AND-gated protease-activated nanosensors for programmable detection of anti-tumor immunity

Abstract

The forward design of biosensors that implement Boolean logic to improve detection precision primarily relies on programming genetic components to control transcriptional responses. However, cell- and gene-free nanomaterials programmed with logical functions may present lower barriers for clinical translation. We report the design of activity-based nanosensors that implement AND-gate logic without genetic parts via bi-labile cyclic peptides. These actuate by releasing a reporter if and only if cleaved by a specific pair of proteases. AND-gated nanosensors that detect the concomitant activity of the granzyme B protease secreted by CD8 T cells and matrix metalloproteinases overexpressed by cancer cells identify the unique condition of cytotoxic T cell killing of tumor cells. In preclinical mouse models, AND-gated nanosensors discriminate tumors that are responsive to immune checkpoint blockade therapy from B2m^–/– tumors that are resistant to it, minimize signals from tissues without co-localized protease expression including the lungs during acute influenza infection, and release a reporter locally in tissue or distally in the urine for facile detection.

Main

Advances in synthetic biology have historically focused on the genetic circuit paradigm to assemble sense-and-respond biocircuits operating under transcriptional regulation^1,2,3. Sophisticated functions have been implemented in prokaryotic and eukaryotic cells with applications across cell therapies, drug delivery, molecular imaging, and biosensors. For example, bacteria have been genetically engineered to detect clinically useful biomarkers^4,5 or produce synthetic biomarkers in biofluids^6,7,8. In mammalian cells, tumor-antigen sensing circuits that employ AND-, OR-, or NOT-gate logic have enabled programmable control of T cell responses to reduce toxicity^9,10,11.

While the gene circuit analogy will remain at the forefront of development, cell- and gene-free biocircuits are gaining attention for their potential to lower barriers for clinical translation while retaining the ability to perform logical functions. For example, RNA-based switches that regulate protein translation in response to cell-type-specific biomarkers^12,13 or small molecule inducers^14,15 could be delivered via lipid nanoparticles without genetic engineering. Similarly, protein circuits that do not primarily rely on transcriptional regulation have been engineered as logic gates, bandpass filters, and regulatory cascades for antigen sensing⁹, protein secretion^16,17, and activation of enzymes and transcription factors^18,19. Recent work also highlights the growing potential of synthetic nanomaterials that, despite being cell- and gene-free, can apply logic to increase detection precision, often using cell-secreted proteases. These include hairpin peptide prodrugs that digitize drug delivery based on protease activity^20,21, imaging probes such as ‘hub-and-spoke’ fluorogenic peptides that require proteolysis of multiple quenchers to highlight tumor margins²², and activity-based sensors that utilize protease promiscuity as an analog framework to classify biological samples^23,24.

Here, we describe a cell- and gene-free design of AND-gated nanosensors that require a pair of proteases for activation, increasing specificity. AND-gated nanosensors are implemented using asymmetric bi-labile cyclic peptides that release reporters if and only if both flanking substrates are cleaved. The cyclic peptides are multivalently displayed on iron oxide nanoparticles (IONPs) to improve catalytic efficiency and tissue retention. In mouse models, we demonstrate that AND-gated nanosensors activated by a combination of matrix metalloproteinases (MMPs) in tumors and granzyme B (GzmB) secreted by cytotoxic T cells can selectively detect anti-tumor responses during immune checkpoint blockade therapy (ICBT) while minimizing activation from off-tumor sites including the lungs during acute viral infection.

Cyclic peptide nanosensors implement AND-gate logic

We postulated that a bi-labile cyclic peptide design could implement AND-gate logic by releasing a reporter only when both substrates are cleaved (Fig. 1a). To test this design, we synthesized a GzmB-selective linear peptide (IEFDSG^25,26) and a cyclic peptide containing two substrate copies. Both constructs used a FRET pair (fluorophore: 5(6)-FAM, quencher: TQ2) for monitoring proteolysis by fluorimetry (Supplementary Fig. 1). Incubation with GzmB rapidly increased fluorescence of the linear substrate (Supplementary Fig. 2a), resulting from cleavage based on the appearance of two mass fragments (Supplementary Fig. 2b). By contrast, whereas GzmB digestion of the cyclic peptide resulted in rapid fluorescence, mass analysis revealed two fully cleaved fragments and a third species corresponding to a linearized form of the cyclic peptide (shift of +18 Da; Supplementary Fig. 2c,d), which we attributed to cleavage of a single arm of the peptide. To further test that dual substrate cleavage is required for reporter release, we replaced either substrate with an isoform using d-amino acids to prevent proteolysis. This significantly reduced fluorescence compared to the fully labile l-amino acid cyclic peptide (Supplementary Fig. 2e,f), further confirming reporter release requires cleavage of both substrates.

Figure 1. Asymmetric bi-labile cyclic peptide nanosensors implement AND-gate logic.

a, Design of AND-gated nanosensors, consisting of iron oxide nanoparticles decorated with asymmetric bi-labile cyclic peptides that release a fluorescent reporter only when both substrates are cleaved, thereby actuating AND-gate logic. b-c, Fluorescence upon treatment of quenched linear substrates for granzyme B (b, GzmB; substrate: IEFDSG) and matrix metalloproteinases (c, MMP; substrate: PAALRA) with recombinant GzmB or MMP8 for 15 minutes (one-way analysis of variance (ANOVA) with Dunnett’s post-test and correction for multiple comparisons). d, Mass spectral analysis by MALDI-MS (matrix-assisted laser desorption ionization mass spectrometry) of an asymmetric bi-labile cyclic peptide bearing substrates for GzmB and MMPs (i.e., an AND-gated peptide) confirms production of a cleaved reporter fragment (1407.6 Da) only after treatment with both GzmB and MMP8. e-f, Representative kinetic curve (e) and fluorescence at 2 hours (f) for quenched fluorescent AND-gated nanosensors treated with GzmB and/or a panel of MMPs (e, two-way ANOVA with Dunnett’s post-test and correction for multiple comparisons; f, one-way ANOVA with Tukey’s post-test and correction for multiple comparisons). (e, **P=0.0031; a-f, ****P<0.0001, n=3 technically independent wells, error bars depict mean ± s.e.m.; ns, not significant; RFU, relative fluorescent units)

