Programming Moisture-Driven Depot Degradation to Open the SPP1 Resistance Niche

Injectable polymer depots — soft, water-rich gels loaded with drugs — are widely proposed to deliver immunotherapy locally to tumors that stop responding to checkpoint inhibitors. But almost everything we know about how these depots fall apart in water comes from soaking them in a still, pH-neutral buffer on a bench. Two literatures that never talk to each other say this is exactly wrong. Materials engineers have spent decades showing that water is an aggressive chemical actor: it plasticizes polymers by prying apart the hydrogen bonds between chains, it strips hydrophobic binders off hydrophilic filler surfaces, and it can catalyze runaway chain scission that collapses a polymer's molecular weight from 126.7 kDa to 1.4 kDa in a single processing step (Preparation of Calcium Phosphate-Biodegradable Polymer Composites with Ion-Releasing Ability for Enhancing Bone Formation). Meanwhile, spatial-omics studies of liver cancer have shown that immunotherapy fails for a physical reason: non-responding tumors wall off their lymphoid aggregates — clusters of immune cells — behind an organized, SPP1-rich stromal boundary, so the immune cells never touch the tumor (478 Spatial omics and deep learning define an immunotherapy resistance niche in hepatocellular carcinoma).

The proposal here is to treat moisture-induced interface degradation not as a failure mode to be suppressed but as the actuator of the therapy. We will build an injectable depot whose water-driven swelling, interfacial stripping, and hydrolysis chemistry are deliberately tuned so that, as it absorbs tumor interstitial fluid, it mechanically and chemically opens that SPP1 barrier and buffers the local microclimate — and we will evaluate it with spatial, in vivo endpoints rather than the bulk tumor-shrinkage numbers that hide whether the resistance niche actually came apart.

Core Claim

A biodegradable composite depot can be engineered at the filler–matrix interface so that moisture ingress — the same water uptake that slashes the yield strength of water-saturated polyamide composites by more than 15% (Strain Rate Dependence of Compressive Mechanical Properties of Polyamide and Its Composite Fabricated Using Selective Laser Sintering under Saturated-Water Conditions) — produces a programmed sequence in vivo: swelling pressure that permeates the stromal exclusion boundary, followed by controlled hydrolysis whose pH microclimate is held near physiological by magnesium-substituted filler chemistry, followed by release of Ca²⁺/Mg²⁺ ions and a checkpoint-sensitizing payload. The measurable claim is that this degradation-programmed depot converts "interfaced" (wall-sequestered) lymphoid aggregates into "exposed" ones at higher rates than an otherwise identical depot whose degradation is unprogrammed, and that this conversion, not bulk drug release, is what predicts tumor response. This directly attacks the translation gap: the degradation kinetics that determine efficacy will be characterized under dynamic interstitial flow and mechanical shear, not static sink buffer, and the primary efficacy endpoint is a spatial-architecture metric validated on human clinical biopsies rather than an in vitro cytotoxicity number.

The closest existing work is the FFC injectable hydrogel, a Pluronic/chitosan gel co-encapsulating a copper-based cuproptosis inducer and glucose oxidase, which achieved over 80% tumor volume regression as a monotherapy and over 90% with systemic anti-PD-L1 in 4T1 tumors in mice (Clinically friendly smart hydrogel boosts cuproptosis and PD-L1 upregulation to enhance anti-tumor immunotherapy). That work exploits bulk thermal gelation and pH-driven erosion, but it never characterizes the interfacial processes — water displacement of the drug–matrix interface, filler-catalyzed hydrolysis, degradation-product acidification — and it never asks whether the depot changes the spatial relationship between immune aggregates and tumor cells. Hydrogel-vaccine reviews make the same omission, flagging "uncontrolled burst release caused by moisture-induced structural erosion" as a bottleneck while assuming released agents uniformly reach their targets (Hydrogel-Based Vaccines: A Promising Approach for Cancer Immunotherapy). We take the opposite stance: the moisture-induced processes are the design variables, and spatial immune architecture is the endpoint.

Research Program

The program has three coupled parts.

