The results from an earlier Gibson assembly were difficult to interpret. The positive control produced no colonies on plates, the water control produced many small colonies and growth appeared in several liquid cultures. Because the controls did not behave as expected, we could not confidently attribute growth to successful assembly of the SSB library.

Rather than continuing to troubleshoot on uncertain foundations, the team introduced stricter contamination checks. Media flasks would be labelled and incubated after use, work surfaces and pipettes would be sterilised daily, and suspect consumables were replaced. We also repeated the Gibson assembly with support from a neighbouring laboratory. This allowed the project to move forward while preserving time for the selection stage.

Improved purification protocols produced DNA with more consistent purity, and the SSB library and original SSB control were assembled and transformed. Plasmids were extracted at usable concentrations, and the library sample was sent for sequencing. Sending the constructed library for sequencing was a key checkpoint: before applying selection, we need to know whether the sequences we designed are actually represented in the biological population.

The team also prepared degenerate De Novo and NEXT tags and calculated the DNA ratios required for further assemblies. Low yields from one gel-extraction kit reinforced an earlier lesson, and we chose an alternative purification method rather than allowing the same bottleneck to dominate another week.

Alongside construction, we refined our plans for final validation. Several possible target proteins were considered for testing whether selected IDP tags improve both solubility and function. After reviewing experimental precedent and feasibility, citrate synthase and lactate dehydrogenase emerged as practical candidates for the tardigrade-inspired assays. This complements the high-throughput beta-lactamase selection by providing more direct measurements of enzyme protection or activity retention.

The week brought the project closer to selection, but its most important outcome was stronger experimental discipline. Reliable controls, contamination monitoring, sequence verification and meaningful validation targets are what allow an apparent improvement to become a defensible result.

Selective and fluorescent plating confirmed that the required transformed cells had been obtained. We then attempted to convert the single-stranded degenerate tag oligos into double-stranded DNA and used the SSB library in a Gibson assembly. Although the assembly mixture appeared to contain abundant DNA, the transformed cells initially failed to grow. Gel analysis later showed that the direct single-strand-to-double-strand conversion had produced a smear, whereas an amplified product gave a clear band.

This result changed the immediate workflow. Rather than relying on direct conversion alone, we moved to amplification before Gibson assembly, then repeated the assembly and transformation with positive and negative controls. The failure was therefore informative: it connected a downstream transformation result to an earlier library-preparation step and showed why concentration measurements alone cannot confirm that a DNA sample contains the intended product.

The error-prone PCR branch also continued, although the project timeline forced us to narrow its scope. We established that the reaction could be reproduced and compared purified with unpurified templates. Purified material performed better, while unpurified reactions produced a short unwanted product. Further mutation rounds were completed for SSB and NEXT, but limited time meant that the number of rounds and variants entering selection would need to be prioritised.

For the tardigrade-inspired branch, we finalised protocols for heat shock and osmotic stress and adjusted the desiccation setup after thin filter paper curled during drying. These practical changes were guided by the same principle as our library work: the assay must be physically reproducible before it can tell us whether an IDP provides protection.

The team also met with potential reagent partners, securing support that could reduce the cost of future orders and troubleshooting.

By the end of the week, we had not yet completed selection, but we had crossed an important boundary. The designed sequence libraries were no longer only computational objects; they were being amplified, assembled and transformed. Each failure was helping us define the conditions needed to preserve the link between sequence design and biological testing.

The beta-lactamase construct had produced transformed colonies, allowing us to extract plasmids and prepare for a second linearisation. Adjusting the primer concentrations improved this PCR and produced clear bands. Sequencing confirmed that beta-lactamase had been inserted correctly into one construct, supporting the next assembly stage. However, another sequencing result showed that a sample believed to contain the original pBAD plasmid was actually the beta-lactamase-containing version. A growth-curve dry run also behaved more like DH10B than the intended BL21 strain.

Together, these results shifted our emphasis from simply progressing through the protocol to verifying strain and plasmid identity at each stage. Fluorescence under arabinose induction, selective plating, diagnostic PCR and sequencing became complementary checks rather than optional confirmation at the end.

The error-prone PCR branch faced a different problem. We tested DMSO, betaine and extension time to reduce mispriming and identified conditions that initially produced the expected fragment. Yet the next round again generated smears or failed amplification, even when repeated by another team member. This suggested that the problem was inherent to the reaction or template rather than a single pipetting mistake. We therefore began developing a computational tool to analyse possible mispriming sites and paused plans to model mutation bias until enough later-round sequencing data could be generated.

We also continued refining the tardigrade-inspired experiments. Recovery and plating conditions were varied after the first desiccation dry run produced either no colonies or lawns that were too dense to interpret. The team learned to use a plate reader and drafted a growth-curve protocol, but contamination and a power interruption made the first run inconclusive.