To confirm asymmetric cyclic peptides can perform AND-gate logic, we synthesized a cyclic peptide containing the GzmB substrate and the substrate APAALRAA, whose linear peptide had broad activity for MMPs and negligible activity for GzmB and other off-target proteases (Fig. 1b,c, Supplementary Fig. 3). We incubated this FRET-labeled cyclic peptide with GzmB, MMP8, or both proteases. Both proteases were required to produce a reporter detectable by mass spectrometry or fluorescence (Fig. 1d, Supplementary Fig. 4), confirming AND-gate logic. Given the short circulation half-lives of free peptides^27,28, we formulated AND-gated nanosensors by conjugating cyclic peptides to IONPs – which can extend circulation and reduce renal clearance of peptides^6,25 – using PEGylated crosslinkers to limit uptake by the reticuloendothelial system^25,29. Peptide conjugation to IONPs resulted in increased particle size and an absorbance peak at ~500 nm due to the FRET pair (Supplementary Fig. 5). Using this design, we confirmed that AND-gated nanosensors produced highest reporter signals when GzmB activity was concomitant with MMPs (Fig. 1e,f). We further evaluated the modularity of our design by substituting the MMP substrate with fPRGS³⁰, a thrombin-responsive sequence, or replacing the GzmB substrate with substrates cleavable by legumain (AAN³¹) or fibroblast activation protein (ASGPAGPA³²). Nanosensors using all substrate combinations successfully implemented AND-gate logic (Supplementary Fig. 6,7).

Multivalent presentation improves efficiency of proteolysis

Whereas restricting the conformation of peptides via cyclization reduces susceptibility to proteolysis^33,34, multivalent presentation on a nanoparticle increases local substrate concentrations and can enable enzymes to “hop” between substrates on a nanoparticle^35,36, increasing cleavage rates^25,35. We therefore evaluated differences in the catalytic efficiencies (kcat/Km) of proteolysis between the symmetrical GzmB-cleavable cyclic peptide and its linear form. We determined catalytic efficiency based on release of the fluorescent reporter (i.e., the cyclic peptide requires cleavage of two substrate copies). We found that the catalytic efficiency of GzmB for the cyclic peptide was ~85.2% less than the linear substrate (6.2 versus 41.9 M⁻¹s⁻¹). However, conjugating peptides to IONPs to create multivalent constructs with ~40 peptides per IONP increased the catalytic efficiency by ~21.7-fold for the cyclic peptide compared to ~6.6-fold for the linear peptide (Fig. 2a,b). This substantially reduced the difference in kinetics between the cyclic and linear peptides to ~51% (134.5 versus 276.2 M⁻¹s⁻¹), indicating multivalent presentation of cyclic peptides can result in larger increases in proteolytic efficacy compared to linear peptides.

Figure 2. Multivalent presentation of AND-gated peptides improves proteolysis kinetics.

a-b, Michaelis-Menten analysis of rate of GzmB catalysis for linear and symmetric bi-labile cyclic peptides when free in solution (a) or multivalently presented on iron oxide nanoparticles (b). c, Initial GzmB cleavage velocity for nanoparticles (NPs) labeled with varying numbers of symmetric bi-labile cyclic peptides per NP. d-e, Dose dependence of GzmB-mediated cleavage fluorescence for nanosensors displaying linear (d) and symmetric bi-labile cyclic (e) peptides. Dashed line indicates detection of statistically significant signal relative to blank. f-g, Cleavage fluorescence (f) and Michaelis-Menten analysis (g) for AND-gated nanosensors displaying asymmetric bi-labile cyclic peptides when cleaved by varying concentrations of GzmB and/or MMP9. (a-g, n=3 technically independent wells, error bars depict mean ± s.e.m.)

To identify an optimal surface valency, we synthesized nanosensors with valencies ranging from 2 to 197 peptides per IONP (Supplementary Fig. 8a,b). We observed that initial cleavage velocities increased with valency to a maximum of ~44 peptides per IONP after which further increases in valency decreased cleavage kinetics, likely from steric effects of overcrowding (Fig. 2c, Supplementary Fig. 8c,d). We therefore selected a valency of ~44 for future studies. To quantify the limit of detection, we exposed sensors to decreasing concentrations of GzmB. Both linear and cyclic peptide nanosensors exhibited dose-dependent responses with lowest detectable signals at 1.6 nM GzmB (Fig. 2d,e). Finally, we tested the limit of detection of the GzmB/MMP AND-gated nanosensor and found that reporter fluorescence was dependent on both proteases, with detectable signal at as low as 3.5 nM GzmB and 10 nM MMP9 (Fig. 2f,g). This data suggests AND-gated nanosensors retain fast activation rates and high analytical sensitivity to dual protease inputs.

AND-gated nanosensors detect T cell killing of tumor cells

We next evaluated whether AND-gated nanosensors can differentiate physiologically unique conditions using cell-secreted proteases. Upon recognition of cells expressing a cognate antigen, CD8+ T cells release effector molecules including GzmB that initiate a cascade of events leading to target cell death^37,38. Given the ubiquity of MMP secretion by cancer cells to promote angiogenesis, tumorigenesis, and metastasis³⁹, we hypothesized that AND-gated nanosensors requiring both MMP and GzmB would respond only to the condition where T cells are killing cancer cells and not when only tumor cells or T cells are present.

To test this, we collected media conditioned by either activated murine CD8+ T cells or MC38 colorectal cancer cells (Supplementary Fig. 9a), confirmed the presence of secreted GzmB or MMP9 by ELISA (Supplementary Fig. 9b,c), and validated orthogonal activation of the GzmB and MMP linear substrates (Supplementary Fig. 9d,e). We tested our AND-gated peptide against the biological truth table for GzmB and MMP9 and found that pooling media from T cells and cancer cells increased activation of AND-gated peptides ~3-fold compared to medium from T cells or cancer cells alone (Supplementary Fig. 9f). We next assessed whether AND-gated nanosensors differentiate anti-tumor T cell killing from other co-culture conditions. We mixed OT-1 CD8+ T cells with MC38 cells pulsed with either the OT-1 cognate peptide antigen OVA_257–264 or the non-cognate LCMV gp_34–41 peptide and verified antigen-specific cytotoxicity by release of lactate dehydrogenase (Fig. 3a,b). We then collected media from T cells only, tumor cells only, T cells with tumor cells, or T cells with cognate antigen-pulsed tumor cells. Antigen-pulsed tumor cells were required for GzmB secretion by T cells, and all conditions with tumor cells showed elevated MMP9 secretion (Fig. 3c,d). Incubation of AND-gated nanosensors led to significant activation from media conditioned by T cells with antigen-pulsed tumor cells within 2.5 minutes (P<0.0001), with no significant activation from all other conditions after 30 minutes (Fig. 3e). Thus, AND-gated nanosensors can process protease activity to discriminate T cell killing of cancer cells.

