NeuroSplice - Project: Engineering

The NeuroSplice Engineering Process

We didn't just build a sensor—we forged a precision diagnostic tool through relentless Design, Build, Test, Learn (DBTL) iteration.

Our core mission was to conquer the challenge of detecting the sIL7R exon 6 skipping variant, a known susceptibility factor for Multiple Sclerosis (MS), using a toehold switch in a cell-free system. This required seamless integration of Wet Lab validation, Dry Lab computational design, and Human Practices insights. Every pivot detailed below—from reporter swaps to sequence tuning—was crucial in transforming our in silico concept into a robust, real-world ready biosensor.

Overall Project Process Diagram showing the DBTL cycle flow from Design to Learn across Wet Lab, Dry Lab, and Human Practices.

Wet Lab: Experimental Validation & Optimization

Our experimental journey focused on maximizing the ON/OFF ratio and increasing the speed of the toehold switch by systematically modifying the reporter protein and the surrounding genetic sequence.

Toehold Switch (OFF State) Diagram

Diagram of the toehold switch in the repressed OFF state, showing the hairpin structure blocking the ribosome binding site.

The hairpin structure prevents translation initiation in the absence of the trigger RNA.

Control Toehold Switch (ON State) Diagram

Diagram of the toehold switch in the active ON state, showing the trigger RNA binding and linearizing the switch.

The trigger RNA binds, unzipping the hairpin and enabling ribosome access to the CDS.

Trial 1: gBlock + amilCP (The Chromoprotein Challenge)

Design: We initiated our work with a proof-of-concept construct that encoded amilCP, a chromoprotein reporter. This reporter was selected because it would present a visible, colorimetric readout that could be observed with the naked eye, and did not require any instrumentation. The coding sequence was concealed within a toehold switch hairpin, such that only sIL7R (the variant enriched in MS patients) trigger RNA would expose the Ribosome Binding Site (RBS) and start codon. Build: A linear DNA gBlock containing the toehold with the amilCP reporter sequence was synthesized and resuspended to 160 nM, rather than the 96 nM recommended in TXTL protocols, to account for addition of the sIL7R trigger RNA. DNA was used at approximately 5 nM final concentration. Test: TXTL reactions were assembled with four conditions: Experimental (with trigger), Positive Control (T7 deGFP), Negative Control 1 (no trigger), and Negative Control 2 (Randomized trigger). Plates were incubated at 27 °C for 18 hours with readings set to an interval of every 10 minutes. Learn: We observed that amilCP expression was slow and inconsistent even under similar conditions, making it difficult to distinguish clear ON and OFF states. This variability was likely caused by the longer folding and chromophore maturation times required for chromoproteins. Based on these findings, we immediately pivoted to Green Fluorescent Protein (GFP) for a more reliable, quantitative, and real-time measurement of gene expression (Alieva, Naila O et al., 1994).

Trial 2: gBlock + GFP (Quantification & Leakiness)

Design: We replaced amilCP with GFP to improve sensitivity and measurement resolution, as GFP matures rapidly and produces a stable fluorescent signal (Chalfie, M et al., 1994). The GFP coding sequence remained embedded within the toehold switch hairpin, keeping the RBS and AUG masked until sIL7R trigger RNA recognition. Build: A new linear DNA gBlock was synthesized. The DNA was used at approximately 5 nM final concentration. Trigger RNA was added at 2 to 10 times molar excess to evaluate switch responsiveness across a dynamic range. Test: Reactions were incubated at 27 °C with fluorescence recorded every 10 minutes at 485/510 nm. This allowed continuous, quantitative tracking of ON and OFF states. Learn: We observed strong ON-state signal, confirming switch function. However, we detected modest background leakiness in the OFF condition. This basal signal suggested incomplete hairpin formation or transient unfolding at the 5' untranslated region, momentarily exposing the RBS and permitting low-level translation initiation (Elizaveta Razumova et al., 2025). This underscored the need for design optimization to improve regulatory tightness.

