Highly active myeloid therapy (HAMT) for cancer

Abstract

Tumor-associated macrophages (TAM) interact with cancer and stromal cells and are integral in sustaining many cancer-promoting features. Therapeutic manipulation of TAM could therefore improve clinical outcomes and synergize with immuno- and other cancer therapies. While different nano-carriers have been used to target TAM, a knowledge gap exists on which TAM pathways to target and what payloads to deliver for optimal anti-tumor effects. We hypothesized that a multi-part combination involving the Janus Tyrosine Kinase (JAK), non-canonical nuclear factor kappa light chain enhancer of activated B cells (NF-κB), and toll-like receptor (TLR) pathways could lead to a highly active myeloid therapy (HAMT). Thus, we devised a screen to determine drug combinations that yield maximum IL-12 production from myeloid cells to treat the otherwise highly immunosuppressive myeloid environments in tumors. Here we show the extraordinary efficacy of a triple small-molecule combination in a TAM-targeted nanoparticle for eradicating murine tumors, jumpstarting a highly efficient anti-tumor response by adopting a distinctive anti-tumor TAM phenotype and synergizing with other immunotherapies. The HAMT therapy represents a recently developed approach in immunotherapy and leads to durable responses in murine cancer models.

Graphical Abstract

The biology of tumor-associated macrophages (TAM) has received increasing attention over the last decade as immunotherapies have taken center stage in cancer treatment. The emerging consensus is that TAM are abundant, are recruited primarily from bone marrow (myeloid origin)¹ and the vast majority of these cells are immunosuppressive and support and enable tumor growth. Conversely, a much smaller subtype of TAM (“M1-like”; M9)² have anti-tumor effects. However, consensus on which TAM phenotypes have optimal anti-tumor function remains uncertain.^3-6 Therefore, manipulation and reprogramming of TAM could enable improved disease control in patients where TAM are abundant and functionally active.

Various therapeutic options exist to modulate TAM function (myeloid therapeutics), including elimination, containment, or suppression of tumor-promoting macrophages, amplification or activation of anti-tumor macrophages, subtype switching, or any combinations.^{1, 7-10} Most therapeutic approaches today are based on antibodies targeting cell surface proteins (CSF1-R, CD40, TREM2, CD47) or small molecules targeting growth factors (CSFR1).^{8, 11} A less explored option is to harness the phagocytic function of TAM to efficiently deliver small-molecule immune modulators and thus improve the therapeutic index of those drugs. Most prior research has used nano-formulations to deliver small-molecule toll-like receptor (TLR) and cGAS/stimulator of interferon genes (STING) pathway agonists to TAM. Some of these nano-formulations have shown anti-tumor efficacy in mouse models of cancer^12-17 while others have limited efficacy.^{18, 19} and/or exhibit unacceptable systemic toxicity.²⁰ The diverging efficacy data are not entirely unexpected and are evidence that we still do not fully understand TAM function and re-programmability. This is on top of the emerging realization that i) TAM are plastic, ii) have multiple phenotypic states, and iii) have redundant wiring mechanisms to escape therapeutic pressures.

In highly active antiretroviral therapy for AIDS, combinations of three or more antiretroviral drugs control the virus, whereas using one or two drugs alone often results in treatment failure and the development of drug resistance. We hypothesized that a similar combination treatment strategy could bypass the limited success of previous TAM-directed monotherapies in cancer by packaging multiple small-molecule immune pathway modulators into the same nanoparticle. We also argued that selective pharmacological manipulation of TAM could lead to other drug-induced TAM phenotypes, generally not seen with natural biological polarization. A number of different anti-tumor TAM phenotypes have been identified by profiling, immunohistochemistry, scRNAseq and other methods. These “M1-like”^{21, 22} phenotypes include i) IL-12 producing TAM,^23-25 ii) CXCL9 producing TAM^26-28 among a few select others. Given the prominence of the interleukin (IL)-12 program, we used it in the current study (Fig. 1; Fig. S20).

Fig. 1: Highly effective myeloid therapy (HAMT).

A. Proof-of-concept scheme consisting of a myeloid cell targeting carbohydrate nanoparticle containing a triple therapeutic payload to affect the wiring of tumor-associated macrophages. B. In this work, we show that targeting three pathways (TLR (red dot), non-canonical NF-κB (dark blue dot), and JAK/STAT (light blue dot) is synergistic and drives macrophages toward a distinctive phenotype resulting in high IL-12 production and T cell priming. Overall pathways towards an M1 phenotype are activated (green check marks) and pathways towards an immunosuppressive M2 phenotype (red stop sign) are turned off. The increased IL-12 production, jumpstarts an effective anti-tumor T cell response. See Fig. S20 for detail.

We screened several combinations using IL-12 reporter assays, which is a major anti-tumor cytokine produced by macrophages. We focused on known modulators of the pro-inflammatory IL-12 and anti-inflammatory IL-10 pathways. We found that a three-part combination involving the Janus Tyrosine Kinase (JAK), the signal transducer and activator of transcription (STAT), the non-canonical nuclear factor kappa light chain enhancer of activated B cells (NF-κB), and the toll-like receptor (TLR) pathways leads to a highly active myeloid therapy (HAMT) and can be used to manage and treat the immunosuppressive myeloid environment in tumors (Fig. 1; Fig. S20). HAMT potentiates anti-tumor immune responses in aggressive mouse models of cancer. We show that this approach leads to a distinctively activated TAM phenotype that can neither be categorized as conventionally “M1” nor “M2” but confers potent anti-tumor functions.

Results & Discussion

Design of multi-drug loaded CANDI nanoparticles

There were a few design criteria we had in mind for an improved nanomaterial drug delivery system as a driving force for multiple drug loading of nanoparticles. This included high therapeutic drug payload for multi-drug loading, long circulation times, effective uptake in tumor-associated macrophages, and non-toxicity. To optimize for these parameters, we initially explored different dextran, cyclodextrin (CD), and hybrid nanoparticles,¹² confirming that pure CD has acceptable characteristics of TAM recognition and drug-loading ability. Despite these optimizations, prior preparations of cross-linked CD nanoparticles (mean size 37 nm, penta-succinyl-β-CD) had limited payload capacity and solubility. To improve these parameters, we changed the nanoparticle size and improved solubility by custom synthesis of bi/tri-succinyl-β-CD (s-β-CD). Based on prior work on optimal size determinations for such materials,²⁹ we chose a mean size of 17 nm, similar to another dextran nanoparticle construct currently in clinical trials by us (NCT04843891). The 17 nm s-β-CD nanoparticle, compared to a 37 nm analog, has a ~2-fold higher surface area to volume ratio (0.35 for 17 nm nanoparticle vs. 0.16 for 37 nm nanoparticle), which facilitates drug loading. The resultant CANDI-300 series design thus had theoretical and pharmacological advantages (see below) over prior and other nanomaterials for TAM targeting. Subsequent experimental work was then done to confirm these advantages (Table S1).