Figure 3. AND-gated nanosensors selectively report on T cell killing of cancer cells.

a, Schematic of co-culture assay using TCR-transgenic OT-1 T cells and MC38 cancer cells that were pulsed with the cognate OVA_257–264 antigen (SIINFEKL) or a LCMV gp_34–41 (AVYNFATC) as a negative control peptide. b, Cytotoxicity quantified by release of lactate dehydrogenase (LDH) (one-way ANOVA with Tukey’s post-test and correction for multiple comparisons). c-d, Concentrations of GzmB (c) and MMP9 (d) in co-culture media determined by ELISA (two-way ANOVA with Tukey’s post-test and correction for multiple comparisons). e, Cleavage fluorescence of AND-gated nanosensor when treated with co-culture media (two-way ANOVA with Dunnett’s post-test and correction for multiple comparisons). (d, *P=0.0227, **P=0.0011, ***P=0.0002; a-e, ****P<0.0001, n=3 biologically independent wells, error bars depict mean ± s.e.m.)

AND-gated nanosensors detect responses to ICBT

ICBT has transformed patient outcomes across many cancer types, yet objective response rates remain below 40% for most indications^40,41. GzmB is a key driver of anti-tumor responses^38,42, and several activity-based probes have been developed to assess ICBT responses using GzmB activity^26,43,44. We therefore tested whether AND-gated nanosensors discriminate anti-tumor responses in a preclinical model of ICBT.

We treated mice bearing syngeneic MC38 flank tumors with a triple-drug combination of anti-programmed cell death protein 1 antibody, anti-cytotoxic T lymphocyte-associated protein 4 antibody, and interleukin-2. This therapy induced significant tumor regression after three doses compared to mice treated with isotype control antibodies, with significant elevations in the frequency and number of GzmB+CD8+ tumor-infiltrating lymphocytes (Fig. 4a–d, Supplementary Fig. 10,11). To compare sensor activation in the tumor and non-tumor organs during ICBT responses, we designed AND-gated nanosensors labeled with a near-infrared fluorescence (NIRF) FRET pair (sulfo-Cyanine7 and TQ7WS) such that the quencher is released upon dual cleavage, leading to NIRF at the site of activated NPs (Fig. 4e). These NIRF AND-gated nanosensors produced elevated fluorescence only when both GzmB and MMPs were present and were activated by media conditioned during T cell killing of MC38 cancer cells (Supplementary Fig. 12). We administered NIRF AND-gated nanosensors intratumorally in mice bearing MC38 tumors, one day after the third ICBT dose, and quantified the biodistribution of activated sensors three hours post-injection. In mice treated with isotype antibodies, tumor fluorescence was ~4.9% of the total fluorescent signal across major organs, whereas in ICBT-treated mice, the relative fluorescence was ~7-fold higher in tumor tissue (34.3%, P<0.001; Supplementary Fig. 13). No significant changes in NIRF were observed in major organs (brain, heart, kidneys, liver, lungs, tumor-draining lymph nodes, spleen; Fig. 4f,g). Tumor fluorescence produced by AND-gated nanosensors also differentiated ICBT responses from resistant B2m^–/– tumors, which lack antigen presentation and escape recognition by CD8+ T cells, resulting in reduced GzmB activity^45,46 (Supplementary Fig. 14). Therefore, AND-gated nanosensors detect protease activity in ICBT-responding tumors.

Figure 4. AND-gated nanosensors detect anti-tumor responses during immune checkpoint blockade therapy.

a, Tumor growth kinetics for C57BL/6J (B6) mice bearing subcutaneous MC38 tumors. Starting on day 13, mice were treated every 3 days for up to 4 doses of immune checkpoint blockade (ICB) therapy or isotype (Iso) control treatment. ICB therapy consisted of intravenous administration of 0.2 milligrams each of αPD1 and αCTLA4 antibodies and intraperitoneal administration of 10 μg of IL-2 cytokine. Iso control treatment consisted of intravenous administration of 0.2 mg each of IgG1 and IgG2b isotype control antibodies (n=8 (ICB) or 7 (Iso) biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). b-d, Representative flow plots (b), frequency (c), and number per gram of tissue (d) of GzmB+CD8+ T cells isolated from MC38 tumors in mice treated with ICB therapy or Iso. Flow analysis was performed one day after the third treatment dose (n=5 (Iso) or 7 (ICB) biological replicates, two-tailed Student’s t-test). e-g, Schematic (e), representative images (f), and quantification (g) of near-infrared fluorescence (NIRF) in MC38 tumors and major organs (liver, lungs, heart, spleen, kidneys, brain, and tumor-draining lymph nodes (tdLNs)) upon intratumoral (i.t.) administration of quenched NIRF AND-gated nanosensors to tumor-bearing mice one day after the third ICB or Iso treatment dose. Images were taken 3 hours after sensor administration. Relative fluorescence in each organ (RFU of organ/total RFU from all excised organs for each mouse) was normalized to average signal for Iso-treated mice (n=6 (Iso) or 7 (ICB) biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). h-j, Urinalysis schematic (h), quantification (i), and receiver-operator characteristic (ROC) curves (j) of carboxyfluorescein (5(6)-FAM) reporters in urine of tumor-bearing mice receiving i.t. administration of 5(6)-FAM-labeled AND-gated nanosensors one day after the third treatment. Urine was collected 3 hours after sensor administration. Reporters were purified from urine by immunoprecipitation and quantified by fluorescence, which was normalized to the average signal for Iso-treated mice at each timepoint. B2m^–/– tumors were generated from MC38 cancer cells with knocked out expression of B2m (beta-2 microglobulin) (n=6 (Iso), 8 (ICB), or 3 (ICB, B2m^–/– tumor) biological replicates, two-way ANOVA with Tukey’s post-test and correction for multiple comparisons). (c, ***P=0.0005; d, *P=0.0349; g, ***P=0.0009; i, *P=0.0220, **P=0.0045; a-j, ns, not significant, error bars depict mean ± s.e.m.; AUC, area under the ROC curve)