This week did not produce a simple forward march. Instead, it exposed several hidden assumptions about what was in our tubes and which organisms were being measured. Finding these problems changed how we approached the project: identity checks, controls and reproducibility now needed to be built into the workflow before selection results could be interpreted with confidence.

We compared digested and undigested PCR products, varied elution temperature and incubation time, and examined gel extraction alongside PCR purification. The results showed that DpnI digestion was not responsible for the contamination. A longer, warmer elution step produced more promising DNA yields, although the first apparent success could not immediately be reproduced. Further tests suggested that PCR purification gave a better balance of yield and purity than gel extraction for some samples, while gel extraction remained useful when an unwanted band had to be physically removed.

These experiments finally produced material good enough to move forward. The team completed a Gibson assembly using the linearised pBAD backbone and beta-lactamase insert, transformed the product into bacteria and began isolating colonies. This was an important shift: the project had moved beyond preparing separate DNA fragments and into constructing the plasmid needed for selection.

The error-prone PCR workflow also became more defined. By repeating the manganese and magnesium optimisation gel, we identified conditions that produced clearer amplification. Later rounds still showed mispriming, so we compared gel-extracted and PCR-purified templates and began developing tools to process sequencing data. These steps reflect an increasingly iterative approach: each round is not only intended to generate diversity, but also to reveal which conditions preserve enough usable material for the next round.

Progress continued on the tardigrade-inspired branch. We prepared induced cells and ran a first desiccation trial using saline. When diluted samples failed to produce colonies, we reconsidered the recovery conditions, including elution volume, vortexing and plating. Rather than treating the dry run as a definitive test of protection, we used it to identify practical weaknesses in the assay itself.

A power outage ended wet-lab work early on the final day, but it also provided time to review the cloning timeline, mispriming problem and stress-assay design. By the end of the week, the purification bottleneck had not disappeared entirely, but it was no longer preventing all downstream progress. We had reached assembly, generated new sequencing data and developed clearer questions for both experimental branches.

The pBAD linearisation produced a clear band at the expected size, showing that the PCR itself had worked. However, purification after DpnI digestion gave DNA at low concentration and with substantial contamination. Error-prone PCR products showed similar problems. We therefore moved from column-based PCR purification to gel extraction, expecting that isolating the correct band would improve sample quality.

Instead, gel extraction repeatedly failed. Low yields and strong contamination signals appeared across different samples and operators, including a supervised test using a separate sample. Because Gibson assembly requires DNA of sufficient concentration and purity, these failures prevented us from progressing to plasmid assembly and transformation. What first looked like a minor cleanup problem had become the rate-limiting step for the beta-lactamase branch.

The team responded by treating the failure as a structured troubleshooting problem. We increased the number of PCR cycles to raise the starting DNA concentration, compared buffer systems, varied handling steps and considered whether the extraction kit itself was at fault. As repeated attempts ruled out simple operator error, we decided to source a different kit and return to a TAE gel system compatible with it.

In parallel, we investigated the smeared products produced during error-prone PCR. Tests with different polymerases and annealing temperatures showed that mispriming decreased at higher temperatures, allowing us to identify a more suitable range for subsequent optimisation of manganese and magnesium concentrations.

The week also revealed an incorrect base in the original NEXT tag sequence. Although this delayed that branch, finding the error before assembly prevented it from propagating further through the workflow.

This was a week shaped by failure, but it clarified where progress depended on method reliability rather than design ambition. By the end of the week, we had converted several vague problems into specific hypotheses and experiments. That troubleshooting work became an essential part of building a selection pipeline we could trust.

One important decision concerned how to introduce diversity. We compared commercial error-prone PCR kits with a custom method based on standard Taq polymerase and altered manganese and magnesium concentrations. Commercial kits offered convenience, but their cost would have limited the number of experiments we could run. We therefore developed a lower-cost error-prone PCR strategy that can be optimised experimentally. This also led us to think more carefully about the balance between mutation rate and usable DNA yield: more mutations are not necessarily helpful if amplification becomes unreliable.

We also revised how our designed libraries would be ordered. Double-stranded degenerate DNA was prohibitively expensive, so we moved towards single-stranded synthesis followed by conversion to double-stranded DNA in the laboratory. This reduced costs while keeping the broader design-build-test-learn cycle intact, although it introduced an additional step that would later need careful validation.

At the same time, the team refined both experimental branches. For the beta-lactamase system, we mapped the stages required for plasmid construction, transformation and selection, and began questioning whether liquid selection would give a reliable measure of improved protein performance. For the tardigrade-inspired system, we narrowed the possible stress assays and developed a practical desiccation setup using filter paper and a desiccator. We also began considering controls that could distinguish genuine protection from simple recovery under favourable conditions.