Fig. 1 details the experimental approach. We hypothesized that the above-described myeloid cell targeting carbohydrate nanoparticle (CANDI) containing multiple therapeutic payloads could be used to affect the wiring at multiple immune-modulating pathways of TAM through complementary mechanisms. Until recently, it had been unclear which pathways to best modulate for maximum functional phenotypes. We had previously shown that TLR7/8 agonism and cIAP inhibition in tumor-associated macrophages leads to IL-12 up-regulation,²³ an effect that, unfortunately, also triggers IL-10/STAT3-induced resistance. This negative feedback loop prevents further IL-12 increases. For this reason, we focused here on triple combinations and adding JAK/STAT inhibitors to shut down the IL-10 axis. One challenge was to incorporate three distinct drugs (JAK/STAT inhibitor, non-canonical NF-κB inhibitor, and TLR agonist)^{25, 30} into the same nanoparticle. We identified a total of 6 prototypical small-molecule modulators for the three distinct pathways. These included LCL-161 (1), R848 (2), Ruxolitinib (3), Upadacitinib (4), BP-1-102 (5), and STX-0119 (6, Fig. 2).

Fig. 2: Core structure and scope of pharmaceutical payloads envisioned for HAMT therapies.

A. Synthetic strategy to form ~17 nm nanoparticles consisting of succinyl-β-cyclodextrin (s-β-CD) cross-linked by L-lysine linkers via EDC/NHS chemistry. Each nanoparticle contains an average of ~900 s-β-CD units and forms spherical nanoparticles capable of carrying small-molecule payloads. B. Different views of repeating CD units, including the guest-host interaction of a model therapeutic payload (Ruxolitinib, 3). C. Structures of all prototypical small-molecule (SM) modulators potentially affecting macrophage function (see B on how the lipophilic moieties fit into the s-β-CD). D. The images show the different nanoparticles with payload combinations investigated in this research. The numbers in each nanoparticle refer to the compounds from panel C. Yellow particles have only one payload, orange particles dual payload, and pink nanoparticles have triple payloads. The same scheme is used for subsequent figures.

We first synthesized and tested different analogs of succinyl-β-cyclodextrin (s-β-CD) by varying their degree of substitution (DS) to achieve optimal water solubility and stability (Fig. S1). Using the optimized DS 2.5 analog, we performed nuclear magnetic resonance (NMR) experiments in D₂O lacking and containing the selected pharmaceutical payloads to i) assess the dissolution of the hydrophobic payloads in D₂O determined by the presence of the aromatic protons in NMR titration experiments (Fig. S2), ii) identify which of the hosts protons undergo chemical shifts (inner cavity protons: H₃, H₅ and H₆) upon loading by one-dimensional ¹H-NMR, iii) correlate the interaction between the hosts protons to those of the guests protons by two dimensional-ROESY NMR, and iv) determine which functional groups have a higher affinity toward forming inclusion complexes with s-β-CD (Fig. S3-4). From these experiments, we selected the compatible payloads 1, 2, 3, and 4. We predicted that CANDI nanoparticles require at least an equimolar amount of CD units available to achieve complete dissolution of the payload combinations (Table S1 and Fig. S2).

Host-guest chemistry

The affinity of the selected payloads to form inclusion complexes with the CD unit was examined experimentally using three independent analyses. First, we performed 1D- and 2D- ¹H-NMR spectroscopy to monitor the changes in chemical shifts of both the cyclodextrin and the pharmaceutical payloads when forming inclusion complexes. Most payloads, excluding R848, underwent significant precipitation when dissolved with D₂O (3-26 mM, 10% d6-DMSO) in the absence of s-β-CD. Upon inclusion-complexation of the payloads, we were able to identify and correlate the relative binding strength of the different functional groups contained in each molecule to determine the payloads possible binding orientations (Fig. S3-4). Second, we performed UV-Vis spectroscopy and assessed changes in turbidity of loaded CANDI solutions to confirm the degree of payload solubility and to determine their loading capacity as a measurement of amount of payload per particle (Fig. 3 and Table S1). Third, we quantified the binding affinity (K_a) of each drug towards the s-β-CD host in stopped-flow experiments (Fig. S5 and Fig. 3).

Fig. 3: Triple small-molecule loading and release kinetics.

A. Dynamic light scattering (DLS) of empty (“0”) or triple-loaded HAMT (“1+2+3”; see Table S1 for detail). Drug loading at 0.3 μmol per mg of particle does not change the nanoparticle size (Z-Av.) or polydispersity index (PDI). B. Transmission electron microscopy (TEM) of multiple HAMT particles stained with aqueous uranyl acetate (2%). C. Turbidity assay to measure drug loading into the CANDI nanoparticle assessed by an increase in absorbance. We determined the loading capacity for HAMT to be ~0.19 mg of payloads 1+2+3 (0.5 μmol) per mg of CANDI nanoparticle. For subsequent biological studies, we loaded 0.11 mg (0.3 μmol) of 1+2+3 per mg of CANDI. D. Release kinetics of triple small molecules from HAMT using a closed-dialysis system with a porous membrane (3 kDa) in PBS (1x) at 37 °C (shown in picture). The cumulative release (%) of each drug was determined as the integrated area of the peak corresponding to each compound (diode array, UV) at nine time points (t = 0, 1, 1.5, 2, 2.5, 4, 5, 6 and 24 h) compared to the total amount of compound found in CANDI loaded with 1+2+3 (no membrane). Aliquots were analyzed by liquid chromatography coupled to a mass spectrometer (LC-MS) with a UV detector. k_off values were determined by fitting each curve to a one-phase decay. All three compounds show similar release kinetics and achieve a plateau at t = 4 h. All experiments were done in triplicates (n = 3).

The binding affinity of each payload in HAMT (1, 2 and 3) was determined from the association (k_on) and dissociation (k_off) rates at physiological pH (PBS).³¹ Using a fixed concentration of each payload, we measured the change in absorbance (λ = 532 nm) as a function of time in the absence or presence of solutions containing s-β-CD (k_on, Fig. S5). The fastest binding rate was determined for LCL-161 (k_on = 902 M⁻¹s⁻¹), followed by R848 (k_on = 18.8 M⁻¹s⁻¹) and Ruxolitinib (k_on = 8.8 M⁻¹s⁻¹ , Fig. S5). The dissociation rate (k_off) was estimated from the cumulative release of each component in a closed-dialysis setup (k_off = 0.81, 0.74 and 0.48 s⁻¹ for LCL-161, Ruxolitinib and R848 respectively, Fig. S5). We determined that LCL-161 exhibited remarkably strong binding affinity (K_a = 3.81x10⁶ M⁻¹) to s-β-CD, comparable to known adamantyl-containing payloads.^{19, 32} R848 and Ruxolitinib also exhibited moderately strong binding affinities (K_a = 4.75x10⁴ M⁻¹ and 3.28x10⁴ M⁻¹) within the expected range of previously studied β-CD-complexed systems (Fig. S5).^{33, 34}

We next aimed to increase the surface area to volume ratio by making a much smaller cyclodextrin particle than in a prior report.¹² A secondary objective was further improving pharmacokinetics and enabling better renal clearance.^{29, 35} For this reason, we screened multiple cross-linking reaction conditions by varying the amount of L-lysine linker (0.5-0.25 eq to carboxylate). The 1:4 CD to L-lysine ratio yielded stable particles with a ~17 nm hydrodynamic diameter (Fig. 2). We estimated that each nanoparticle could contain at least an average of ~900 s-β-CD units and form spherical nanoparticles capable of carrying up to 0.5 μmol of total payloads per mg of particle (Table S1).