While activatable fluorescent probes are under evaluation to highlight residual disease during surgical resection^47,48,49, fluorescence-based in vivo assays are not routinely used in the clinic. Therefore, we evaluated whether intratumoral injection of AND-gated nanosensors followed by quantification of released reporters in urine could assess responses to ICBT, given that protease-cleaved peptide fragments are below the renal filtration size limit and rapidly concentrate in urine^6,30. We designed AND-gated nanosensors to release fluorophore-labeled fragments into urine for purification by immunoprecipitation and quantification by fluorimetry. Three hours after intratumorally administering nanosensors, urinary reporters were significantly elevated in mice treated with ICBT compared to mice receiving isotype antibodies and ICBT-treated mice bearing B2m^–/– tumors (Fig. 4h,i). Urinary reporters indicated response with high diagnostic accuracy (AUROC=0.85 versus isotype, 1.00 versus B2m^–/–, Fig. 4j). These results demonstrate that local activation of AND-gated nanosensors can be quantified distally in urine to indicate anti-tumor T cell responses.

AND logic increases specificity to co-localized proteases

T cells secrete GzmB not only during anti-tumor responses but also pathologies like viral infection, autoimmunity, and transplant rejection^50,51. However, we postulated that AND-gated sensing will increase specificity by minimizing signal in tissues with T cell activity but without MMP upregulation. To test this, we made artificial tissue mixtures by spiking GzmB into MC38 tumor homogenates (MMP-high) or healthy tissue homogenates (MMP-low). Nanosensor activation was 1.45-fold higher in tumor samples than all other samples (Supplementary Fig. 15). We further tested specificity in vivo using a mouse model of acute viral infection by PR8 influenza A. We confirmed infection by decreased body weight (Fig. 5a, Supplementary Fig. 16) and increased GzmB+CD8+ T cells in the bronchoalveolar lavage fluid and lungs 8 days post-infection compared to naïve mice (Fig. 5b,c). We then intravenously administered AND-gated or linear GzmB nanosensors to separate cohorts and quantified NIRF in major organs. Linear GzmB nanosensors resulted in 3.0-fold higher fluorescence in the excised lungs of infected mice compared to naïve mice (P<0.0001) whereas NIRF signals were statistically identical across all tissues for mice receiving AND-gated nanosensors (P>0.05; Fig. 5e,f, Supplementary Fig. 17). These results indicate that AND-gated nanosensors improve local tissue specificity for T cell activity.

Figure 5. AND-gated nanosensors increase specificity by requiring co-localized proteases for activation.

a, Schematic and kinetics of body weight for mouse model of viral infection using intranasal administration of PR8 influenza A (30 plaque-forming units) (n=10 biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). b-c, Representative flow plots (b) and number (c) of GzmB+CD8+ T cells isolated from bronchoalveolar lavage fluid (BALF) and lungs of naïve and PR8-infected mice eight days post-infection. Dump channel includes cells stained with αCD4, αCD19, and αNK1.1 antibodies (n=3 biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). d, Schematic of expected detection specificity for AND-gated sensor. A linear GzmB sensor is expected to activate during both anti-viral responses to influenza infection and anti-tumor responses to ICB. By contrast, an AND-gated sensor is expected to be tumor-specific and remain inactive during anti-viral responses. e-f, Representative images (e) and quantification (f) of NIRF (RFU) in lungs upon intravenous (i.v.) administration of quenched NIRF linear GzmB nanosensors or AND-gated nanosensors to PR8-infected mice 8 days post-infection. Images were taken 3 hours after sensor administration (n=5 biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). g, Schematic of possible outcomes for AND-gated nanosensors in distal tumors after i.t. administration to a local tumor in a bilateral tumor model. If sensors were partially cleaved by GzmB in the local tumor, they would activate distally in both resistant B2m^–/– and responsive wild-type (w.t.) tumors independent of GzmB expression in the distal tumor. By contrast, if sensors remained predominantly intact upon arrival to distal tumors, they would retain the ability to discriminate resistant and responsive tumors dependent on GzmB expression. h-i, Representative images of local and distal tumors (h) and quantification of NIRF (RFU/area) in distal tumors (i) for bilateral tumors upon i.t. administration of quenched NIRF AND-gated nanosensors to local tumors (n=8 (w.t./w.t. and w.t./B2m^–/–) or 6 (B2m^–/–/w.t.) biological replicates, one-way ANOVA with Dunnett’s post-test and correction for multiple comparisons). (a, **P=0.0075; i, **P=0.0032 (w.t./w.t. versus w.t./B2m^–/–) or 0.0037 (w.t./B2m^–/– versus B2m^–/–/w.t.); a-i, ****P<0.0001, ns, not significant, error bars depict mean ± s.e.m.)

Next, we assessed whether AND-gated nanosensors require co-localized input proteases, as opposed to a mechanism whereby after partial activation by a protease at one tissue site, nanosensors traffic to a distal site where a second protease completes activation. We assessed this in mice bearing bilateral MC38 flank tumors using a combination of ICBT-responsive wild-type (w.t.) tumors (MMP-high/GzmB-high) or treatment-resistant B2m^–/– tumors (MMP-high/GzmB-low) (Fig. 5g). We reasoned that high NIRF signals in distal B2m^–/– tumors after intratumoral delivery of nanosensors to local w.t. tumors would indicate “false positive” signals, resulting from GzmB cleavage in w.t. tumors followed by MMP activation in distal B2m^–/– tumors. By contrast, high NIRF signals in local w.t. tumors and low NIRF signals in distal B2m^–/– tumors would support a co-localized protease activation mechanism. Experimentally, we observed significantly lower NIRF signals in distal B2m^–/– tumors compared to distal w.t. tumors (P<0.01) after local injection in w.t. tumors, supporting the latter hypothesis. We also tested the reverse condition using local B2m^–/– tumors and observed increased NIRF signals in distal w.t. tumors (Fig. 5h,i), providing additional support.