First, a degradation-engineering axis. We synthesize polyester-based composite depots with bioactive-glass fillers of matched composition except for the alkali species, following the Na-for-Mg substitution that the bone-composite work showed to be decisive: sodium-bearing glass catastrophically catalyzes hydrolysis (matrix molecular weight collapsing from 126.7 kDa to 1.4 kDa, soak pH below 5), while the magnesium analogue retains 71.3 kDa, holds Tris buffer near pH 7.4, and releases about 90% of its Ca²⁺/Mg²⁺ with silicate release throttled to roughly 7% (Preparation of Calcium Phosphate-Biodegradable Polymer Composites with Ion-Releasing Ability for Enhancing Bone Formation). We quantify each formulation with the field's own instruments: gel permeation chromatography for molecular weight, surface-energy measurements of the thermodynamic work of adhesion and stripping at the filler–matrix and matrix–drug interfaces, in the spirit of the tuff-aggregate surface-energy analysis (Water Damage Resistance of Tuff Asphalt Mixtures with Admixtures Based on Surface Energy Theory), and gravimetric/chemo-mechanical tracking under three conditioning regimes of escalating realism: static Tris-HCl sink, a bioreactor imposing interstitial flow and cyclic compressive shear on a collagen-rich stroma (justified by the >15% yield-strength loss that water saturation alone inflicts on polyamide under load, Strain Rate Dependence of Compressive Mechanical Properties of Polyamide and Its Composite Fabricated Using Selective Laser Sintering under Saturated-Water Conditions), and implanted tumors. The first hypothesis to test is that static-sink release curves mispredict dynamic and in vivo release — a direct instantiation of the translation gap. The electrospun peptide-membrane work supplies the cautionary precedent: most of its active peptide dissolved away within 48 hours of water exposure, a burst failure invisible in dry characterization (Composite Membranes of Recombinant Silkworm Antimicrobial Peptide and Poly (L-lactic Acid) (PLLA) for biomedical application).

Second, an interface-niche axis. We build a 3D co-culture in which SPP1-expressing stromal cells and macrophages form a boundary layer around lymphoid aggregates, mimicking the resistance niche that the spatial-omics study defined in human hepatocellular carcinoma, where a graph neural network on Akoya Phenocycler and 10x Visium data classified exposed versus interfaced aggregates with an AUROC of 0.93 in pre-treatment biopsies, and where the same signature appeared in non-responding breast cancer (478 Spatial omics and deep learning define an immunotherapy resistance niche in hepatocellular carcinoma). Our primary in vitro endpoint is the exposed-aggregate fraction as a function of depot degradation state, cross-referenced against payload-only and swelling-only controls to separate mechanical permeation from chemical effects. For prioritizing payload candidates we use TISMO, a consolidated syngeneic mouse database of 2,123 RNA-seq samples (605 in vitro, 1,518 in vivo; 832 from immune-checkpoint-blockade studies; 49 cell lines, 68 tumor models, 19 cancer types) (TISMO: syngeneic mouse tumor database to model tumor immunity and immunotherapy response), asking which degradation-responsive pathways — acidosis response, SPP1 expression, ICOS signaling — track with checkpoint response in vivo, so the in vitro arm never floats free of in vivo evidence.

Third, an in vivo axis in the 4T1/BALB-c setting used by the FFC study (Clinically friendly smart hydrogel boosts cuproptosis and PD-L1 upregulation to enhance anti-tumor immunotherapy), with spatial readouts at multiple time points past day 21, where the FFC study stopped. One arm must confront the counter-regulatory risk head-on: depot-induced immunogenic stress could feed the very feedback loop in which activated CD8⁺ T cells secrete IL-2 that stimulates immunosuppressive ICOS⁺ regulatory T cells, the mechanism implicated in melanoma PD-1 resistance (455 Treg-associated ICOS as a possible biomarker for PD-1 therapy resistance in melanoma patients). If ICOS⁺ Treg frequency rises in depot-treated tumors, an ICOSL-blockade combination arm resolves whether the degradation microclimate helps or backfires.

Baselines

The naive reference point is untreated tumor growth and systemic bolus anti-PD-L1 alone in the same 4T1 model; the file reports no number for either reference arm on Lacuna, so we will establish them ourselves under identical conditions rather than quote one. The strongest existing competitor is the FFC injectable hydrogel of Pluronic F127/F68 and chitosan carrying the cuproptosis inducer and glucose oxidase (Clinically friendly smart hydrogel boosts cuproptosis and PD-L1 upregulation to enhance anti-tumor immunotherapy), which reported over 80% tumor volume regression as monotherapy and over 90% combined with systemic anti-PD-L1 in subcutaneous 4T1 tumors in female BALB/c mice. We will run it head-to-head in the same model, same mouse strain and tumor line, same anti-PD-L1 dose and schedule, matched depot mass and payload loading, and an equal material-synthesis and formulation-optimization budget, so any difference is attributable to degradation programming rather than effort. Beating it means matching or exceeding the >90% combination regression while also achieving something it never tested: sustained, controlled release and intact spatial efficacy beyond day 21, a significantly increased exposed-aggregate fraction on spatial imaging, and no local pH excursion below physiological — with the acidifying sodium-glass variant serving as the kill-criterion demonstration. The key ablation is an architecturally identical depot whose filler chemistry is left unprogrammed: either the sodium-bearing analogue (predicted to acidify below pH 5 and collapse molecular weight, per the bone-composite results) or an inert, non-ion-releasing filler. If the programmed and unprogrammed depots deliver the same payload at the same total dose but differ in exposed-aggregate conversion and response, the causal ingredient is the degradation program, not local delivery per se. A secondary analytical baseline is the MIF-GNN spatial classifier at AUROC 0.93 on pre-treatment biopsies (478 Spatial omics and deep learning define an immunotherapy resistance niche in hepatocellular carcinoma): our exposed/interfaced quantification pipeline must reach comparable classification accuracy on pre-treatment samples before its post-treatment measurements can be trusted.