Characterization of CANDI nanoparticles

Fig. 3 and Table S1 summarize the size distribution for the particle formulation characterized by dynamic light scattering (DLS) for empty, mono- and combinational therapies. The empty nanoparticle (CANDI^E) had an average size of 16.7±2.73 nm and a zeta potential of −8 mV. Loading of single payloads to saturation (0.5 μmols per mg of particle) led to small fluctuations in particle diameter: 18.8±0.65 nm (1), 18.3±1.63 nm (2), 16.9±3.1 nm (3), and 21.1±4.5 nm (4) with respect to the empty nanoparticle (Table S1). Conversely, loading experiments at isotonic conditions and at a physiological pH of 7.4 with payloads 5 and 6 resulted in unsuccessful dissolution and sedimentation, indicating sub-optimal loading and inefficient inclusion-complex formation with s-β-CD. Further investigations both simple (pH, salt conditions) and more complex (chemical modifications) are warranted to explore the full potential of drug payloads.

Particles loaded with two and three components showed similar results in size (1+2 d = 17.3±2.3 nm, 1+2+3 d = 17.3±1.8 nm, and 1+2+4 d = 18.5±11 nm) and polydispersity index (PDI, 1+2 = 0.258, 1+2+3 = 0.338 and 1+2+4 = 0.390, Table S1). Transmission electron microscopy (TEM) showed nanoparticles as single structures without any apparent sign of aggregation, especially when kept at physiological conditions (1x PBS, Fig. 3). We then developed a turbidity assay to measure the loading efficiency of individual or combination of drugs based on our previous NMR experiments (Fig. 3 and Table S1). In essence, nanoparticle loading via guest-host interactions resulted in the immediate dissolution assessed by the loss of turbidity. On average, loading of payloads 1, 2, 3, and 4 alone or combined up to 0.19 mg (0.5 μmol) per mg of CANDI particle yielded stable formulations (71.4% of the particles theoretical loading capacity). Due to the excellent loading ability and particle size of the 1+2+3 combination, we determined the cumulative drug release for each pharmaceutical component by LC-MS using a closed-dialysis set-up. Analysis of all time points resulted in a similar average half-life (t_1/2 = 1-1.5 h) for all three payloads reaching a release plateau after 4 h (Fig. 3).

Cell-based screens identify hits

Since the ~17 nm CANDI nanoparticle had not been tested biologically, we first determined whether it had macrophage affinity. We utilized an immortalized mouse macrophage cell line (iMAC) for these experiments and incubated these macrophages with different amounts of a fluorescently labeled CANDI^AF647 analog. Flow cytometry (data not shown) and fluorescence microscopy (Fig. S7) showed high uptake and internalized punctate structures in ~99% of all iMACs.

Having prepared different nanoparticle combinations, we next wanted to determine their effects on macrophage phenotypes (Fig. 4). Since IL-12 is a well-known myeloid cell-produced factor with potent anti-tumor activity, we tested the ability of drug-loaded CANDI to induce IL-12 in macrophages derived from IL-12 eYFP reporter mice by culturing whole bone marrow with recombinant macrophage colony-stimulating factor (M-CSF). The different amounts of drugs were chosen based on therapeutic efficacy,^{12, 23, 25, 36} synergy with each other, CANDI payload capacity and results from in vitro IL-12 induction screens (Table S1 summarizes the final amounts formally tested). Four different JAK/STAT inhibitors (Ruxolitinib (3), Upadacitinib (4), BP-1-102 (5), STX-0119 (6)) were tested, differing in their JAK1/JAK2/STAT3 selectivity,^{37, 38} pharmacological profile and clinical translation. Different cIAP inhibitors are also known and had previously been screened by us,^{24, 25} yielding LCL-161 as a top candidate. Finally, R848 is a prototype TLR7/8 agonist but has not progressed beyond early-stage clinical data in 2013.³⁹ Fig. 4 summarizes the results from these screens.

Fig. 4: Screening for therapeutic efficacy in cells.

A. Isolation and differentiation scheme of bone marrow-derived macrophages (BMDM) from IL-12 eYFP reporter mice with subsequent in vitro stimulation assay. B. Stimulation of BMDM with different nanoparticle formulations (see Fig. 2 and Table S1; single treatments: 0.5 μM LCL-161 (1), 0.32 μM R848 (2), 0.65 μM Ruxolitinib (3); same concentrations for combination therapies) leads to induction of IL-12 in vitro. Dual therapy (“1+2”) can be significantly boosted by adding Ruxolitinib (3) into the same nanoparticle (HAMT, “1+2+3”). Each data point represents a separate FOV with the % of eYFP positive cells of an average of ~1500 cells; * = p = 0.0131; **** = p < 0.0001; 100 ng/mL LPS, 50 ng/mL IFNγ. C. Representative microscopy images showing eYFP positive cells (corresponding to IL-12 induction) after cell stimulation assay in vitro (normalized contrast in 488 nm channel, 100 ms exposure time) as shown in panel B. Scale bar = 200 μm.

Empty nanoparticles showed no IL-12 positive cells, demonstrating the low background of the IL-12 assay, while the positive control using LPS/IFNγ resulted in an average of 5% IL-12 positive cells per well (p = 0.0078). Similar results were observed with R848 (2) loaded monotherapy particles. Dual loading of nanoparticles induced approximately 10% of cells to produce IL-12 (p < 0.0001). This number increased to ~30% for triple-loaded therapy (Fig. 4). The highest effect was identified for nanoparticles containing LCL-161, R848, and Ruxolitinib (HAMT; “1+2+3”), which was significantly higher than for any other combination or monotherapy (p < 0.0001). These results were also confirmed by flow cytometry analysis of bone marrow-derived macrophages (BMDMs) differentiated from C57BL/6J wild type mice (Fig. S8). Given these results, we performed subsequent in vivo experiments with this particular HAMT preparation. We also confirmed that HAMT up-regulates IL-12 in bone marrow-derived dendritic cells (BMDCs, Fig. S8) using the in vitro IL-12 induction assay by fluorescence microscopy. The strong induction of IL-12 in HAMT-treated cells was quantified by flow cytometry resulting in ~50% total IL-12 positive BMDCs. In contrast, cells treated with single-loaded CANDI particles were significantly lower, resulting in 30% for R848 (2) and almost no IL-12 induction for LCL-161 (1) and Ruxolitinib (3) relative to the control (PBS). We also performed dose-dependent toxicity experiments identifying 0.1 mg/mL as a highly efficient and non-toxic dose (Fig. S9).