The week ended with the first preparations for experimental work: antibiotic plates were poured, bacterial strains were cultured and chemically competent cells were produced. We also revised our schedule and accepted that two well-supported rounds of evolution would be more realistic than three rushed rounds.

The main outcome of the week was not a single experiment, but a more grounded project. Our early ideas had become a set of testable workflows, with clearer constraints and more explicit assumptions to examine in the weeks ahead.

A major decision was to simplify the tardigrade branch. Instead of using LDH as the main selection system, we are moving towards survival-based testing after desiccation or freeze-thaw stress. This should make the workflow more consistent with the rest of the project and easier to scale.

We also updated the tardigrade tag design. The team plans to use a species where truncated tardigrade proteins have already been reported to retain function, and to focus first-round mutations on a better-defined motif. The current design rationale is based on helical propensity, controlled hydrophobicity, and avoiding changes that could harm bacterial growth.

Finally, we discussed the difference between selection and validation. Selection can tell us which variants survive or become enriched, but validation is needed to show why. Our next challenge is to connect survival improvements back to protein solubility using assays such as soluble-versus-insoluble beta-lactamase measurements.

Overall, the project is moving from broad design planning into more concrete ordering, cloning, selection, and validation decisions.

Beta-lactamase selection became the most developed experimental branch. The basic idea was to attach candidate IDP tags to beta-lactamase, expose cells to antibiotic stress, and identify variants that allow better survival. This gave us a potential high-throughput selection system.

However, the details became more complicated. Liquid selection looked attractive because it could screen many cells at once, but we identified a major possible artefact: beta-lactamase released from lysed cells could reduce antibiotic pressure for the whole culture. This would make weak variants look better than they really are. Because of this, plate-based or gradient-based selection remained an important backup.

We also looked more carefully at cloning. Gibson assembly was still attractive, but primer design and possible mispriming against beta-lactamase made it less straightforward than expected. Restriction cloning remained a fallback, although we needed more advice on cost, yield, and timeline.

The computational side also became more disciplined. Instead of relying on one very broad degenerate library, we began considering sequence binning and compression. The aim was to keep libraries within a realistic size while tracking how much sequence information was lost.

We began by narrowing our choice of target protein. An earlier chloramphenicol-related system became less attractive because of structural complications, so we shifted our attention toward an ampicillin-resistance enzyme as a more practical selectable target. The idea was to use a weakened or aggregation-prone version of the enzyme, then ask whether different IDP tags could improve its performance.

We also started refining our tag-generation strategy. Rather than mutating entire tags blindly, we planned to identify likely mutation regions first, then feed selected variants into our reverse-translation workflow. This was important because very large degenerate libraries quickly become impossible to sample properly.

A major discussion point was the trade-off between coverage and off-target sequences. Degenerate codons can represent many variants, but they can also generate sequences we did not intend to test. Our early goal was therefore to reduce unnecessary variation while still keeping enough diversity for selection.

By the end of this stage, we had early DNA designs for SSB, NEXT/N11, and the beta-lactamase target, but several decisions still needed confirmation: whether the target system was final, whether the mutation strategy was acceptable, and whether our reverse-translation approach gave a sensible balance between library size and sequence coverage.

This matters because proteins are central to biotechnology. They can act as enzymes, reagents, diagnostics, therapeutics, and research tools, but only if they can be expressed, handled, and stored in a functional form. If a protein aggregates easily, even a promising biological idea can become difficult to turn into a useful system.

We became interested in intrinsically disordered proteins, or IDPs, because they offer a different way of thinking about this problem. Unlike structured proteins, IDPs remain flexible and dynamic. Their sequences are often rich in charged and polar residues, which may help them interact with water, reduce aggregation, or keep neighbouring proteins apart.

This led us to the idea of using IDP-inspired sequences as solubility tags. Instead of redesigning every difficult target protein directly, we asked whether a transferable tag could improve the behaviour of a target protein.

Several ideas shaped our early thinking:

• known solubility tags, such as SSB and NEXT/N11;
• de novo IDP-like sequences designed around charge balance and hydrophilic character;
• tardigrade-inspired proteins, which may help protect biological material during stress such as drying;
• possible mechanisms such as entropic bristle effects, protective matrices, or cage-like behaviour around a target protein.

At this stage, the project was still broad. We were not yet sure which target protein, selection method, or validation assay would work best. But the core direction was already clear: generate candidate IDP tags, test whether they improve protein performance, and use the results to guide the next design cycle.

Our early goal was not just to find one working sequence. We wanted to build a workflow that could connect sequence design, experimental selection, and biological interpretation. That remains the central idea behind the project.