Trial 3: gBlock + GFP (Upstream Buffer Sequences)

Design: To reduce OFF-state leak, we incorporated short, neutral buffer sequences between the promoter and the toehold switch. These spacers were designed to insulate the hairpin from upstream promoter effects that might interfere with its proper folding or unintentionally expose the RBS, aiming to stabilize the switch structure. Build: The modified construct, containing the added buffer sequences, was synthesized as a new gBlock and used at approximately 5 nM in the TX-TL system. Test: We ran TX-TL reactions using the buffered construct in the ON condition to assess whether the upstream insulation affected activation efficiency and fluorescence intensity. Learn: Adding buffer sequences made GFP expression more consistent than in previous trials, suggesting improved stability. However, the ON-state fluorescence still showed noticeable variability between replicates. This inconsistency suggested that sequence-related factors downstream of the start codon might still be influencing translation efficiency or mRNA folding (Green et al., 2014).

Trial 4: gBlock + GFP (Guanine Content Reduction)

Design: We hypothesized that G-rich regions downstream of the GFP coding sequence could form strong secondary structures that interfere with ribosome movement. To improve translation elongation, we designed a new construct with a reduced guanine (G) content in the trailing region to minimize stable RNA hairpins and improve ribosome processivity.

Build: A new gBlock with the altered trailing sequence was synthesized and prepared under the same DNA concentration protocol. Test: Fluorescence kinetics were tracked at 27 °C. Time-to-half-maximum signal was compared with earlier trials to assess improvements in ribosome progression. Learn: Reducing the G content improved translation kinetics and resulted in a modest increase in ON-state fluorescence intensity (Picard et al., 2023). However, residual OFF-state fluorescence was still detectable. This trial confirmed that both coding and regulatory sequence compositions must be co-optimized to achieve strong activation while minimizing unintended background expression.

Trial 5-10: gBlock + sfGFP (Final Optimization)

Design: All previous components of the toehold were maintained but the GFP variant was swapped (sfGFP). This was selected for faster maturation and stronger fluorescence intensity. This was once again extracted from an established sequence data based and optimized to fit our toehold switch design. Build: The sfGFP coding sequence was synthesized as a new gBlock while keeping the same optimized toehold switch structure. DNA was used at around 5 nM in TX-TL reactions. Test: Reactions were incubated at 27 °C and fluorescence kinetics were compared to earlier GFP versions to see how the new variant affected signal speed and intensity. Learn: Using sfGFP provided the clearest ON/OFF difference across all trials. The brighter signal and faster maturation time made sfGFP essential for achieving diagnostic sensitivity and reliable, real-time output. This trial showed that optimizing reporter selection is just as important as refining the toehold design itself.

Our Team in Action: Wet Lab Continuous Gallery

Dry Lab: Computational Modeling & Design Acceleration

Our computational work ran parallel to the wet lab, using NUPACK and RNAstructure to predict structural integrity and inform critical design changes before expensive gBlock synthesis.

Computational Modeling and Validation

NUPACK Predicted Secondary Structure (OFF State)

NUPACK prediction diagram showing the stable hairpin structure of the toehold switch in the OFF state.

Initial model predicting strong secondary structure stability (Delta G = -18.5 kcal/mol).

Trigger vs. Folding Energy Curve

Graph showing the change in free energy (Delta G) of the toehold system as trigger RNA concentration increases, demonstrating the switch mechanism.

Computational prediction of system free energy, confirming the favorable binding with the target sequence.

Computational Insight Leading to Design Fix

The NUPACK simulation for Wet Lab Trial 2 (initial GFP construct) showed a predicted partial opening near the ribosome binding site (RBS) even in the absence of the trigger. This calculated structural 'weakness' immediately informed the wet lab team to incorporate upstream buffer sequences (Wet Lab Trial 3) to physically and thermodynamically isolate the toehold switch from the promoter's influence, directly reducing the experimentally observed leakiness.

Trial 1: Initial Toehold + amilCP Reporter

Design [and Preliminary Steps] Team members researched splice variants associated with various neurological disorders (e.g. BIN1 exon 7 skipping associated with Alzheimer’s, IL7R exon 6 skipping associated with Multiple Sclerosis) and determined the splice variant with the highest confidence → IL7R exon 6 skipping through dataset and literature mining of PubMed and NCBI articles and references. Then, we used GenBank and Geo RefSeq’s databases to extract both the canonical IL7R splice and the variant IL7R splice (exon 6 skipped) in order to template our oligo-RNA trigger sequence. Following this step, we performed robust checks across multiple databases and used Python to ensure our sequence was accurately documented. Using NUPACK and RNAstructure, a toehold switch was designed (after embedding the RBS, i.e. the Shine-Dalgarno sequence) and started codon AUG in a stable hairpin. amilCP sequence was extracted from NCBI and appended to this hairpin to create a complete switch design. The toehold sequence itself was designed as the reverse complementary sequence to the oligo-RNA trigger sequence.