Improved systemic specificity for anti-tumor immunity

We evaluated whether AND-gated nanosensors improve on-tumor specificity compared to linear GzmB nanosensors upon intravenous administration. We first assessed toxicity of systemically delivered AND-gated nanosensors. We performed two intravenous injections in healthy mice a week apart to assess inflammatory responses caused by repeated dosing. We observed no significant changes in weight, temperature, serum metabolites, or inflammatory cytokines, indicating that the nanosensors were well-tolerated under these conditions (Supplementary Fig. 18). Next, we systemically administered linear GzmB nanosensors or AND-gated nanosensors in mice receiving ICBT or isotype antibodies. ICBT resulted in ~1.25-fold higher tumor fluorescence relative to isotype treatment using linear GzmB nanosensors (Supplementary Fig. 19), compared to a significant ~2.1-fold increase using AND-gated nanosensors (P<0.001; Fig. 6a,b, Supplementary Fig. 20). The increased activation with AND-gated nanosensors may be due to improved stability as cyclization can reduce peptide degradation by serum proteases^33,34. We also observed activation of AND-gated nanosensors in the tumor-draining lymph nodes, with no significant changes in other organs. We further compared the tumor selectivity of the sensors. AND-gated nanosensors had significantly higher activation in ICBT-responding tumors than 87.5% of tested non-tumor organs compared to 37.5% for linear GzmB nanosensors (Supplementary Table 1), resulting in higher tumor selectivity for AND-gated nanosensors (Fig. 6c, Supplementary Fig. 21).

Figure 6. AND-gated nanosensors increase selectivity of on-tumor detection.

a-b, Representative image (a) and quantification (b) of fluorescence in MC38 tumors and major organs produced by intravenous (i.v.) administration of NIRF AND-gated nanosensors to tumor-bearing mice one day after the third ICB or Iso treatment dose. Images were taken 3 hours after sensor administration. Fluorescence (RFU/area) in each organ was normalized to average signal for Iso-treated mice (n=6 (Iso) or 7 (ICB) biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). c, Average tumor-to-organ selectivity for each organ produced by linear GzmB nanosensors or AND-gated nanosensors. Tumor-to-organ selectivity is defined as the ratio of isotype-normalized fluorescence in ICB-treated tumors compared to the corresponding normalized fluorescence in a respective non-tumor organ. Average selectivity was calculated using the mean from n=6 (Iso) or 7 (ICB) biological replicates (two-tailed paired Student’s t-test). d, Quantification of fluorescence in MC38 tumors and major organs produced by i.v. administration of NIRF AND-gated nanosensors to tumor-bearing mice that were treated with one dose of oxaliplatin (0.1 mg, i.p.) followed by three doses of ICB or Iso. Fluorescence (RFU/area) in each organ was normalized to average signal for mice treated with Iso only (n=4 (Iso), 6 (ICB), or 7 (Oxa + Iso, Oxa + ICB) biological replicates, two-way ANOVA with Sidak’s post-test and correction for multiple comparisons). e-f, Quantification of fluorescence in MC38 tumors, lungs, and other major organs (e) and representative images (f) after i.v. administration of NIRF AND-gated nanosensors to tumor-bearing mice infected with PR8 influenza A virus and treated with three doses of ICB or Iso. Corresponding images of liver and spleen are in Supplementary Figure 23. Fluorescence (RFU/area) in each organ was normalized to average signal for uninfected control mice treated with Iso only (n=5 (PR8 + Iso), 6 (Iso only), 7 (PR8 + ICB, B2m^–/– tumor), or 8 (PR8 + ICB) biological replicates, two-way ANOVA with Dunnett’s post-test and correction for multiple comparisons). (b, *P=0.0382, ***P=0.0005; c, **P=0.0073; d, **P=0.0042; f, *P=0.0377, **P=0.0069; a-f, ****P<0.0001, error bars depict mean ± s.e.m.; ndLNs, non-tumor-draining lymph nodes; tdLNs, tumor-draining lymph nodes)

While ICBT is FDA-approved for first-line use in several cancer types, many approved indications for ICBT are either in combination with or following chemotherapy^52,53. Chemotherapeutics can induce off-tumor inflammation and remodeling, which are mediated in part by MMPs^54,55. We therefore evaluated whether AND-gated nanosensors maintain specificity for anti-tumor responses after preconditioning with one dose of oxaliplatin, a platinum compound used in standard-of-care treatments for colorectal cancer⁵⁶. While oxaliplatin followed by ICBT resulted in ~2.7-fold higher tumor fluorescence relative to treatment with oxaliplatin and isotype antibodies, we did not observe activation in off-tumor organs for either cohort when compared with mice receiving only isotype antibodies (Fig. 6d, Supplementary Fig. 22), indicating that AND-gated nanosensors maintain on-tumor specificity after chemotherapy.

Finally, we evaluated whether AND-gated nanosensors selectively activate at sites where protease pairs are co-localized using a comorbidity model of ICBT response and PR8 influenza infection. Infected mice with tumors that are not responsive to ICBT (via isotype treatment or B2m^–/– resistance) have elevated GzmB and MMP at different tissue sites (lungs and tumor, respectively). Therefore, ICBT-responsive tumors represent the unique tissue site where both proteases are co-localized. Upon intravenous administration, NIRF AND-gated nanosensors did not activate in the tumor, lungs, liver, or spleen of PR8-infected mice bearing B2m^–/– tumors, or of infected mice bearing w.t. tumors and receiving isotype antibodies, when compared to uninfected, isotype-treated controls (Fig. 6e,f, Supplementary Fig. 23). By contrast, ICBT in infected mice bearing w.t. tumors resulted in significantly higher tumor fluorescence compared to all control groups, without statistical changes in the lungs, liver, or spleen. Collectively, our data supports that AND-gated nanosensors remain inactive when target proteases are expressed in different tissues and selectively detect co-localized proteases to discriminate anti-tumor responses from off-tumor anti-viral immunity.

Conclusions

Genetically engineered biocircuits have enabled biosensors that conditionally activate in tumors with high specificity by implementing synthetic logic. Our work demonstrates the application of activity-based nanosensors that implement AND-gate logic without genetic components to increase detection precision for anti-tumor immune responses. While ICBT can induce T cell immunity off-tumor via reactivation of anti-viral T cells^57,58, immune-related adverse events^40,59, and opportunistic infections upon treatment of adverse events^60,61, our study shows that AND-gate logic improves specificity to anti-tumor responses and minimizes detection of T cell-secreted GzmB without co-localized MMP activity during off-tumor anti-viral immunity. While genetic circuits continue to drive major advances in cell therapies and diagnostics, clinical translation of cell-based systems for in vivo applications often requires reducing immunogenicity of foreign components, such as by humanizing synthetic receptors⁶², and may have restricted accessibility for larger patient populations, driving the development of “off-the-shelf” allogeneic cell therapies^63,64 and in vivo manufacturing to bypass ex vivo pipelines^65,66. As such, cell- and gene-free biosensors that implement logic could present lower barriers for translation.