Pharmacological behavior

To better understand the in vivo behavior of the CANDI-300 series, we performed biodistribution experiments with a ⁶⁴Cu-labeled version (Fig. S12). Whole body excretion was primarily via urine and feces. Taking into account the very small size of the nanoparticle (17 nm), the remaining whole-body radioactivity was ~25% injected dose within 1 day after administration and similar to other carbohydrate nanopreparations of similar size.⁴⁰ Major CANDI uptake was observed in the macrophage-rich organs of the reticuloendothelial system, such as the liver. This is characteristic of most materials with high molecular weights, such as antibodies, proteins, and nanoparticles. Tumoral uptake in whole tumors varied across animals and ranged from 6.54 to 1.32 %IDGT. Uptake in remaining organs (lung, heart, brain, digestive tract, muscle, fat and reproductive organs) was low. Autoradiography of removed MC38 tumors shows CANDI accumulation in peripheral zones, which contained the highest amounts of TAM. In these focal areas, uptake was similar to values observed in the liver.

Within tumor tissues, CANDI was almost exclusively associated with macrophages that also internalize dextran, a known macrophage marker (Fig. S14),⁴¹ and other myeloid cells like dendritic cells (Fig. S15). Nearly all HAMT-containing cells showed IL-12 induction (Fig. S14). We further investigated the mechanism of cellular CANDI uptake by performing typical uptake experiments in the presence of specific inhibitors (Fig. S19). The biggest effects were observed with scavenger receptor inhibition (Fucoidan).

Finally, we determined the toxicity of HAMT at the cellular (Fig. S9) and whole-body levels (Fig. S10-11). At the in vivo dose chosen (0.25 mg LCL-161, 0.1 mg R848, 0.2 mg Ruxolitinib), we did not observe any significant enzyme or electrolyte abnormalities. However, when the drug combo was given as free drugs (0.25 mg LCL-161, 0.1 mg R848, 0.2 mg Ruxolitinib in DMSO; equivalent dose to HAMT group), only 1 of 3 mice survived and showed extensive liver function (Fig. S10) and morphology (Fig. S11) abnormalities. These findings were not observed with the HAMT preparation.

Anticancer effects

We next determined the in vivo efficacy of HAMT in different mouse tumor models. We first performed in vivo efficacy in the subcutaneous implanted MC38 colorectal mouse model. For these experiments, 2×10⁶ cells were implanted into the flank of recipient mice. At days 8 and 12, with established tumors, HAMT was given systemically through intravenous injection (1 = 0.25 mg, 2 = 0.1 mg, and 3 = 0.2 mg, a total of 0.3 μmol payloads per mg of CANDI). Tumor sizes were measured three times a week by calipers. Empty CANDI nanoparticles were used as a control and compared directly to HAMT therapy. The data show no therapeutic effect on tumor growth for the empty nanoparticle control (Fig. 5). In contradistinction, systemic HAMT eradicated tumors in two-thirds of the mice and lead to a long-term durable response. The remainder ~30% of mice showed a partial response, with significant tumor growth delay compared to control-treated mice (Fig. 5). Interestingly, when complete responders were re-challenged on contralateral flanks with tumor cells at ~2 months following the initial tumor rejection, they were protected from future tumor growth, demonstrating that a long-term memory response to tumor cells persists and is driven by HAMT immunotherapy (Fig. 5). These results indicate that memory T cell effect can be induced by HAMT therapy resulting in long-lasting durable immunity. Immune stimulatory therapies such as HAMT potently activate professional antigen presenting cells (APCs) such as macrophages and dendritic cells. These cells are known to be required for generating robust T cell responses. Likewise, optimally stimulated APCs are known to stimulate effective T cell responses. From these experiments, we observed that the degree of T cell exhaustion in the TME of HAMT treated mice is far lower than the control. Finally, PD1 expressing CD8⁺ T cells that are not terminally exhausted remain sensitive to anti-PD1 therapy (Fig. S16).⁴²

Fig. 5: Therapeutic efficacy in murine tumor models (n = 106).

A. Treatment overview of colorectal MC38 murine model. MC38 tumor cells (2 x 10⁶) were injected into the flank of immunocompetent C57BL/6J mice (day 0). The first HAMT treatment (5 mg triple-loaded nanoparticle, 100 μL PBS (0.5x), red triangle) was injected after tumor growth on day 8, followed by a second treatment on day 12. Complete responders were re-challenged on day 49 and 60. B. Tumor growth curves of control mice receiving the empty nanoparticle (CANDI^E, n = 18). C. Tumor growth curves of mice treated with HAMT (n = 15). 4 mice with residual tumors due to partial response (those above dashed line) were sacrificed on days 13 and 19 so that tumors could be processed for flow cytometry. D. Survival curves for the two different cohorts show significantly longer survival for treated animals (p < 0.0001; n = 21 mice total). E. Re-challenge experiment in n = 13 mice previously treated with HAMT and complete tumor regression. Note that MC38 tumor-bearing mice previously treated with HAMT (green curves, n = 5) are immune to further tumor growth, suggesting T cell memory response (n = 8 control mice). F and G. Treatment overview of flank melanoma (F) and metastatic (G) B16-F10 murine tumor model. In F, 0.5 x 10⁶ fluorescent B16-F10 H2B-mApple cells were injected into the flanks of C57BL/6J mice (n = 41, day 0). After 7 days of tumor growth, mice received a total of four intravenous doses of HAMT (5 mg triple-loaded nanoparticle, 100 μl PBS (0.5x), red triangle) over a period of 2 to 3 weeks. Additional anti-PD1 treatment (200 μg, 50 μl PBS, green triangle) was administered intraperitoneally on the same days as the first two HAMT treatments. In G, 0.2 x 10⁶ fluorescent B16-F10 H2B-mApple cells were injected into mice via tail-vein injection (n = 12, day 0). Mice received a total of three intravenous HAMT or combination treatments with anti-PD1 antibody (amount same as in F) over a period of 2 weeks before sacrificing animals for lung clearing and imaging. H. Tumor growth curves of treatment groups of B16-F10 flank melanoma mice. Control mice (n = 12) received PBS injections, other cohorts anti-PD1 antibody alone (n = 5), HAMT (n = 12) or HAMT + anti-PD1 combination (n = 12) treatment. I. Survival curve of B16-F10 flank tumor model (n = 26 mice) shows significantly longer survival for treated animals: Median survival for untreated mice was 19 days, anti-PD1 treatment alone 28 days, HAMT alone treatment 31 days, and combination treatment 62 days with 3 mice that showed complete response after therapy.

In the metastatic model (Fig. 5), we injected B16-F10 melanoma cells intravenously on day zero and gave three courses of HAMT intravenously over the next week, with or without anti-PD1 therapy. On day 14, animals were sacrificed, and lungs were removed after cardiac perfusion to determine metastatic tumor burden (Fig. S13). The data show greatly diminished tumor growth in animals that had received HAMT (p = 0.0121) or a combination of HAMT with anti-PD1 (p < 0.0001) therapy without signs of toxicity.

Mechanism of action

In the next set of experiments, we explored how the HAMT therapies were so effective in vivo. We first turned to intravital microscopy (IVM) to identify which cells take up the nanomaterial. IVM data showed that 24 hours after administration, the CANDI particles mostly localized in TAM. Fig. S14 shows that the nanoparticles co-localize with systemically injected 2M MW dextran-pacific blue (PB), a standard marker for macrophages.⁴¹ These results were confirmed by flow cytometry analysis of immune cells in MC38 tumors 24 hours after injection of fluorescent CANDI^AF647 showing nanoparticle uptake into myeloid cells but not tumor cells or T and B cells (Fig. S15).