Build Our preliminary toehold–amilCP design was visualized in SnapGene, annotated for trigger-binding sites, RBS, and CDS. No optimization of flanking or buffer sequences was performed yet. A toehold sequence of roughly 24 nucleotides was used, and the hairpin loop included 2 randomized buffer sequences preceding the RBS. BLAST was used to test against cross-reactivity with other non-IL7R trigger sequences in the human transcriptome.

Test (in silico) In order to confirm stable OFF-state hairpin formation i.e. a stable hairpin formed in the absence of the oligo-RNA trigger sequence, NUPACK was utilized. The hairpin was deemed to be stable upon testing with multiple sequences of varying nucleotide lengths (24 base pairs, 36 base pairs, 40 base pairs) at the variant splice junction (Exon 5 → Exon 7) signalled by the Delta G values ranging from -10 to -20 kcal/mol.

Learn This design confirmed the conceptual feasibility of the toehold switch but highlighted amilCP’s slow color development as a poor fit for rapid wet-lab validation. We learned that a fluorescent protein would be better for kinetic testing. Additionally, we recognized that the 24-nucleotide trigger sequence was perhaps too short to create a fully stable hairpin so made key changes in the second trial to rectify this issue.

Trial 2: Toehold + GFP Reporter

Design In place of using amilCP, we extracted the GFP sequence from GenBank (used previously for multiple other established toehold switch designs) for real-time fluorescent readout. The trigger-binding region was extended to improve strand invasion efficiency; we made 2 key changes. The first key change was extending the trigger sequence to be 36 nucleotides; this increased the hairpin stability; 12 nucleotides preceded the stem-loop and 18 were embedded within the stem-loop. Secondly, we removed the buffer sequences preceding the RBS in the loop of the hairpin design and instead used the remaining 6 base pairs (outside of the stem loop) within the hairpin’s main loop to precede the RBS instead.

Build We visualized the new constructs in SnapGene, and RBS-to-start spacing was adjusted to maintain ribosome accessibility once opened.

Test (in silico) We performed NUPACK simulations, which confirmed significantly stronger ON-state opening upon trigger binding, exposing AUG and RBS (increase by 73%). We then verified OFF-state stability with RNAstructure, confirming limited RBS accessibility. Furthermore, NUPACK produced more stable hairpin designs with better defined stem-loop complexes.

Learn Switching to GFP improved theoretical responsiveness for rapid wet-lab validation. However, further analysis suggested potential leakiness from context effects. We recognized the need for upstream buffer sequences to insulate promoter–switch interactions.

Trial 3: GFP with Buffer Sequences

Design In order to reduce theoretical OFF-state leakages, buffer sequences were introduced upstream of the toehold. We adhered to certain conditions for these designs. Firstly, we ensured there were no Shine-Dalgarno sequences or RBS motifs. Secondly, we ensured the absence of start codons. We also maintained the guanine and cytosine content to make up between 40 and 60% of the total base pairs in the buffer sequences to avoid extreme structures. No strong hairpin, terminator, or promoter motifs were included, and we ensured the buffer sequences were low in similarity to host or common plasmids.

Build We wrote robust code in Python to output 5-10 buffer sequence candidates, each of 60 base-pair length, by following certain conditions. Each of the characteristics vital to having appropriate leading and trailing buffer sequences were kept in mind; we encoded them into our Python script (e.g. a line included that no more than 4 of the same base pair can be present consecutively) then outputted 10 completely randomized buffer sequences with 60 base pairs fitting all of these conditions from which we selected 2 for leading and trailing sequence use.

Test (in silico) Upon utilizing NUPACK, it was shown that buffer sequences reduced unintended base-pairing between promoter-proximal nucleotides and the hairpin. The predicted Delta G values for OFF-state became more stable, with reduced probability of partial unfolding.

Learn Insulation improved OFF-state repression without substantially compromising ON-state activation. However, simulations revealed that downstream sequence composition might also affect ribosome accessibility, motivating further changes in subsequent trials.

