcome these limitations, we changed our approach to using Native PAGE for assembly validation. This method not only bypasses the need for purification but also allows for direvt visualization of encapsulin expression and assembly. Due to the inclusion of the eUnaG fluorescent tag, we could simultaneously confirm protein expression and detect assembled nanocages based on their migration pattern and fluorescence signal.

To identify and engineer tyrosinases suitable for melanin synthesis within our synthetic melanosomes, we designed 18 tyrosinase variants covering a range of natural, rationally mutated, and engineered constructs (Pretzler and Rompel, 2024). Our goal was to achieve efficient yet controllable melanin production while minimizing cellular toxicity caused by uncontrolled enzyme activity. We selected six wild-type tyrosinases from diverse organisms - Bacillus megaterium (BmTyr), Hahella (HcTyr), Laceyella sacchari (LsTyr), Streptomyces avermitilis (SavMel), Streptomyces kathirae (SkMel), and Verrucomicrobium spinosum (VsTyr) based on their structural simplicity, documented activity, and availability of structural data. Building on insights from scientific literature, expert consultation with Prof. Pretzler, and computational modeling, we designed additional mutant and engineered versions aimed at modulating enzymatic activity and regulation. To prevent uncontrolled melanin production, we created split Bm tyrosinase variants, in which the enzyme is divided into two inactive fragments that regain activity upon reconstitution - which can be achieved during encapsulin assembly.

We also explored tyrosinase-caddie systems from Streptomyces species (Leu et al., 1992) to faciliate copper delivery under physiological conditions.

Finally, we fused LID-domains to the Vs and Hs tyrosinases (de Almeida Santos et al., 2024; Fekry et al., 2023; Son et al., 2018) to enable reversible inhibition of enzyme actvity and therefore prevent uncontrolled melanin synthesis.

All constructs were designed for rapid prototyping in a cell-free expression transcription-translation (TXTL) system, allowing us to dorectly assess enzyme activity and assembly behavior without the complexity of to enable rapid prototyping of enzyme activity without cellular complexity (Garamella et al., 2019).

For the cell-free expression of our designed tyrosinases, we used a vector containing a TXTL-compatible Linear Expression Template (LET), made up of a T7 promoter, a ribosome binding site (RBS), the respective tyrosinase coding sequence, and a terminator sequence. To enhance protein solubility and stability during expression, as well as to enable potential downstream purification, we included a CAT-Strep-Linker (CSL) upstream of the tyrosinase sequence (Kightlinger et al., 2019). To ensure that the CSL doesn't interfere with the tyrosinase's activity, we added a Factor Xa protease cleavage site downstream of the CSL, allowing for post-translational removal if necessary. The designed tyrosinase sequences were inserted into the LET backbone using Gibson Assembly. Then, assembled constructs were amplified via Rolling Circle Amplification (RCA) and used as DNA templates for TXTL expression.

To confirm the correct assembly of our tyrosinase constructs, we linearized the plasmids and verified their sizes via agarose gel electrophoresis.

Then, cell-free expression was carried out using TXTL. To assess expression efficiency, we quantified total protein yields using a bicinchoninic acid (BCA) assay.

Even though agarose gel electrophoresis confirmed that our tyrosinase constructs assembled correctly, cell-free TXTL expression resulted in protein yields below the detection limit of the BCA assay, indicating either insufficient expression or incorrect maturation of the enzymes under TXTL conditions. To determine whether our tyrosinases were expressed at all and to facilitate troubleshooting, we decided to perform a Western Blot for specific protein detection.

For our first iteration, we based the design of our systems receptor part on the work of the authors of the original MESA (Modular Extracellular Sensor Architecture) receptors (Daringer et al., 2014) while also incorporating insights from subsequent improvements documented in later publications (Dolberg et al., 2021). Daringer et al. identified the TEV protease cleavage sequence ENLYFQ*M, which, compared to the wild-type sequence, enhanced the fold induction of downstream signaling while minimizing background activity. Key advancements from the work of Dolberg et al. involved optimizing the split TEV protease system, particularly identifying N-terminal and C-terminal variants with desirable properties for receptor activation-induced reconstitution. Here, they revealed H75E and H75S to be the most compatible the N-terminal variants and L190K to be the most compatible C-terminal variant with the other components of MESA receptor system. While additional modifications, such as alternative transmembrane domains, were explored in the literature as well, we decided to focus on the well-documented original designs to ensure reproducibility and clarity in our initial experiments.
For our proof-of-concept MESA receptors, we designed constructs incorporating LaM2 and LaM4 anti-mCherry nanobodies as ligand binding domains, CD4 signal peptide for membrane localization of the receptor chains, the CD28 transmembrane domain, the split TEV protease domains and cleavage sequence mentioned above, and TEV-cleavable receptor cargo domains. These cargo domains consist of either tetracycline-controlled transactivator (tTA) for induction of tetracycline response element (TRE)-controlled reporter genes, and two controls applying reverse tetracycline-controlled transactivator (rtTA) for tetracycline-inducible reporter expression and Tet repressor (TetR) as a sort of negative control. To be monitor the localization of these transcription factors in receptor-bound and released state, we tagged them with blue fluorescent protein (BFP).
To refine our design and ensure biological plausibility, we arranged an expert meeting with Prof. Josh Leonard, whose group originally developed the MESA system. His feedback confirmed that our designs were promising and recommended starting with soluble, rapamycin-inducible MESA receptors for rapid proof-of-concept testing. These soluble constructs are easier to express and measure, enabling faster iteration and more reproducible data. Based on his advice, we decided to include soluble rapamycin-inducible MESA as a parallel approach alongside our anti-mCherry designs, to validate the system before attempting more complex receptor architectures.
References