We next determined in vivo whether there was IL-12 induction within the tumor microenvironment as suggested by in vitro experiments. Intravital imaging data in IL-12 eYFP reporter mice shows that this was indeed the case (Fig. 6). The baseline tumor microenvironment is mainly devoid of IL-12 signals. However, within 48 hours of HAMT treatment, there was a marked up-regulation of IL-12 in the entire tumor microenvironment observed in all animals tested (Movie 1). We also confirmed that the HAMT treatment cohort had fewer terminally exhausted T cells (PD-1⁺ TIM-3⁺) in the tumor microenvironment, suggesting a more effective T cell anti-tumor response (Fig. S16).

Fig. 6: Insight of HAMT efficacy by serial intravital imaging in the MC38 model.

Dorsal window chambers were implanted into IL-12 eYFP mice, followed by tumor cell injection into the window chambers (10⁵ MC38 H2B-Apple cells, red). Imaging was performed serially before drug administration (pre) and after drug administration (t = 10 min to 5 h), and again 24 and 48 hours after systemic HAMT administration. Immediately after systemic administration, HAMT was detected in tumor microvessels, followed by cellular uptake 1 to 5 hours after intravenous administration. For cellular distribution of HAMT, see Fig. S14 and Movie 1. Note the marked induction of IL-12 after HAMT administration. For mechanistic insight, see Fig. S17-20. Scale bar = 40 μm.

Fig. S17 shows pathway analysis of single compounds in HAMT therapy by western blotting. As expected, TLR agonism through R848 led to p38 MAPK (Thr180/Tyr182) activation. LCL-161 showed an increase in NIK levels, indicating activation of non-canonical NF-κB signaling as constitutive proteosomal degradation of NIK is inhibited and signaling therefore activated. Finally, IL-10 is a known negative regulator of IL-12 signaling. Therefore, we hypothesized that IL-10 would diminish activating signals to macrophages. This was indeed shown by blocking the phosphorylation of STAT3 (Tyr 705) with Ruxolitinib in the presence of IL-10 (Fig. S17).

To determine the effects of pathway modulation on TAM, we next performed cytokine analysis (Fig. S18). These data showed that acute inflammatory cytokines (e.g. IL-12 and TNFα) and myeloid activation markers (MCP-1, MIP-1α, and MIP-1β) were highly up-regulated. Finally, we performed RNA sequencing on HAMT-stimulated bone marrow-derived macrophages (Fig. S18), which showed a TAM phenotype characterized by an over-expression of IL-12, surface expression of macrophage receptor with collagenous structure (MARCO), DC-SIGN, and SIGNR7. Interestingly, interferon-stimulated genes (ISG) were largely absent in this TAM phenotype, suggesting that IL-12 expression can occur independently of ISG responses. Furthermore, interferon-inducible inhibitory mechanisms such as Pdl1 (Cd274), Ido1, and Ido2 were absent in HAMT-treated macrophages. The mechanism of action is summarized in Fig. S20.

Here we describe the design and synthesis of a carbohydrate nano-carrier able to encapsulate multiple different small-molecule drugs into a single formulation to i) concentrate drug combinations in tumor-associated macrophages and ii) to affect multiple immune modulatory pathways in targeted cells. We found that this multi-target engagement is a highly effective method to modulate TAM function and is much more effective and better tolerated than single-agent therapeutics since the latter would require much higher doses. The reasoning behind developing HAMT is similar to the success stories of highly effective retroviral combination therapies. The combination of three drugs allows stimulation or antagonism of multiple pathways, likely eliminating resistance mechanisms (e.g. counteractive IL-10 effects during IL-12 production),³⁶ and the use of much lower drug doses of each component. This method of simultaneous targeting of linked immune stimulatory and counter-regulatory pathways in one therapeutic particle can potently reprogram TAM towards anti-tumor phenotypes.

A variety of different approaches have previously been used to deliver IL-12 to tumors using protein, viral vectors, and mRNA.^{6, 25, 43, 44} Many of these prior approaches have failed clinically for efficacy or toxicity reasons⁴⁵ while murine work supports the efficacy of IL-12-based therapies.⁴⁶ However, many of these prior approaches rely upon the direct delivery or production of IL-12 through therapeutic vectors. The challenge then is to develop more efficient and locally active IL-12 therapeutics. Immune signaling is complex but involves multiple regulatory pathways acting in concert to provide strong immune stimulation. For example, T cell stimulation requires at least three discrete signals such as T cell receptor (signal 1), co-stimulation (signal 2), and cytokines (signal 3). These signals are provided in context by activated antigen-presenting cells such as dendritic cells or macrophages, and perhaps the failure of earlier IL-12 modulating therapeutics was due to the IL-12 signal acting in isolation. The work described here is fundamentally different from the above approaches in that we stimulate local myeloid cells rather than deliver exogenous proteins, fusion proteins, protein-encoding sequences, or engineered cells. This approach has multiple advantages, including simplicity, much higher efficacy, and importantly, multi-pathway modulation to more effectively stimulate myeloid cells, all while lacking measurable toxicity. Oddly, the HAMT approach using a specific triple-drug combination did not enhance other co-stimulation (CD80/86), cytokines (IL-15, IL-18, IL-27), and chemokines (Cxcl9/10) and was largely devoid of ISGs (Fig. S18); nonetheless conferred potent anti-tumor phenotypes in vivo. A possible explanation for this observation could be that normally TAM antigen presentation to cytotoxic T cells drives T cell exhaustion,⁴⁷ activation of IL-12 in TAM without enhanced antigen presentation could therefore be a feature of HAMT therapy. HAMT therapy could also have an effect on naïve T cell priming for immune cold tumors whereby, defective antigen presentation early on results in diminished T cell responses. Our data with CD8⁺ T cells shows that HAMT treatment reduces terminally exhausted T cell numbers in tumors, so in fact, it does not trigger exhaustion. T cell-driven effects seen in our models likely rely upon efficient antigen-presenting cell (macrophage and dendritic cell) activation within the tumor microenvironment as HAMT therapy does not accumulate within T cells nor tumor cells in vivo (Fig. S15). In summary, these multiple mechanisms are likely the reason why the featured approach is superior to simple IL-12 delivery, which is often modest and by itself may lead to T cell exhaustion since T cell stimulatory signals in isolation do not produce robust T cell responses.

