Thiol-Ene Chemistry as a Hidden Variable in 3D Radiobiology

When researchers study how tumors become resistant to radiation, they increasingly grow cancer cells inside soft polymer gels rather than on flat plastic, because the gel's stiffness changes how much DNA damage the cells accumulate after a dose. But the reaction used to build many of these gels may not be a neutral bystander. Thiol-ene "click" crosslinking stitches biopolymer chains together by joining sulfur-hydrogen groups with carbon double bonds under light, and it leaves behind precisely the sulfur chemistry that ionizing radiation stirs up: radiation floods the sample with reactive oxygen species, and leftover thiols and sulfur-containing crosslinks can either consume those radicals — shielding the embedded cells — or be cleaved by them, dismantling the scaffold on a timescale of weeks that short assays never observe. The scaffold may therefore both fake a radioresistance phenotype and lose, over time, the very stiffness that produced it.

The problem compounds at the measurement layer. Radiation-response readouts are typically gene-expression measurements normalized against housekeeping genes — internal yardsticks assumed constant — and these yardsticks are documented to drift under exactly the conditions a 3D radiobiology experiment imposes: radiation, hypoxia, matrix mechanics, and long culture times. The strongest existing 3D radiobiology platform, which grew patient-derived glioblastoma cells in methacryloyl-photocrosslinked and enzyme-crosslinked gels, showed that matrix stiffness dominates DNA-damage accumulation but used only non-thiol chemistries, only acute endpoints, and never validated its normalizers (Biomimetic model for the identification of distinctive microenvironmental factors in glioblastoma radiosensitivity). We propose to close all three gaps at once by building a thiol-ene matrix platform whose chemistry is the variable under test.

Core Claim

Thiol-ene crosslinked matrices are active participants in radiobiology, not inert culture substrates. Any measured radiation-response fold-change decomposes as

$$\hat{F}(t) \;=\; F(t)\;\cdot\; A(\rho, D, t)\;\cdot\; B_{\mathbf{r}}(t),$$

where $\hat{F}$ is what the assay reports, $F$ is the true biology we want, $A$ is a chemistry-mediated attenuation factor depending on residual thiol density $\rho$, delivered dose $D$, and elapsed time $t$, and $B_{\mathbf{r}}$ is the normalization error introduced by using reference gene set $\mathbf{r}$. Current practice assumes $A = 1$ and $B = 1$. We will estimate both terms independently, then test three falsifiable hypotheses: that at matched stiffness, thiol-ene matrices blunt acute DNA-damage readouts relative to non-thiol gels in proportion to $\rho$; that irradiation progressively softens the network, so the stiffness-dependent "shield" decays over weeks and long-term survival rankings diverge from acute ones; and that classic housekeeping genes fail across this multi-factor landscape, distorting reported radioresistance-marker changes. The premise that cellular ROS handling shapes radioresistance is established — reactive-oxygen-mediated MAVS oligomerization changes cellular radioresistance (The Reduced Oligomerization of MAVS Mediated by ROS Enhances the Cellular Radioresistance) — but no one has asked whether the scaffold's own sulfur groups feed into the same budget.

Research Program

The platform adapts the silk-fibroin thiol-ene chemistry from a wound-healing hydrogel that withstands 100% strain while keeping cell viability above 95% after three days of co-cultivation (Innovative wound management: creating dynamic Alg-Mg/SF hydrogels for controlled Mg<sup>2+</sup> release in wound healing); we drop the magnesium-coordination component so that the covalent thiol-ene network is the only chemistry in play. Gel storage modulus is tuned to span the reference study's range, from soft matrices to the 6.7 kPa stiff regime, and a methacrylate-photocrosslinked gel at identical modulus serves as the matched non-thiol control, following the GelMA formulation of the glioblastoma work (Biomimetic model for the identification of distinctive microenvironmental factors in glioblastoma radiosensitivity). Residual thiol density is titrated by crosslinker stoichiometry and quantified colorimetrically, giving a $\rho$ dial that is independent of stiffness.

