Individual Final Project

Casein Metamaterials: Programmable Bio-Actuated Proteins for 4D Fabrication

Aditya Retnanto | ChitownBio | Chicago, Illinois USA | aretnanto@arizona.edu

This writeup was developed with the assistance of Claude (Anthropic) as an AI research and writing collaborator. The author arrived with substantial work already completed: a fully executed Jupyter notebook with Kyte-Doolittle analysis and ESMFold structure predictions across four protein variants; multiple ESMFold PDB structure files; a Twist Bioscience codon-optimized FASTA sequence; a Benchling construct with annotated restriction sites; a pre-written abstract and motivation grounded in Sutherland’s Ultimate Display; three fully articulated project aims; and prior presentation materials from the HTGAA course. The workflow was conversational and iterative: Claude reviewed all of these materials, then prompted the author with targeted questions section by section. The author provided conceptual framing, corrections, and key ideas — including the primitives model for Aim 2, the personal biofabrication ethics framing, the jitter algorithm as the central synthesis challenge, hygroscopic actuation terminology, and the baseline casein bioplastic as the experimental starting point — while Claude structured and articulated those ideas into prose, scaffolded missing sections, and flagged inconsistencies for review. All scientific ideas, experimental results, design decisions, and framing are the author’s own.

Section 1: Abstract

This project advances Ivan Sutherland’s vision of the “Ultimate Display” — where matter is programmable and reactive — by introducing a novel framework for molecular-level 4D printing. While personal fabrication has democratized geometric design, traditional 3D-printed objects remain static and ecologically persistent. We propose a paradigm shift where the “code” of an object is written into the primary sequence of bovine beta-casein proteins.

Our approach involves designing chimeric proteins that alternate between elastic resilin and structural keratin motifs. By using computational tools including ESMFold and Kyte-Doolittle hydrophobicity analysis, we predict the structural integrity of these chimeras and design specific hygroscopic actuation outcomes. This allows for a new form of 4D printing where motion — bending, twisting, or blooming — is programmed at the peptide level and triggered by environmental humidity.

As biomakerspaces continue to rise, this research provides an accessible means for users to move beyond plastic filaments toward sustainable, autonomous bio-materials. The expected outcome is a set of design tools and protein-based mechanisms that allow makers to fabricate objects that are not only functional and reactive but also fully biodegradable. By bridging the gap between high-level computational design and decentralized bio-fabrication, this work opens a new design space at the intersection of synthetic biology and human-computer interaction.

Section 2: Project Aims

Aim 1: Experimental Aim (Course Scope)

The first aim of my final project is to design and express a chimeric bio-actuated protein by utilizing computational protein folding (ESMFold), systematic codon optimization, and heterologous expression in E. coli. This aim encompasses the transition from digital sequence design to physical protein “printing.” Specifically:

• Design: Engineer a β-casein chimera featuring alternating resilin (GGRPSDSYGAPGGGN×6) and keratin (CCQP×8) motifs, inserted at residue 35 and 180 respectively. A custom Python-based systematic codon jitter tool cycles through synonymous codons to eliminate DNA homology and bypass synthesis complexity flags.
• Simulate: Use ESMFold to predict the tertiary structure of the chimera, ensuring hydrophilic elastic domains remain surface-accessible. Kyte-Doolittle hydrophobicity analysis confirms the +48% improvement in cumulative water-binding potential vs. original β-casein.
• Synthesis: Codon-optimize the sequence for E. coli and order the clonal gene via Twist Bioscience in a pET-29b(+) vector using NdeI/XhoI restriction sites. Benchling design
• Expression: Transform the construct into E. coli BL21(DE3) and utilize standard IPTG induction protocols for protein expression and Ni-NTA purification via the His-tag.
• Characterization: Develop a bio-material substrate from the expressed protein and quantify actuation rate and mechanical response under varying humidity conditions.

Aim 2: Development Aim (Post-Course Progression)

Having verified one actuation direction (bending) with the Resilin-casein chimera, Aim 2 extends this into a library of actuation primitives — a finite set of pre-designed chimeric proteins, each encoding a distinct mechanical behavior: bending, twisting, curling, angular control, and temporal control (delayed actuation via degradable crosslinks).