While our study focused on detection of T cell-mediated anti-tumor responses, multiple inflammatory and immunosuppressive cell types contribute to responses to immunotherapy^67,68. For example, imaging probes have been described that bind to or are cleaved by proteases expressed by macrophages⁶⁹, neutrophils⁶⁹, and cancer-associated fibroblasts⁷⁰ during cancer progression or anti-tumor responses. Our demonstration of platform modularity using substrate substitutes for other tumor-relevant endoproteases such as legumain, thrombin, and fibroblast activation protein warrants future studies to determine whether our approach can be extended for detection of multiple cell types.

In our study, we validated detection strategies using both intratumoral and systemic delivery of AND-gated nanosensors. ICBT has received first-line approvals for both superficial tumors and cancers originating from deeper tissues^52,53. Looking forward, we envision that intratumoral assessment may be rapidly applied in an outpatient setting for superficial tumors, complementing imaging-based approaches like CT and PET. Systemic delivery may be applicable for cancers in deeper tissues, which could be assessed using orthotopic models in future work. In addition, chemotherapy is commonly used in first-line combinations with or prior to second-line ICBT^52,53. While chemotherapy can induce MMP-mediated processes like wound repair and tissue remodeling^54,55, we found that AND-gated nanosensors maintained minimal off-tumor activation when platinum chemotherapy preceded ICBT while producing higher on-tumor signals, consistent with evidence that immunogenic cell death caused by chemotherapy may sensitize tumors to ICBT and increase anti-tumor responses^71,72.

Overall, our results demonstrate AND-gated nanosensors implement synthetic logic without genetic circuitry to produce detection signals after activation by protease pairs, which has implications for localized detection of disease and drug responses at specific tissue sites.

Methods

Animals

Female mice (6- to 10-week-old) were used at the outset of all experiments. OT-1 (C57BL/6-Tg(TcraTcrb)1100Mjb/J) transgenic mice were bred in-house using breeding pairs purchased from The Jackson Laboratory (Jax, 003831). C57BL/6J (B6) mice were purchased from Jax (000664) and were used as recipients for all tumor models. A maximal tumor size of 1.5 cm in any dimension was used as an endpoint for all tumor models and not exceeded. All animal procedures in this study received ethical approval by the Georgia Tech Institutional Animal Care and Use Committee (protocol no. KWONG-A100190, KWONG-A100191, and KWONG-A100193).

Peptide synthesis

Peptides were either synthesized in house or ordered from a vendor (Lifetein, Genscript, or CPC Scientific). In-house synthesis was performed by Fmoc solid phase peptide synthesis on a Liberty Blue peptide synthesizer (CEM) using low-loading rink amide resin (CEM). Briefly, for each amino acid cycle, Fmoc deprotection was performed using 20% (v/v) 4-methylpiperidine in dimethylformamide, and amino acids (5 equivalents (eq.); Chem Impex) were coupled in the presence of 10 eq. diisopropylcarbodiimide (DIC) and 5 eq. Oxyma Pure. After synthesis, peptides were cleaved off resin using 92.5% trifluoroacetic acid (TFA), 2.5% water, 2.5% triisopropylsilane, and 2.5% 3,6-dioxa-1,8-octane-dithiol for 30–45 minutes at 41C using Razor (CEM). After cleavage, peptides were precipitated in ice-cold diethyl ether and vacuum dried overnight. Crude peptides were purified by reverse phase high-performance liquid chromatography (HPLC; 1260 Infinity II, Agilent) using a Zorbax C18 column (Agilent) using an elution gradient from 5% to 100% acetonitrile in 0.05% TFA. Acetonitrile was removed from purified fractions by rotary evaporation, and samples were frozen overnight and lyophilized.

To make cyclic peptides, linear peptides were first synthesized with Lys(Alloc) near the C-terminus and Glu(OAllyl) at the N-terminus. Fmoc-protected resin was then treated with phenylsilane and tetrakis(triphenylphosphine)palladium(0) in dichloromethane at 35C to restore Lys and Glu side chains, followed by cyclization using DIC and Oxyma.

For labeling with fluorophores and quenchers, resin-coupled peptides were N-terminally labeled with 5(6)-carboxyfluorescein (FAM) directly on Liberty Blue with DIC and Oxyma pure after Fmoc deprotection. Alternatively, Fmoc deprotection was performed on Liberty Blue, and resin-coupled peptides were N-terminally labeled with Tide Quencher 7WS (TQ7) succinimidyl ester (AAT Bioquest) in dimethyl sulfoxide with 4–10 eq. triethylamine at 37C for 2 hours. Peptides, which were synthesized with propargylglycine, were cleaved from resin and labeled with Tide Quencher 2 (TQ2) azide (FAM peptides; AAT Bioquest) or sulfo-cyanine7 azide (Lumiprobe) by copper(I)-catalyzed azide-alkyne cycloaddition (CuAAC). CuAAC reaction was performed using copper(II) sulfate, tris(3-hydroxypropyltriazolylmethyl)amine (THPTA), ascorbic acid, and aminoguanidine in phosphate-buffered saline at 37C overnight. Peptides were purified by HPLC after dye labeling.

Mass spectrometric analysis of synthetic peptides and cleavage products

Mass analysis of crude and purified peptides synthesized in house was performed by liquid chromatography-mass spectrometry (LC/MS; 1260 Infinity II and InfinityLab LC/MSD, Agilent; OpenLab CDS ChemStation v2.7 software) on positive ion mode electrospray ionization (ESI/MS) using a Zorbax or Poroshell C18 column (Agilent) and an elution gradient from 5% to 100% acetonitrile in either 0.05% TFA or 0.05% formic acid (FA). Mass analysis of peptide cleavage products was performed by ESI/MS (Agilent) or by matrix-assisted laser desorption/ionization mass spectrometry (MALDI/MS; Autoflex, Bruker) using an α-cyano-4-hydroxycinnamic acid matrix. Prior to analysis, cleavage products were diluted to 75% acetonitrile and refrigerated for 30 minutes to precipitate proteins, followed by centrifugation at 4000xg for 5 minutes.