A cell-free radiation-chemistry arm irradiates the gels at 0, 2, and 8 Gy (single and fractionated 2 Gy schedules) and measures radical-scavenging capacity, thioether-link integrity, and serial rheology out to 28 days. This yields $A(\rho, D, t)$ directly, without cells in the loop. A cell arm then embeds the same patient-derived glioblastoma lines used in the reference study (GBM6, GBM8, GBM12, GBM34, GBM39, GBM44, and U87), irradiates at 8 Gy delivered at 6 Gy/min and 0.4 Gy/min under matched oxygen conditions below 5% O₂, and tracks acute γ-H2AX foci (fluorescent markers of DNA double-strand breaks), clonogenic survival, and delayed outgrowth to 28–42 days. The reference study found low dose rate more damaging in radiosensitive EGFRvIII-mutant lines but not in radioresistant ones; we test whether that ranking survives once the scaffold itself softens under accumulated dose. This is where limited follow-up duration enters as a design principle, not an afterthought: the clinical literature documents how truncated windows hide late dynamics, from 120-day lymphocyte tracking that stops before immune recovery (Clinical Considerations of Splenic Dose Constraints to Mitigate Radiation‐Induced Lymphopenia) to ablation cohorts followed 3–6 months that miss late recurrences (A predictive model using left atrial function and B‐type natriuretic peptide level in predicting the recurrence of early persistent atrial fibrillation after radiofrequency ablation); we treat the acute-assay window in 3D radiobiology as the same failure mode.

The normalization arm profiles roughly ten candidate reference genes spanning glycolytic, cytoskeletal, ribosomal, and other classes across the full factorial of chemistry, stiffness, hypoxia, dose, and time, using the standard stability algorithms geNorm, NormFinder, and BestKeeper. We then re-quantify canonical radioresistance markers — SIK2, RAD51, vimentin, CDC6 — that published studies measured against unvalidated housekeeping genes in two-dimensional culture (Salt‐inducible kinase 2 confers radioresistance in colorectal cancer by facilitating homologous recombination repair; ADAR1 enhances tumor proliferation and radioresistance in non-small cell lung cancer by interacting with Rad18), comparing GAPDH-only normalization against a validated panel. Sparse RNA-seq at anchor timepoints cross-checks the panel against rank-based normalization, which sidesteps housekeeping genes entirely by exploiting ordinal expression ranks (Less is more: relative rank is more informative than absolute abundance for compositional NGS data). Finally, a compact translational check reanalyzes the GSE133057 rectal-cancer chemoradiotherapy cohort, in which the SIK2 study correlated expression with overall survival using standard Cox regression, under a censoring-aware survival model that disentangles event risk from dropout risk (Explicitly Modeling Censoring Produces Superior Survival Predictors); no reported result found on Lacuna for the numerical SIK2–overall-survival association in GSE133057, so we compute it fresh under both models.

Baselines

The naive reference point is the field default: two-dimensional monolayer culture, GAPDH normalization, and acute MTT or foci endpoints, as used in the SIK2 radioresistance study (Salt‐inducible kinase 2 confers radioresistance in colorectal cancer by facilitating homologous recombination repair). We run it on the same cell lines and irradiation schedules so the comparison is fair by construction. It is not a serious mechanistic competitor — the reference 3D study already showed stiffness shifts foci formation from 85% to 12%, an effect invisible on plastic — but it defines the magnitude of what the field currently accepts; we beat it only if corrected and uncorrected readouts differ beyond assay error, and if they do not, the default is vindicated, which is itself a reportable result.

The strongest existing competitor is the GelMA/GelFA patient-derived glioblastoma platform, which reported that 85% of cells accumulated γ-H2AX foci after 8 Gy in soft matrices versus 12% in stiff 6.7 kPa matrices, under hypoxia below 5% O₂ and dose rates of 6 and 0.4 Gy/min (Biomimetic model for the identification of distinctive microenvironmental factors in glioblastoma radiosensitivity). We replicate its exact protocol on the same lines, same doses, same dose rates, and same foci-counting window, with our gels matched to its storage moduli and assayed in the same batches. Beating it means two things: at matched 6.7 kPa stiffness, thiol-ene gels produce a foci fraction that deviates from the GelMA value in proportion to residual thiol density, and the deviation pattern at day 28 contradicts the acute one because the network has softened. Its acute-only design cannot, even in principle, detect the second effect.