Each primitive would be validated as a standalone film, then made available as a preset “filament” in biomakerspaces. Just as a makerspace stocks PLA and TPU for different geometric needs, it would stock a small set of bio-filaments, each with a known, reproducible actuation signature. A maker selects the primitive that matches the behavior they want, fabricates with it, and gets a predictable output — without needing to understand the underlying molecular design.

The longer-term step is CAD integration: libraries that allow makers to map these primitives onto complex 3D geometries, compositing different actuation behaviors into a single object. A structure could bend at one joint, twist at another, and hold rigid at a third — all from the same biological substrate, just different chimeric sequences.

Aim 3: Visionary Aim (Long-Term Impact)

The long-term vision is decentralized, de novo bio-fabrication — shifting the paradigm from centralized manufacturing to localized, on-demand material synthesis:

• On-the-Fly Material Synthesis: Individuals with access to “biological compilers” — desktop-scale DNA synthesis and microbial expression — could create novel functional materials on demand, bypassing global supply chains.
• De Novo Molecular Design: Moving beyond nature’s templates toward entirely synthetic, de novo protein patterns — mathematically optimizing sequences for specific physical properties rather than biological function.
• The Ultimate Display of Matter: Realizing Sutherland’s vision by transitioning HCI from pixels on a screen to programmable patterning of matter. Objects will be designed to sense, react, and eventually biodegrade, aligning the lifecycle of our technology with the nitrogen and carbon cycles of the natural environment.

Section 3: Background

Literature Context

Lazaro et al. (2025) demonstrated that unmodified bovine casein — a dairy protein readily available in grocery stores — can be cast into films that exhibit reversible hygroscopic actuation, bending in response to humidity gradients. The work showed that the asymmetric distribution of hydrophilic phosphoserine residues in β-casein drives differential swelling across the film, producing curvature without any embedded electronics or external actuators. This established casein as a viable bioplastic for responsive textile fabrication. However, the actuation magnitude and speed are constrained by the native protein sequence, which was not designed with mechanical performance in mind. (Lazaro et al., ACM CHI 2025. doi:10.1145/3714394.3750704)

Elvin et al. (2005) synthesized recombinant pro-resilin — the elastic protein found in insect flight muscles and wing hinges — and demonstrated that crosslinked resilin exhibits near-perfect elastic resilience (>97% energy return) and a highly hydrophilic character driven by its repetitive GGRPSDS motifs. The protein’s exceptional fatigue resistance and water-swelling properties make it an ideal candidate for engineering reversible actuation. Critically, resilin’s elastic modulus can be tuned by varying the number of repeat units, providing a design parameter for programming mechanical response. This work established resilin repeats as modular, synthetically accessible elements for protein-based material engineering. (Elvin et al., Nature 2005. 437:999–1002. doi:10.1038/nature04085)

Novelty

This project is novel in two key respects. First, rather than using casein as-found, it engineers the protein’s primary sequence to amplify its hygroscopic actuation capacity — introducing resilin repeats directly adjacent to the native phosphoserine cluster to create a stronger differential between the actuating and passive domains. Second, by framing the entire workflow — from sequence design to ESMFold prediction to Twist order to E. coli expression — as a generalizable “compiler” pipeline, this work extends synthetic biology tooling into the domain of programmable material fabrication accessible to non-expert makers.

Impact

The problem this project addresses is the ecological and functional limitation of conventional 3D printing: plastic objects that are static, non-degradable, and unable to respond to their environment. This is significant because the personal fabrication movement has democratized geometric design but has not yet democratized material intelligence. A biodegradable, humidity-responsive material that can be programmed at the molecular level would represent a step change in what makers can produce.

Beyond fabrication, the broader societal contribution is the alignment of material lifecycles with biological cycles — casein-based films are fully biodegradable, in contrast to the PLA and ABS plastics that currently dominate personal fabrication. For scientific capability, the chimeric design framework demonstrated here could be extended to any protein with known structural-mechanical relationships, enabling a new class of computationally-designed, biologically-expressed functional materials. If the aims are achieved, this could shift makerspace tooling from filament selection toward molecular sequence design, fundamentally changing the relationship between makers and the materials they work with.