Why not just…?

Why not just give systemic checkpoint inhibitors plus adjuvants? For a start, systemic delivery cannot address a physical problem: the resistance niche excludes effector cells geometrically, and antibodies do not remove walls (478 Spatial omics and deep learning define an immunotherapy resistance niche in hepatocellular carcinoma). Clinically, patients with the stromal tumor-microenvironment class did worse on systemic therapy in the multi-cohort biology-guided deep-learning study (Biology-guided deep learning predicts prognosis and cancer immunotherapy response), and systemic adjuvant co-administration carries the immune-related adverse-event burden that hydrogel confinement was designed to reduce (Hydrogel-Based Vaccines: A Promising Approach for Cancer Immunotherapy). That said, this alternative is genuinely adequate for tumors without a spatial-exclusion phenotype, and we will report the systemic arm honestly so the proposal's scope is limited to stroma-dense, non-responsive indications.

Why not just use the existing smart hydrogel? This is the real threat, and we treat it as the head-to-head baseline rather than arguing around it. The FFC depot already achieved over 90% regression with systemic anti-PD-L1 in the exact model we use (Clinically friendly smart hydrogel boosts cuproptosis and PD-L1 upregulation to enhance anti-tumor immunotherapy). Two things distinguish us, and both are testable rather than rhetorical: FFC never measured whether its depot changed immune-cell spatial architecture, so its efficacy may or may not have involved niche opening — our exposed-aggregate endpoint will show whether it does; and FFC never characterized interfacial degradation past 21 days or under tumor-mechanical load, so it is unknown whether its release is sustained or collapses into the burst-release failure mode documented for other hydrated depots (Hydrogel-Based Vaccines: A Promising Approach for Cancer Immunotherapy). If the FFC depot matches us on every spatial and durability endpoint, the contribution collapses to a measurement study — and we will say so.

Why not just discover resistance-reversing molecules with 2D screens, or just predict responders with the existing deep-learning models? Both are cheaper, and both are insufficient for the same reason: they observe or exploit a static system rather than change it. Static 2D cultures miss the vascular and spatial determinants of response — reflectance confocal phenotyping of living lesions found that response depends on a combined inflammation/vasculature phenotype at 76% accuracy, with high-vasculature lesions failing despite inflammation (In vivo tumor immune microenvironment phenotypes correlate with inflammation and vasculature to predict immunotherapy response) — and they miss multi-cell feedback loops such as the IL-2/ICOS Treg backfire (455 Treg-associated ICOS as a possible biomarker for PD-1 therapy resistance in melanoma patients). The imaging models are excellent at prediction and powerless at intervention: the biology-guided deep-learning model predicts immunotherapy objective response with an AUC of 0.753, outperforming the approved PD-L1 Combined Positive Score (Biology-guided deep learning predicts prognosis and cancer immunotherapy response), but a patient whose aggregates are walled off still needs something that removes the wall. We use these predictors as stratification and endpoint tools, not substitutes.

Decisive Checks

Three results would make or break the proposal. The translation-gap check: static-sink release curves must be shown to mispredict in vivo release for at least one formulation — if static and dynamic kinetics agree, the degradation-engineering premise weakens and the work reduces to standard delivery. The mechanism check: exposed-aggregate fraction must rise in the degradation-programmed depot relative to the unprogrammed ablation at matched payload dose, and that rise must mediate response in a mediation analysis — if regression occurs without niche opening, the spatial story is decorative. The safety check: local microclimate pH must stay near physiological in vivo, with the sodium-glass analogue's acidification (pH below 5, molecular-weight collapse, per Preparation of Calcium Phosphate-Biodegradable Polymer Composites with Ion-Releasing Ability for Enhancing Bone Formation) serving as the explicit negative control, and ICOS⁺ Treg frequency must not increase — if it does, the ICOSL-blockade arm decides whether the concept is salvageable. A final translational sanity check applies the same spatial-exposure analysis to human pre-treatment biopsies of a stroma-rich tumor type, using the HCC cohort's 0.93 AUROC as the analytical benchmark, to ask which patients this depot is actually for.