Nanosensor conjugation

Amine-functionalized iron oxide nanoparticles (IONPs) were synthesized in-house as previously described^25,73 and purified by fast-performance liquid chromatography (FPLC; AKTA Pure, Cytiva; Unicorn v7.11 software) using a Superdex 200 Increase 10–300 GL column, resulting in particles of ~20 nm average hydrodynamic diameter as determined by dynamic light scattering (Zetasizer, Malvern). For conjugation to peptides, nanoparticles were labeled with SM(PEG)₆ (ThermoFisher) for 2 hours, then crosslinked to peptides containing free cysteines overnight. Reactions were performed in phosphate-buffered saline (PBS; pH 7.2–7.4) at room temperature; 2 mM ethylenediaminetetraacetic acid was added to the peptide crosslinking reaction to reduce formation of disulfide bonds between peptides. Nanosensors were purified by FPLC. Nanosensor valency (i.e., peptide-to-nanoparticle ratio) was quantified by UV/Vis spectroscopy, using 400 nm for IONPs and 490 nm for FAM/TQ2-labeled peptides or 750 nm for Cy7/TQ7-labeled peptides. All valency calculations were performed using optical density (OD) values below 2.0 to ensure linearity of the Beer-Lambert Law. Samples were diluted in PBS to achieve OD<2.0 when necessary.

Cell culture

MC38 cells (kind gift of the National Cancer Institute and Dr. Dario Vignali, University of Pittsburgh) and B2m^−/− MC38 cells (developed in previous work²⁶) were cultured in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (PenStrep; ThermoFisher). Cells were grown to 70–90% confluence before being trypsinized and passaged with a split ratio of 20–40:1. For OT-1 T cells, OT-1 transgenic mice were sacrificed, primary splenocytes were harvested, and CD8 T cells were isolated (CD8a+ T cell isolation kit, Miltenyi). T cells were activated for 48 hours using αCD3e and αCD28 antibodies (BD Biosciences; clones 145–2C11 and 37.51, respectively) and cultured for up to 4 days at <3 million cells per mL, using T cell media (RPMI 1640 supplemented with 10% FBS, 1% PenStrep, 1% non-essential amino acids, 1% sodium pyruvate, 4 ppm (v/v) beta-mercaptoethanol, and 30 units/mL IL-2 (Roche)).

Fluorogenic substrate assay

Granzyme B (GzmB; Peprotech 140–03), thrombin (Thrb; Prolytix HCT-0020), fibroblast activation protein (FAP; R&D Systems 3715-SE), gamma-glutamyltransferase 1 (GGT1; R&D Systems 10977-GT), granzyme A (GZMA; Enzo Life Sciences ALX-201–118), ADAMTS1 (R&D Systems 2197-AD), tissue-type plasminogen activator (TPA; R&D Systems 7449-SE), urokinase-type plasminogen activator (UPA; R&D Systems 1310-SE), and plasmin (PLG; Prolytix HCPM-0140) were diluted directly in PBS. Matrix metalloproteinases (MMPs; Enzo Life Sciences BML-AK013 and BML-AK014) were diluted in MMP activation buffer (50 mM Tris, 150 mM NaCl, 10 mM CaCl₂, 0.1 mM ZnCl₂, 0.05% Brij-35, pH 7.4) and activated for 30 minutes at 37C prior to dilution in PBS. Legumain (LGN; R&D Systems 2058-CY) and cathepsins (CTSD, CTSE, CTSS; Enzo Life Sciences BML-SE199, R&D Systems 1294-AS, R&D Systems 1183-CY) were diluted in LGN activation buffer (100 mM NaOAc, 100 mM NaCl, pH 4.5; 20 mM DTT was added for CTSS) and activated at 37C prior to dilution in PBS to final pH ~5.5. Caspases (CASP1, CASP3, CASP8; Enzo Life Sciences BML-SE168, BML-SE169, BML-SE172) were diluted in caspase activation buffer (50 mM HEPES, 100 mM NaCl, 20 mM DTT, 2 mM EDTA, 0.1% Tween-20, 10% glycerol) and activated for 30 minutes at 37C prior to dilution in PBS. Conditioned media was isolated freshly from cultured cells by centrifugation at 1000xg for 5 minutes. Peptides or nanosensors were mixed with recombinant proteases or conditioned media and incubated at 37C. For peptides labeled with FAM and TQ2, sample fluorescence was measured by Cytation 5 plate reader (Biotek; Gen5 software). For peptides labeled with Cy7 and TQ7, sample fluorescence was measured by Odyssey CLx Imager (LI-COR; ImageStudio v5.2 software). Peptide concentrations were 1–10 μM for free peptides in solution and 0.1–1 μM for nanosensors. Protease concentrations were 50–250 nM unless otherwise specified.

T cell cytotoxicity assay

Five days after harvesting OT-1 T cells, MC38 cells were trypsinized and incubated at 1 million cells per mL with 30 μM antigen peptide (OVA or gp34) in OptiPro for 1 hour at 37C. Antigen-pulsed MC38 cells were washed in T cell media, and 60,000 cells were seeded in a 96-well plate. After 4–6 hours, OT-1 T cells were washed in T cell media, and 300,000 cells were added to the tumor cells (effector:target ratio=5:1). After overnight incubation (12–16 hours), conditioned media was isolated by centrifugation. Conditioned media was either used for fluorogenic substrate assay or analyzed by lactate dehydrogenase assay (Abcam) or enzyme-linked immunosorbent assay for GzmB (ThermoFisher) or MMP9 (Abcam).

Mouse models of cancer immune checkpoint blockade therapy

B6 mice were subcutaneously inoculated into the left flank with 1 million MC38 cells or B2m^−/− MC38 cells. Both left and right flanks were inoculated for the bilateral tumor model. Tumor burden (0.52 × length × width × depth) was monitored twice weekly, and treatment was initiated at 300–400 mm³. 0.2 milligrams each of αPD1 antibody (kind gift of Dr. Gordon Freeman, with the help of Dr. Rafi Ahmed of Emory University, clone 8H3; original source: Dana-Farber, not commercially available) and αCTLA4 antibody (BioXCell; clone 9H10) or matched IgG1 and IgG2b isotype controls (BioXCell; clones MOPC-21 and MPC-11, respectively) were intravenously administered by tail vein injection every 3 days for up to 4 doses. 10 μg of IL-2 (Peprotech) was intraperitoneally administered to mice receiving αPD1/αCTLA4 treatment on the same schedule. For chemotherapy, 0.1 mg of oxaliplatin (Thermo Fisher J66586.MF) was intraperitoneally administered 3 days before initiation of ICBT. For tumor-bearing mice infected with PR8 influenza A, mice were infected as described below one day before initiation of ICBT. Flow cytometric analysis and intratumoral or intravenous injection of nanosensors for near-infrared fluorescence (NIRF) imaging or urinalysis were performed one day after the third treatment dose.