Ethical Implications

The central ethical concern in this project is not material sourcing but the long-term vision of personal biofabrication. To produce meaningful yields of a chimeric protein, expression must occur through in vivo systems — living bacteria that replicate and metabolize in order to manufacture the target protein. In a centralized lab or biomakerspace, this is well-controlled. But the visionary aim of this project imagines a future where individuals have desktop-scale fabricators, raising the question: what happens when bacterial expression systems become as accessible as inkjet printers? The risk is not malicious intent but accidental release — engineered organisms escaping containment, horizontal gene transfer, or poorly maintained cultures producing unintended byproducts. This is a non-maleficence concern at the systems level, not just the experiment level. The principle of responsibility requires that democratization of biofabrication tools be matched by democratization of biosafety literacy.

The primitives framework proposed in Aim 2 is itself an ethical design choice, not just a technical one. By limiting what can be fabricated to a curated set of pre-validated actuation behaviors, the biomakerspace model constrains the design space in a way that reduces risk without eliminating access. A maker choosing from a library of five actuation primitives cannot accidentally design a pathogen — the guardrail is architectural. This mirrors how community labs like ChitownBio already operate: not as open-access DNA synthesis facilities, but as curated environments with institutional oversight, training requirements, and defined scope. The proposed action is to keep protein fabrication within these institutional boundaries during development, expand access incrementally as biosafety understanding matures, and integrate sequence screening (e.g. SecureDNA) into the compiler pipeline before any Twist order is submitted. The key uncertainty is whether this institutional guardrail model can scale — and whether it will remain meaningful as the technology becomes cheaper and more accessible outside formal community lab settings.

Section 4: Experimental Design

Timeline and Experimental Plan

Phase 1 — Baseline Bioplastic (~1 day)

1. Prepare a baseline casein bioplastic using commercially available skimmed milk casein, washing soda (Na₂CO₃, pH ~11 to denature and crosslink), and glycerol as a plasticizer. Cast into a flat film, allow to dry 24h. Expected result: free-standing casein film that bends toward the denser side under humidity.
2. Characterize baseline actuation by placing the film at varying humidity levels (30%, 50%, 70%, 90% RH). Photograph and measure curvature with ImageJ. Expected result: reproducible, quantifiable bending — establishes the control benchmark.

Phase 2 — Chimeric Protein Design (~3–4 days)

3. Retrieve β-casein sequence (UniProt P02666). Identify the phosphoserine cluster (residues 16–35) as the insertion anchor — this is the naturally hygroscopic N-terminal domain.
4. Design the Resilin insert (GGRPSDSYGAPGGGN×6, 90 aa) inserted after residue 35 to amplify the hydrophilic actuating domain. Design the Keratin insert (CCQP×8, 32 aa) inserted at residue 180 to stiffen the passive C-terminal domain via disulfide crosslinks.
5. Run Kyte-Doolittle hydrophobicity analysis on all variants. Compute cumulative water-binding potential to quantify predicted actuation improvement. Expected result: Resilin+Keratin chimera shows ≥40% improvement over original β-casein.
6. Submit all four sequence variants to ESMFold API. Visualize in py3Dmol colored by KD score. Confirm resilin insert remains surface-exposed and disordered. Expected result: low pLDDT (<50) in resilin region, consistent with intrinsically disordered protein character.

Phase 3 — DNA Design & Ordering (~2 days + ~1 week turnaround)

7. Open Benchling, import the chimeric amino acid sequence, and annotate insert regions. Add NdeI (CATATG) restriction site at the 5′ end and XhoI (CTCGAG) at the 3′ end for directional cloning into pET-29b(+). Confirm no internal NdeI/XhoI sites exist within the chimera sequence.
8. Export sequence to Twist Bioscience codon optimizer. Run codon optimization for E. coli BL21(DE3) expression. Apply systematic synonymous codon cycling across the resilin repeats to minimize DNA homology flags. Expected result: 909 bp construct, no complexity warnings.
9. Submit Twist order: 909 bp gene in pET-29b(+) with C-terminal His-tag. Estimated turnaround: 1 week.