Trial 4: GFP with Reduced G-rich Trailing Sequence

Design Upon determining that our trailer sequence was G-rich, we identified the design modifications that needed to be implemented. We modified the trailing sequence after GFP to reduce G runs, which risked forming stable secondary structures or G-quadruplexes that stall translation.

Build We added to our script to avoid long homopolymers with low complexity to prevent leakages; a condition was implemented outlining that our buffer sequence may contain no more than 3 consecutive guanine base pairs in any section. The modified constructs were prepared in SnapGene and re-analyzed in NUPACK.

Test (in silico) Secondary structure predictions showed reduced stable downstream structures, with more accessible 3' ends. This was predicted to improve translation elongation efficiency.

Learn Reducing G-rich regions decreased predicted stalling and improved ribosome flow. Nonetheless, we motioned to optimize the GFP variant used to obtain better visual results (stronger fluorescence intensity).

Trial 5: Toehold with Alternative GFP Variant

Design All previous components of the toehold were maintained but the GFP variant was swapped (sfGFP). This was selected for faster maturation and stronger fluorescence intensity. This was once again extracted from an established sequence data based and optimized to fit our toehold switch design.

Build Risk scores were computed using BLAST and NUPACK to ensure no cross reactivity one final time and to ensure stable ON/OFF state structures. Constructs were visualized in SnapGene, and Shine-Dalgarno-to-AUG spacing was confirmed across GFP variants. All previous tools from previous trials’ builds were repurposed for this final trial to aim for robusticity.

Test (in silico) NUPACK confirmed that the toehold operated similarly across variants, with no unintended folding introduced by codon changes. Predicted ON-state exposure remained strong.

Learn Changing the reporter did not alter the switch mechanism but was predicted to enhance measurable output in the wet lab. This reinforced the importance of reporter selection as part of the design cycle. We were able to get significant results with increased observed fluorescence and a successfully stable overall toehold switch design.

Human Practices: Social & Ethical DBTL Cycle

We grounded our design in real-world needs by engaging with experts across Community Health, Clinical Care, and Academic Neurology. Our framework ensured NeuroSplice was not only technically sound but also ethical, transparent, and accessible. This work was conducted under an ethical stakeholder engagement framework.

D

Designing the Framework

Ethics & Format: Used stakeholder feedback to design the project's ethical guidelines (informed consent, anonymity) and communication plan. Designed the final low-cost paper-based format based on community accessibility needs.

B

Building the Assets

Engagement Tools: Built the core community engagement assets, including formal interview protocols, ethical consent forms, and the "My Busy Brain" children's book to educate children about neurobiology.

T

Testing the Feedback

Feedback Integration: Tested the clarity and impact of our planned messaging by coding interview data against five themes and assessing the children's book pilot with educators. This validated gaps in deployment strategy.

L

Learning and Refining

Refining Deployment: Learned that accessibility, clarity, and trust-building are paramount for clinical deployment. Refined the final implementation strategy to prioritize equity and transparency alongside technical precision.

Stakeholder Integration: Feedback to Design Change

The feedback below directly resulted in mandatory protocol and communication changes, moving NeuroSplice from a lab concept to an ethically informed diagnostic tool.

Clinical and Community Usability

Stakeholders: VA Nurse & Community Practitioner

Insight: Loss of trust stems from unclear communication (timelines, rationale) and financial barriers. Also, highlighted technical risk of RNA instability.
Design Change: Developed simple, human-centered explanation templates and timeline disclosures. We reinforced the choice of a low-cost, paper-based, cell-free format for high accessibility. Implemented tighter sample handling assumptions and risk controls.

Academic and Diagnostic Clarity

Stakeholders: Neurology Expert (Dr. Sabatino) & PhD Candidate (Jake Oxendine)

Insight: The tool's function (screening vs. confirmatory) must be defined clearly. Also, highlighted technical risk of biological variability.
Design Change: Adopted stringent validation requirements for repeatability, sensitivity, and specificity. Ensured all outreach materials clarify the result as a "risk signal" not a definitive diagnosis (defining it as a screening tool).

Societal and Cultural Awareness

Stakeholders: Educators (Early Learning Center)

Insight: Affirmed that transparency and accessible language are crucial. The need to promote empathy for neurological differences at an early age.
Design Change: Conducted outreach using the My Busy Brain children's book to teach young audiences about neurobiology, promoting empathy and kindness.

Project Engineering & Iteration

Design, Build, Test, Learn (DBTL) Cycle