While we tested multiple different triple combinations, we found that loading CANDI particles with LCL-161 (1), R848 (2), and Ruxolitinib (3) (HAMT) lead to the most stable nanoparticle formulation and strongest IL-12 induction. The cIAP inhibitor LCL-161 essentially works through the non-canonical NF-κB pathway, while the R848 agonist works through the canonical NF-κB pathway²³ by stimulating TLR7/8. From a host-guest chemistry point of view, we observed a two-phase drug release behavior which was expected for a drug delivery systems that utilize macromolecular host-guest interactions such as small-molecule payload (guest) complexing to cyclodextrins (host).^{13, 48} The two-phase release can be often desirable to provide an initial rapid therapeutic effect followed by a slower sustained and prolonged release.⁴⁹ The initial rapid-release phase could be mainly attributed to weaker host-guest interactions occurring on the particles surface. There, the molecules can undergo fast adsorption-desorption cycles and are prone to competitive displacement from other targets present in solution. From our 2D-ROESY NMR experiments, we confirmed that the small-molecule payloads in HAMT (LCL-161, R848, and Ruxolitinib) undergo multiple orientations during complexation with the cavity of s-β-CDs (Fig. S3-4). We observed that cyclohexyl group in LCL-161 and cyclopentyl in Ruxolitinib, as well as ethylene glycol groups found in R848, displayed the strongest spatial interactions with the protons found in the hydrophobic cavity of the s-β-CDs and induced a chemical shift in H_a and H_b (Fig. S3-4). The slower, prolonged release of the payloads can be attributed to molecules found closer to the particles core governed by strong inclusion complexes or those that are surrounded by higher number of free s-β-CDs.

It was surprising that JAK inhibition through Ruxolitinib significantly improved the efficacy of IL-12 induction at much lower doses. Stimulation of macrophages with TLR agonists results in the secretion of TNFα, IL-6, and IL-12, which is also controlled by multiple feedback pathways. Importantly, macrophages also produce IL-10, inhibiting pro-inflammatory cytokine production via the JAK/STAT3-dependent pathway.³⁰ Prior work has shown that Ruxolitinib can block the IL-10-mediated feedback inhibition on cytokine transcription in macrophages.³⁶ Overall, these results suggest that inhibition of JAKs may increase the inflammatory potential of macrophages stimulated with TLR agonists.

The current research shows that combinatorial loading of myeloid-avid nano-carriers offers an attractive venue for more efficient and potentially complementary cancer therapy. A particularly interesting observation was that the pharmacologic manipulation resulted in TAM phenotypes orthogonal to biological systems. HAMT-induced macrophage re-programming can be neither defined as the “M1” or “M2” axis. Instead, we show that this approach leads to a distinctively activated TAM phenotype that confers anti-tumor functions. This TAM phenotype is characterized by high IL-12 production, surface expression of macrophage receptor with collagenous structure (MARCO), DC-SIGN, and SIGNR7; however, interferon-stimulated genes (ISG) were largely absent in this TAM phenotype, suggesting that IL-12 expression can occur independently of ISG responses. Since the M1 and M2 phenotype classification does not capture the optimal anti-tumor TAM phenotype, we argue that IL-12 induction by myeloid cells should be a guiding principle of anti-tumor TAM re-programming. Using high throughput, image-based screening we identified HAMT as a therapeutic combination that most strongly induces IL-12 and exhibits potent efficacy in multiple pre-clinical cancer models. This platform for TAM re-programming agent selection has the potential to fine-tune desirable effector functions while avoiding undesirable features such as compensatory immune suppression (e.g. PD-L1). While we show high efficacy in multiple murine models, we anticipate future research and optimization to result in further improvements. For example, broader screens might identify further synergistic combinations, while radiometric loading could be optimized to titrate macrophage activation. Finally, it is possible to target CANDI formulations to more specific myeloid cell subtypes¹ for further enhanced efficacy.

Conclusion

The development of the HAMT carbohydrate nano-carrier, represents an important approach to cancer therapy. By encapsulating a specific triple-drug combination, HAMT demonstrates significant potential in modulating tumor-associated macrophages (TAMs) towards anti-tumor phenotypes. This innovative strategy combines the stimulation of multiple immune modulatory pathways within targeted cells, resulting in superior efficacy and enhanced tolerance compared to single-agent therapeutics. Importantly, HAMT's unique mechanism of action avoids the pitfalls of previous IL-12 delivery methods, which often acted in isolation and risked inducing T cell exhaustion. Instead, HAMT stimulates local myeloid cells, achieving multi-pathway modulation and potent activation without measurable toxicity. The choice of this triple-drug combination, LCL-161, R848, and Ruxolitinib, provides a stable nanoparticle formulation with strong IL-12 induction, making it a promising candidate for cancer therapy.

Methods/Experimental

Synthesis and characterization of HAMT

Materials.

All reagents and solvents were purchased from Thermo Fischer or Sigma-Aldrich and used as received. Small molecules (LCL-161 (1), R848 (2), Ruxolitinib (3), Upadacitinib (4), BP-1-102 (5), STX-0119 (6)) were purchased from MedChem Express, dissolved in DMSO accordingly and used without any further purification. MilliQ water was obtained from the Waters filtration system.

Synthesis of TAM targeting CANDI nanoparticle.

The synthesis of cyclodextrin nanoparticles (CANDI) was further developed from a previously reported method.²³ In contradistinction, we used better controlled s-β-CD and made much smaller nanoparticles (17 nm vs. 37 nm). We de novo synthesized s-β-CD with a well-defined degree of substitution (DS 2.5) since commercially available products had a large degree of substitution and performed differently from batch-to-batch. Briefly, β-cyclodextrin (Sigma, 1.3 g, 1.2 mmol) was dried at 60 °C for 72 h.⁵⁰ In a pressure vessel (50 mL) charged with a magnetic stirrer, succinic acid (Sigma, 1 g, 8.2 mmol) was dissolved in water (1 mL) followed by sodium hypophosphite monohydrate (Sigma, 65 mg, 0.6 mmol), and the previously dried β-cyclodextrin (1.3 g). The reaction was heated to 120 °C under constant stirring and monitored by LC-MS (ELSD signal) until reaching the desired degree of substitution (t = 24 h). The clear solution was cooled to room temperature and triturated with ethanol (50 mL, 200-proof). The solids were sonicated and washed with abundant ethanol (~150 mL), filtered, and dried at 60 °C for at least 24 h. The desired succinyl-β-cyclodextrin (DS ~2.5, 1.3 g) was obtained as white crystals (75% yield). ¹H NMR (400 MHz, D2O) δ = 5.05 (s, 7H, H1) reference proton, 4.56–5.53 (m, 2H, Ha), 4.26–4.21 (m, 2H, Hb), 4.04–3.98 (m, 2H, H5’), 3.94–3.90 (m, 12H, H3 & H5), 3.85–3.75 (m, 10H, H6), 3.65–3.56 (m, 21H, H2 & H4), 2.68–2.63 (m, 10H, H_suc) ppm. For the synthesis of CANDI nanoparticles, the succinyl-β-cyclodextrin (DS ~2.5, 250 mg, 1.0 eq to carboxylate) was dissolved in MES buffer (6 mL, 50 mM, pH = 6.5) and activated with N-(3-(dimethylamino)propyl)-N′-ethyl carbodiimide hydrochloride (EDC) (Fisher; 1.5 g, 10.0 eq to carboxylate) and N-hydroxysuccinimide (NHS) (Sigma; 550 mg, 5.0 eq to carboxylate) for 30 min at 25 °C under constant stirring in a closed 20 mL scintillation vial charged with a magnetic stirrer. A solution of L-lysine (Sigma; 35 mg, 0.25 eq to carboxylate) in MES buffer (1.5 mL) was added in a drop-wise manner, and the reaction was allowed to stir for 18 h at 25 °C. The particles were precipitated with absolute ice-cold ethanol (70 mL, 99.9%), yielding a white precipitate that was decanted and dissolved in water (14 mL). The particles were purified with 10 kDa MWCO centrifugal filters (Amicon; 10,000 g for 8 min), and lyophilized for 48 h. The dry particles (~320 mg) were characterized by DLS (2 mg/mL, 1x PBS) and Zeta potential (2 mg/mL, 0.1x PBS) and stored as solids at −20 °C until further use.