Phase 4 — Protein Expression & Film Casting (~1–2 weeks)

10. Transform pET-29b(+) construct into E. coli BL21(DE3) via heat shock. Plate on LB + kanamycin and pick colonies. Verify that the correct construct was incorporated using standard sequence confirmation methods. Expected result: confirmed colonies carrying the chimeric construct.
11. Grow confirmed culture and induce protein expression with IPTG. Lyse cells and confirm the chimeric protein is present at the expected size. Purify the protein using the His-tag affinity handle built into the pET-29b(+) vector.
12. Cast chimeric protein films using the same washing soda/glycerol protocol as Phase 1. Allow to dry 24h. Expected result: chimeric films show greater curvature than baseline casein films at matched humidity conditions.

Workflow Diagram

Figure: Experimental workflow from sequence design to material characterization. (diagram pending)

Techniques Checklist

Relevant techniques applied in this project:

• ✅ Bioethical Considerations
• ✅ Protein Design (ESMFold, KD hydrophobicity analysis)
• ✅ Databases (UniProt, NCBI)
• ✅ DNA Construct Design
• ✅ Designing a Twist Order
• ✅ Restriction Enzyme Digestion (NdeI/XhoI cloning sites)
• ✅ Chassis Selection (BL21(DE3))
• ✅ Plasmid Preparation
• ✅ Bacterial Culturing
• ✅ Protein Purification
• ✅ Pipetting / Lab Safety
• ✅ Use of Benchling
• ✅ Models and Notebooks (Python, Jupyter)

Two Techniques Expanded

Protein Design (ESMFold + Kyte-Doolittle Analysis): ESMFold was used to predict the tertiary structure of all four protein variants — original β-casein, Resilin-casein, Keratin-casein, and the combined chimera — via the ESM Atlas API. Each sequence was submitted as a plain amino acid string and returned as a PDB file, which was visualized in py3Dmol with residues colored by KD hydrophobicity score. This allowed visual confirmation that the resilin insert (highlighted in green) remained surface-exposed and disordered, consistent with its expected intrinsically disordered protein (IDP) character. The KD analysis quantified the impact of each modification: the Resilin insert (avg KD = −1.21) increased cumulative water-binding potential by +36%, while the combined chimera achieved +48% over the original sequence.

DNA Construct Design (Systematic Codon Jitter): A custom Python tool was developed to translate the chimeric amino acid sequence into a DNA sequence optimized for E. coli expression while minimizing homology across the six tandem resilin repeats. Rather than selecting the single most-preferred E. coli codon for each amino acid (which would produce near-identical DNA for each of the six GGRPSDSYGAPGGGN repeats), the tool cycles through a table of synonymous codons in order, ensuring each repeat uses a different DNA “spelling.” The resulting 909 bp sequence was further refined using Twist’s own codon optimization engine and flanked with NdeI (CATATG) and XhoI (CTCGAG) restriction sites for directional cloning into pET-29b(+).

Industry Council Companies

• Twist Biosciences — gene synthesis and codon optimization of the 909 bp chimera construct
• Benchling — sequence design, annotation, and construct visualization

Section 5: Results & Computational Validation

Validation Approach

The computational design pipeline was validated through three complementary analyses: Kyte-Doolittle hydrophobicity profiling of all four sequence variants, cumulative water-binding potential comparison, and ESMFold tertiary structure prediction. Together these confirm that the engineered chimera has substantially higher predicted actuation capacity than native β-casein, and that the DNA sequence was successfully designed and ordered via Twist Bioscience.