For our proof-of-concept experiments, we constructed seven mCherry-inducible MESA receptor variants. Two of these receptors incorporate the LaM4 anti-mCherry nanobody, CD4 signal peptide, CD28 transmembrane domain, and either the H75E or H75S variant of the split TEV protease N-terminal domain. The remaining receptors replace LaM4 with the LaM2 anti-mCherry nanobody and incorporate the L190K C-terminal TEV protease variant. Among these C-terminal receptor chains, one does not include a cargo domain and was intended solely to activate the synthetic MESA pathway in later experiments. The other three chains include the previously mentioned transcription factors tTA, rtTA, and TetR. Each of them is fused to BFP to facilitate visualization and tracking.
For the soluble rapamycin-inducible MESA receptors we constructed two receptor vatiants. The constructs for a whole receptor were designed as a single expression unit, meaning that both receptor chains are part of the same construct and sperated by a T2A peptide. This esnures expression of unfused receptor chains while minimizing cloning effort. The constructs compose an HA-tag for possibility of purification, FKBP as first rapamycin-binding domain, either the H75E or H75S variant of the split TEV protease N-terminal domain, T2A peptide, FRB as second rapamycin-binding domain and the L190K variant of the C-terminal domain of split TEV protease, accompanied by linkers for proper flexibility between the different receptor domains.
For the soluble rapamycin MESA receptors, we also created a TEV-cleavable membrane-bound tTA that would be released upon soluble MESA dimerization, translocate to the nucleus and induce reporter gene expression. The consruct for the membrane-bound tTA composes CD4 signal peptide for membrane targeting, HA-tag for possibility of purification, CD28 transmembrane domain, TEV cleavage sequence and tTA fused to BFP for fluorescence-tracking of this transcription factor. The different functional receptordomain are fused together by different linkers to provide appropriate flexibility.

To evaluate our soluble MESA constructs, we transiently transfected HEK293T cells with different combinations of receptor components and a YFP reporter plasmid under rapamycin-induced and non-induced conditions.
Control experiments confirmed low background fluorescence and proper construct expression. However, the YFP reporter showed a noticeable basal signal even in the absence of the transcription factor, indicating leakiness of the tetracycline-responsive promoter.
When testing both receptor chains together, we observed that cells expressing the tTA-BFP transcription factor exhibited strong and constitutive YFP expression, independent of rapamycin addition. This suggests that the transcription factor was active without requiring cleavage by the split TEV protease, likely due to non-specific proteolytic release or spontaneous activation. The presence of Ntev and Ctev domains did not significantly affect expression patterns, and no rapamycin-dependent increase in fluorescence was detected.
Overall, while the constructs were successfully expressed and produced strong reporter activity, the system failed to exhibit controlled, ligand-dependent activation. These findings indicated that our current receptor architecture allowed background transcriptional activation, masking any inducible signal. Similarly, the membrane-tethered constructs showed functional expression of their components. While co-transfection of Ntev and Ctev fragments led to increased BFP fluorescence, confirming TEV reconstitution and release of the Ctev-BFP domain, this activation occurred regardless of mCherry ligand presence. Both induced and non-induced samples showed comparable nuclear BFP localization, indicating that the receptor system was constitutively active. The most likely causes are spontaneous receptor dimerization or proximity-induced reconstitution of the split protease at the membrane.
Across both systems, the MESA constructs demonstrated strong expression and detectable downstream signaling, but the ligand-dependent control central to MESA function was absent. In both cases, constitutive activation overshadowed any inducible effects. These findings revealed the complexity of reliably balancing receptor stability, membrane localization, and protease control, underscoring the need for more systematic design strategies to predict and prevent unintended background activation.

From this iteration, we learned that while MESA receptors are highly modular and conceptually powerful, manually designing functional receptors is challenging due to the vast design space and intricate interactions among receptor domains. The difficulty in obtaining robust reporter activation highlighted the need for computational support to systematically explore compatible component combinations, linker lengths, and split protease configurations.
This experience directly motivated the development of MESA Designer, a software tool intended to streamline receptor design, ensure compatibility of modular components, and reduce trial-and-error in wet-lab experiments. By automating sequence assembly, validating component compatibility, and providing annotated, production-ready constructs, MESA Designer addresses the key limitations revealed in our first engineering cycle.