Small-molecule loading of nanoparticles (Table S1).

A solution of empty CANDI (CANDI^E; 5 mg) in PBS (0.1x, 90 μL) was used for payload loading to a final DMSO concentration of 10%. The following nanoparticle compounds were prepared: CANDI-200 containing LCL-161 (0.5 mg) and R848 (0.2 mg); CANDI-301: BP-1-102 (0.41 mg), LCL-161 (0.25 mg), and R848 (0.1 mg); CANDI-302 containing Ruxolitinib (0.2 mg), LCL-161 (0.25 mg), and R848 (0.1 mg); CANDI-303 containing Upadacitinib (0.25 mg), LCL-161 (0.25 mg), and R848 (0.1 mg). The solutions were vortexed rapidly until the complete dissolution of the drugs. All solutions were filtered through a 0.22 μm sterile filter (VWR) and used immediately for characterization, in vitro assays, or stored at −20 °C until further use.

Turbidity assay.

CANDI^E stock solutions (0–100 mg/mL, 0.5x PBS) were prepared at pH = 7.4 and small molecules were dissolved in DMSO (200 mM LCL-161, R848, Ruxolitinib, Upadacitinib; 100 mM BP-1-102) to prepare payload stocks. Loading of CANDI particles with payloads (8 mM and 4 mM, respectively, 10% DMSO) was quantified by absorbance scan measurement (λ_abs = 350–700 nm) after thoroughly mixing the solutions. Total loading of the payload was determined by the complete loss of absorbance. Data were normalized to payload-free control and experiments were performed in triplicates (n = 3).

Loading assessment by nuclear magnetic resonance (NMR) and liquid chromatography-mass spectrometry (LCMS).

NMR spectra were recorded on a Bruker Avance UltraShield 400 MHz spectrometer. ¹H NMR chemical shifts are reported in ppm relative to SiMe₄ (δ = 0) and were referenced internally concerning residual protons (δ = 4.79 for D₂O). Peak assignments, calculated chemical shifts and peak integrals are based on reference solvent peaks. Two-dimensional Rotating Frame Overhauser Enhancement Spectroscopy (2D-ROESY) experiments were performed to assess the interactions between dipolarly coupled hydrogens, and integrals were normalized to the reference hydrogen (H¹) of the s-β-CD. All experiments were performed in D₂O (0.7 mL) at a fixed s-β-CD concentration (26 mM) with 10% (CD₃)₂SO. High-performance liquid chromatography-mass spectrometry analysis (HPLC-MS, LCMS) was performed on a Waters instrument equipped with a Waters 2424 ELS Detector, Waters 2998 UV-Vis Diode array Detector, and a Waters 3100 Mass Detector. Separations employed an HPLC-grade water/acetonitrile (0.1% formic acid) solvent gradient with XTerra MS C18 Column, 125Å, 5 μm, 4.6 mm X 50 mm column; Waters XBridge BEH C18 Column, 130Å, 3.5 μm, 4.6 mm X 50 mm.

Loading Capacity.

The total amount of payload (mg or μmols) loadable in one mg of particle (CANDI). Assuming complete cross-linking and semi-quantitative yield of CANDI synthesis (> 95% weight recovery), we estimated the following values to determine the loading capacity:s-β-CD per particle = ~900 s-β-CD content per mg CANDI = 87%, maximal theoretical loading capacity per mg of CANDI (0.7 μmols). For HAMT = LCL-161, R848, Ruxolitinib 1+2+3 molar ratios (1.6:1:2); ~0.19 mg of payloads (0.5 μmol) per mg of CANDI nanoparticle by turbidity assessment (Fig. 3).

Fluorescent CANDI analogs.

Lyophilized CANDI^E was dissolved in carbonate buffer (0.1 M, pH = 8.5) and AF647 succinimidyl ester (ThermoFisher, 2 mg/mL in DMSO) was added to achieve a final concentration of 50 μM. The reaction was stirred for 45 min at 37 °C in a thermocycler (600 rpm). The labeled nanoparticles were purified by buffer exchange into water against 10 kDa MWCO centrifugal filters (Amicon; 10,000 rcf for 5 min; 300 μL water per wash, 3-4x) and the final products were diluted with water or PBS to a final concentration of 50 mg/mL and filtered through a 0.22 μm sterile filter (VWR) prior to use.

Synthesis of ⁶⁴Cu-CANDI.

CANDI particles (60 mg) were dissolved in carbonate buffer (0.1 M, pH = 8.5, 500 μL 0.5x) and DOTA-NHS ester (30 mg, Macrocyclics) was added and stirred at 37 °C for 45 min in a thermocycler (600 rpm). The DOTA-functionalized particles were purified by buffer exchange into water against 10 kDa MWCO centrifugal filters (Amicon; 10,000 rcf for 5 min; 300 μL water per wash, 3-4x) and lyophilized for 24 h to obtain a white powder (50 mg). The DOTA-CANDI (50 mg) were dissolved in citrate buffer (1 mL, pH = 4.5) and ~6 mCi of radioactive ⁶⁴CuCl₂ in 0.1 NHCl was added. The CANDI particles were labelled at 70 °C for 45 min in a thermomixer (700 rpm). After labelling, the particles were were purified by buffer exchange into water against 10 kDa MWCO centrifugal filters (Amicon; 10,000 rcf for 5 min; 300 μL water per wash, 4x). The total radioactivity for the washed ⁶⁴Cu-CANDI was measured (2.9 mCi in 1 mL, 50% loading efficiency). The total dose was divided into 9 injectable doses (each 100 μL of 300 μCi).

Physico-chemical characterization of CANDI nanoparticles

Particle size and surface charge for all nanoparticle formulations were determined by dynamic light scattering (DLS) and zeta potential measured on a Malvern Zetasizer APS at 5 mg/mL in PBS (1x) and 2 mg/mL in PBS (0.1x), respectively measured in DTS1170 cuvettes (Malvern) at 25 °C. All absorbance and fluorescent spectra (e.g, CANDI^AF647 analogs) were performed with a multimode microplate reader (Tecan, Spark 500) using 96-well transparent bottom black polystyrene microplates (Corning).

Drug release kinetics.

Kinetics of drug release was performed in a closed dialysis set-up employing a 3 kDa molecular weight cut-off membrane (Pur-A-Lyzer^™ Midi Dialysis Kit). Solutions of CANDI^E (50 mg) were loaded with LCL-161 (2.5 mg), R848 (1 mg), and Ruxolitinib (2 mg) in PBS (1x, 1 mL) containing 10% DMSO and were dialyzed against PBS (1x, 6 mL) at 37 °C under constant stirring (600 rpm). The percentage of eluted molecules was quantified by analysis of the liquid chromatographs at different time points (t = 0, 1, 1.5, 2, 2.5, 4, 5, 6, and 24 h), injecting a total of 60 μL aliquots into an LC-MS and subsequently replacing the system with additional 60 μL of PBS. Each payload was identified by its distinctive retention time (R848 = 0.78 min, Ruxolitinib = 0.95 min, and LCL-161 = 1.01 min) and mass-to-charge ratio (ES⁻: R848 = 313, Ruxolitinib = 305 M- and LCL-161 = 499). The cumulative drug release was determined as the ratio of the integrated area under the curve for each eluted peak to the total area under the curve of chromatographs obtained from the non-dialyzed solutions. All experiments were performed in triplicates (n = 3).