Protocol

1. Retrieved β-casein sequence (P02666) from UniProt via API call in Python
2. Designed four variants: original, Resilin-casein (+90 aa resilin insert at position 35), Keratin-casein (+32 aa CCQP insert at position 180), and combined chimera
3. Computed per-residue Kyte-Doolittle scores for all variants using the standard KD scale
4. Computed cumulative water-binding potential as the running sum of max(-KD, 0) across all residue positions
5. Generated 11-residue sliding window smoothed hydrophobicity profiles for visual comparison
6. Submitted all four sequences to ESMFold API; saved PDB files for visualization
7. Implemented systematic codon jitter algorithm cycling through synonymous E. coli codons
8. Generated 909 bp Twist-ready DNA sequence; confirmed NdeI/XhoI sites present, no internal cut sites
9. Submitted sequence to Twist Bioscience codon optimizer; ordered construct in pET-29b(+)

Synthetic Biology Techniques Utilized

The protein design workflow used ESMFold (a protein language model-based structure predictor) to generate structure predictions for all variants without requiring experimental X-ray crystallography or NMR data. KD hydrophobicity analysis applied a standard bioinformatics scoring matrix to quantify water-binding capacity — a proxy for hygroscopic actuation potential. DNA construct design used systematic synonymous codon cycling to address a fundamental challenge in gene synthesis: repetitive sequences in tandem repeat proteins cause DNA synthesis failure due to homology-mediated polymerase slippage. The Twist Bioscience order incorporated NdeI and XhoI restriction sites to enable directional ligation into pET-29b(+), a T7 promoter-driven bacterial expression vector with a C-terminal His-tag for affinity purification.

Results

Figure 1: Kyte-Doolittle hydrophobicity profiles for β-casein (original, 224 aa) and Resilin+Keratin-casein chimera (300 aa). Blue regions = hydrophilic; orange = hydrophobic. Resilin insert region (purple shading) shows strongly hydrophilic character (avg KD = −1.21). Keratin insert region (orange shading) introduces Cys residues for disulfide crosslinking. Red dashed lines mark phosphoserine positions.

Figure 2: Per-residue KD scores for the Resilin insert (×6, avg KD = −1.21) and Keratin insert (×8, avg KD = −0.02). Dotted lines mark repeat boundaries. The Resilin insert is predominantly hydrophilic; the Keratin insert alternates hydrophilic/hydrophobic residues to enable disulfide-mediated crosslinking while maintaining solubility.

Figure 3: Cumulative water-binding potential across residue positions for all variants. Resilin+Keratin-casein achieves a total of 493 vs. 332 for original β-casein (+48%). The shaded blue region marks the resilin insert, where the chimera diverges most sharply from the original.

Summary of quantitative results:

The Resilin+Keratin chimera is predicted to achieve maximum bending due to the largest differential between the highly hydrophilic actuating domain (resilin insert + N-terminal phosphoserine cluster) and the stiffer passive domain (keratin insert with 17 Cys crosslink sites in the C-terminal region).

Challenges

The primary challenge was DNA synthesis complexity arising from repetitive sequences. Inserting off-the-shelf motifs like the resilin repeat (GGRPSDSYGAPGGGN×6) directly produces near-identical DNA across all six copies — the same amino acid sequence naively translates to the same codons, creating long stretches of homologous DNA that gene synthesis platforms flag as too complex to manufacture reliably. The systematic codon jitter algorithm was developed specifically to address this: by cycling through synonymous codons for each amino acid, each repeat gets a slightly different DNA spelling even though it encodes the same protein sequence. This eliminated the complexity flags and produced a Twist-orderable 909 bp construct.

A secondary limitation is that the KD-based actuation prediction assumes all residues are equally surface-accessible. In a cast protein film, some hydrophilic residues may be buried in aggregated regions and not contribute to water uptake. Future validation requires casting and testing actual films to confirm whether the computational +48% improvement in water-binding potential translates to a measurable increase in curvature.

Section 6: Additional Information

References

• Lazaro, E. et al. (2025). “Bio-actuated Textiles.” ACM CHI 2025. doi:10.1145/3714394.3750704
• Elvin, C.M. et al. (2005). “Synthesis and properties of crosslinked recombinant pro-resilin.” Nature 437:999–1002. doi:10.1038/nature04085
• UniProt Consortium. β-casein (P02666). UniProtKB. https://www.uniprot.org/uniprot/P02666
• Smialowski, P. et al. (2012). PROSO II — a new method for protein solubility prediction. FEBS Journal 279(12):2192–2200.
• Kyte, J. & Doolittle, R.F. (1982). A simple method for displaying the hydropathic character of a protein. Journal of Molecular Biology 157(1):105–132.

Supply List and Budget

Group Final Project