The key ablation isolates thiol content from everything else: within the thiol-ene family, $\rho$ is titrated at fixed modulus, and a post-crosslinking thiol-capping treatment removes residual thiols without changing mechanics. If the radiation-response shift tracks $\rho$ and vanishes when thiols are capped, the mechanism is sulfur-mediated scavenging rather than any correlated property of the gel. A parallel normalization ablation contrasts single-gene GAPDH against the algorithm-selected panel on identical samples. For the cohort check, the baseline is the Kaplan–Meier and Cox analysis as published for SIK2; the censoring-aware SALaD model already won 109, tied 28, and lost 19 of 156 comparisons across ten datasets including METABRIC (Explicitly Modeling Censoring Produces Superior Survival Predictors), so our bar there is not methodological superiority but whether the SIK2 conclusion itself is robust to the censoring assumption.

Why not just…?

Why not just use 2D plates or commercial Matrigel? The reference 3D study settles the first half: stiffness alone moves DNA-damage accumulation from 85% to 12% of cells, an effect no monolayer can reproduce (Biomimetic model for the identification of distinctive microenvironmental factors in glioblastoma radiosensitivity). Matrigel fails the second half: its batch-to-batch variability makes it impossible to hold stiffness fixed while varying crosslink chemistry, which is the entire point of the experiment. Only a synthetic, defined matrix with an independent dial for residual thiol density can separate the scaffold's chemical effect from its mechanical one.

Why not just normalize to GAPDH like everyone else? The literature on condition-dependent instability is blunt: using unvalidated 18S rRNA across developmental transitions inflated normalized target signals several hundredfold in the peach aphid (Rahman et al.), and even within a single tissue, stability does not transfer across quality cohorts or stressors, requiring validated gene combinations rather than any single gene (Epigenetic-related transcriptional reprogramming elucidated by identification and validation of a novel reference gene combination for RT-qPCR studies in porcine oocytes of contrasting quality). GAPDH is a glycolytic enzyme, and radiation pushes cells toward glycolytic metabolism (Combination treatment with Phloretin enhances the anti-tumor efficacy of radiotherapy in lung cancer models), while ACTB encodes a cytoskeletal protein remodeled by matrix mechanics — both are destabilized by the axes under test. Our factorial sampling measures GAPDH stability directly; if it holds, that is a finding, not a failure.

Why not just run RNA-seq with rank-based normalization and skip qPCR? This is the most serious alternative, because rank-order methods (ROGER and pairwise analysis of gene expression) genuinely bypass housekeeping genes and perform strongly, with diagnostic AUCs of 0.990 and 0.995 on a sepsis benchmark (Less is more: relative rank is more informative than absolute abundance for compositional NGS data). Two limits argue against it as the sole instrument here: sequencing cost makes dense longitudinal sampling across a chemistry-by-stiffness-by-dose-by-time factorial impractical, and rank statistics discard the magnitude and kinetics of expression change, which is exactly what a recovery-trajectory study needs. But we take the threat seriously, which is why the design includes RNA-seq at sparse anchor timepoints: if rank-based and validated-qPCR readouts disagree in sign at any anchor, the simpler route wins and we will say so.

Decisive Checks

The first decisive number is the matched-stiffness comparison: in a 6.7 kPa thiol-ene gel after 8 Gy at 6 Gy/min, does the γ-H2AX foci fraction deviate from the 12% reported for the methacrylate gel at the same stiffness, and does the deviation scale linearly with measured $\rho$ and disappear under thiol capping? A dose-independent, capping-insensitive result would falsify the chemistry-shield hypothesis. The second check is longitudinal: serial rheology must show whether an irradiated 6.7 kPa network decays toward the soft-matrix regime within the culture window, and whether day-28 clonogenic survival across the glioblastoma lines ranks differently from acute foci counts — a rank reversal would demonstrate concretely that acute endpoints mispredict long-term outcome, the in vitro counterpart of the follow-up-truncation bias documented clinically (The Sustainability Gap in Short-Term Clinical Evaluations). The third check quantifies the moving ruler: the fold-error in SIK2, RAD51, vimentin, and CDC6 measurements when GAPDH replaces the validated panel, reported alongside geNorm and NormFinder stability values across the full factorial. The fourth is the cohort flip test on GSE133057: if the SIK2 hazard ratio changes materially between Cox and the censoring-aware model, a published radioresistance biomarker's clinical claim is shown to rest on an assumption the survival literature has already flagged as fragile. If all three in vitro hypotheses fail — matched chemistry, stable mechanics, stable GAPDH — the thiol-ene platform still emerges validated as a mechanically superior, chemically inert scaffold, and the field gains a clean negative result it currently lacks.