Mouse models of influenza A virus

B6 mice were intranasally inoculated with 30 plaque-forming units (p.f.u.) of purified H1N1 influenza A virus (PR8; strain A/PR/8/34). Body weight was measured daily. Flow cytometric analysis and intravenous injection of nanosensors for near-infrared fluorescence (NIRF) imaging were performed eight days post infection.

Organ dissociation and flow cytometry analysis

For tumor-bearing mice, MC38 tumors were enzymatically and mechanically dissociated using mouse tumor dissociation kit (Miltenyi) and gentleMACS dissociator (Miltenyi), respectively. Tumor-infiltrating lymphocytes (TILs) were isolated from the single-cell suspension using a density gradient with Percoll centrifugation media (GE Life Sciences) and RPMI 1640 at 44:56 volume ratio, followed by red blood cell (RBC) lysis. For PR8-infected mice and naïve controls, bronchoalveolar lavage fluid (BALF) was collected using 3 washes with 1 mL of PBS. Cells were harvested from BALF by centrifugation, followed by RBC lysis. Lungs were enzymatically dissociated using collagenase type IV. Lymphocytes were isolated from the single-cell suspension using a 40:60 Percoll:RPMI density gradient, followed by RBC lysis.

For flow cytometry analysis, cells were first stained for surface markers in FACS buffer (1x DPBS, 2% FBS, 1 mM EDTA, 25 mM HEPES), followed by viability staining with LIVE/DEAD Fixable Near-IR dye (Invitrogen). Cells were fixed and permeabilized (eBioscience) prior to intracellular staining for GzmB in permeabilization buffer. Stained cells were analyzed using an LSRFortessa (BD Biosciences; FACSDiva v8 software) or Aurora (Cytek Biosciences; SpectroFlo v3.2.1 software) flow cytometer. To determine absolute cell counts, CountBright Absolute Counting Beads (ThermoFisher) were added to samples immediately prior to analysis.

Antibody clones used were αCD45 (30-F11), αCD3 (145–2C11), αCD8 (53–6.7), αCD4 (RM4–5), αNK1.1 (PK136), αCD19 (6D5), and αGZMB (GB12). αGZMB was purchased from ThermoFisher; all other antibodies were purchased from Biolegend. All antibodies were used for staining at 1:100 dilution from stock concentrations.

Near-infrared fluorescent imaging of sensor activation in whole organs and homogenates

Three hours after injection of NIRF Cy7/TQ7-labeled nanosensors, tumor-bearing or PR8-infected mice were euthanized, and tumor and organs (brain, lungs, heart, liver, kidneys, spleen, tumor-draining and non-tumor-draining lymph nodes) were isolated, rinsed in PBS, and imaged by Odyssey CLx (LI-COR; ImageStudio v5.2 software). Alternatively, tumor and organs were isolated from mice that did not receive NIRF nanosensors and were dissociated using FastPrep-24 homogenizer (MP Biomedicals) with beads (Lysing Matrix D, MP Biomedicals) in tissue protein extraction buffer (T-PER, ThermoFisher). Homogenate was isolated by centrifugation at 14,000xg for 5 minutes and added to GzmB and NIRF nanosensors for fluorogenic cleavage assay as described above.

Detection of urinary reporters

Either one day before the first dose or 24 hours after the third dose of αPD1/αCTLA4 treatment, FAM-labeled nanosensors were intratumorally administered. Urine was collected for three hours. FAM reporters were isolated from urine samples by immunoprecipitation using Dynabeads (ThermoFisher) decorated with anti-fluorescein isothiocyanate antibody (GeneTex GTX42751, clone 19F/2A3, multiple lots used; 1 mg antibody per 25 mg Dynabeads) with elution in acetic acid (5% v/v) and neutralization in Tris buffer (2 M, pH 12). Sample fluorescence was measured using a Cytation5 plate reader (Biotek), and reporter concentrations were determined by using a known FAM ladder.

Toxicology

AND-gated nanosensors were intravenously injected twice, one week apart. Body temperature was measured by rectal probe. One week after the second injection, whole blood was collected in serum separator tube, incubated at room temperature for 30 minutes to allow for clotting, and centrifuged at 2500xg for 10 minutes at 4C, after which serum was isolated. Serum chemistry analysis was performed by Antech Diagnostics, and serum cytokines were analyzed using LEGENDplex assay (Biolegend).

Software and statistical analysis

Graphs were plotted and appropriate statistical analyses were conducted using GraphPad Prism v8 (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; error bars depict mean ± s.e.m.). At least 3 replicates were used for all statistical analyses. Measurements were taken from independent samples, using biological replicates when possible. Data distribution was assumed to be normal, but this was not formally tested. Equal variances were confirmed using F test to compare variances. NIRF images were analyzed using Image Studio v5.2 (LI-COR). Flow cytometry data were analyzed using FlowJo X. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications^25,26. For all animal studies, animals were randomly assigned to various experimental groups, including random assignments for which tumor type was inoculated (e.g., w.t. or B2m^−/−), which intervention was given (e.g., ICBT or isotype), which nanosensor was administered (e.g., AND-gate or linear GzmB), and whether mice were infected with PR8 or naïve. For tumor models, a stratified block randomization strategy was used to ensure similar tumor burden across all cohorts. When feasible during in vivo studies (e.g., tumor measurements, organ harvest, urine collection), data collection was performed blinded by de-identifying mice with ear tag numbers that were matched to group allocation only after data collection and analysis. During other in vivo procedures and in vitro studies, blinding was not feasible as procedures were performed by individual investigators who required knowledge of the experimental groups to administer the allocated condition (e.g., for in vivo studies, administration of allocated treatment or nanosensor; for in vitro studies, addition of allocated protease or cell population). After data collection, all data was analyzed quantitatively without knowledge of group allocation, after which samples were matched to their group for data visualization and statistical analysis. No data were excluded from the analyses. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data Availability

The source data underlying the main text figures (Figs. 1–6) and Supplementary Figs. are provided as source data files. All raw data and calculations used to generate plotted data are provided with this paper in the source data files. Image source data for Supplementary Figs. are provided at the end of the Supplementary Information file.