Transmission Electron Microscopy.

HAMT 1+2+3 particles were freshly prepared (50 mg/mL, PBS 1x) and diluted with water to a final concentration of 0.1 mg/mL. The particle solution was was charged on a TEM grid for 1 min and treated with a 2% aqueous uranyl acetate solution for 15 min, followed by three washing steps with ultra-pure water (x3). Imaging was performed in a transmission electron microscope (JEOL 2100).

Cell experiments

Immortalized cell lines.

The immortalized murine bone marrow-derived macrophages (iMACs)⁵¹ were acquired from Charles L. Evavold (Ragon Institute, Harvard University). B16-F10 cells were obtained from ATCC. MC38 cells were obtained from Kerafast (ENH204-FP). Fluorescent versions of MC38 and B16F10 for intravital microscopy were made using an H2B-mApple vector. Specifically, cells were transfected with pLVX-H2B-mApple lentiviral vector (Clonetech) in the presence of 10 μg mL⁻¹ polybrene (Santa Cruz Biotech). They were then selected using 3 μg mL⁻¹ puromycin.⁵²

All cell lines were cultured following standard cell culture protocols (Table S4). Briefly, iMACs and B16-F10 cells were plated and grown in Dulbecco’s Modified Eagle Medium (DMEM, Corning) supplemented with 10% Fetal Bovine Serum (Corning) and 1% Penicillin Streptomycin (Corning) at 37 °C and 5% CO₂ and MC38 cells were cultured in Iscove’s Modification of DMEM (Corning). Upon reaching confluency, cells were split using 0.05% Trypsin / 0.53 mM EDTA (Corning), and all in vitro assays were performed after the cells reached 90% confluency. Prior to cell culture application, all CANDI preparations were filtered through a 0.22 μm sterile filter (VWR).

Bone marrow-derived cells.

Murine bone marrow-derived cells were isolated from IL-12 eYFP reporter or wild type C57BL/6J mice. To obtain the whole bone marrow, femurs were prepared and flushed with sterile PBS using syringes and a 28-gauge needle. RBC Lysis Buffer (BioLegend) was then used according to manufacturer’s instructions to lyse red blood cells. The remaining cells were counted using a Neubauer chamber and seeded into either transparent (NEST, flow cytometry analysis) or black (Ibidi, glass bottom for imaging) 96 well plates at a density of 1 x 10⁵ cells per well. Bone marrow-derived macrophages (BMDMs) were differentiated by adding 50 ng/mL recombinant murine M-CSF (BioLegend) to cell culture media for 7 days. To obtain bone-marrow derived dendritic cells (BMDCs), 300 ng/mL recombinant mouse Flt3L (BioLegend) and 50 ng/mL GM-CSF were added into RPMI 1640 with L-glutamine cell culture media (optimized with 25 mM HEPES, 10% FCS, nonessential amino acids, sodium pyruvate, B-ME and penicillin/streptomycin) for 9 days. Media was added every 3-4 days.

Cell viability assay.

For testing immortalized cell lines, iMACs were seeded in 96 well plates at a density of 8000 cells per well and incubated for 24 h at 37 °C and 5% CO₂ before use. Stock solutions of compounds in DMSO (200 mM) and in different CANDI nanoparticles (Table S1) were prepared and then diluted in cell culture medium to desired concentrations (0.6 μg/mL to 12.5 mg/mL, DMSO 0.5%). Cells were incubated for 2.5 h with nanoparticles before the medium was exchanged. Cells were further incubated for 48 h at 37 °C and 5% CO₂ before adding MTT solution (5 g/L in FluoroBrite DMEM, 10% final) to each well. After 3 h, the supernatant was carefully removed, and metabolized formazan was dissolved with isopropyl alcohol. Plates were shaken at 500 rpm on a microplate shaker (VWR) for 30 min, and the absorbance of each well was measured (λ_abs = 550 nm). Duplicates of triplicates were sampled for each concentration, and IC₅₀ values were calculated.

Live-cell microscopy.

Cells were treated with various combinations of CANDI loaded with small molecules (0–0.5 μM, DMSO < 0.5%) for 24 h by adding prepared stock solutions to cell culture media. Before imaging, cells were stained with Hoechst 33342 (15 μg/mL, Thermo Fisher) according to the manufacturer’s protocol. Cells were imaged in a 96-well plate. Fluorescence microscopy was performed using an IX81 inverted fluorescence microscope (Olympus, Tokyo, Japan) equipped with a motorized stage (Renishaw, Wotton-under-Edge, England, UK) and fitted with an ORCA-Fusion Digital CMOS camera (Hamamatsu Photonics, Hamamatsu, Japan). Using CellSens Dimension 3.1.1 software (Olympus), multiple fields of view were acquired for each sample with a UPlanSApo ×10 (numerical aperture (NA) 0.75, Olympus) or a UPlanSApo ×40 air objective (NA 0.95, Olympus). In addition to brightfield, five fluorescent channels were acquired: DAPI (345/455), GFP (489/508), YFP (550/565), CY3 (550/565), and CY5 (625/670) were excited with the appropriate optical filters.

Flow cytometry.

Bone marrow-derived cells of IL-12 eYFP reporter mice were stimulated o/n with the respective drug combinations, then trypsinized and washed with PBS. Next, the cells were stained using AquaAmine LiveDead Fixable viability stain (Thermo Fisher) diluted in PBS, followed by treatment with Fc block (BioLegend) and fluorochrome-conjugated antibodies (Table S2-S3) diluted in FACS buffer (1x PBS, 2 mM EDTA, 2% FBS). For intracellular cytokine staining of primary bone marrow-derived cells, samples were incubated for 6 hours with GolgiPlug (BD Biosciences, 1 μL/mL of culture media) after stimulations. Cells were then surface stained and, if necessary, fixed and permeabilized using the BD Cytofix/Cytoperm kit (BD) according to the manufacturer’s protocol and stained for intracellular cytokines. For flow cytometry measurements, cells were resuspended in a FACS buffer. All conditions were measured in triplicates in Attune NxT flow cytometer (Thermo Fisher), and the data was analyzed using FlowJo 10 software (TreeStar).

RNA sequencing.

Bone marrow-derived macrophages were isolated and differentiated as previously described. Cells were then stimulated for 24 h with desired treatments (PBS; HAMT; LPS/IFNγ) to induce activation and RNA was isolated using the RNeasy Plus Micro Kit (Qiagen). Final RNA concentration was determined by absorbance (Nanodrop), and samples were stored at −80 °C until shipment for sequencing (NovoGene).

In vivo experiments

Mouse models.

All mice (n = 123) were bred and housed under specific pathogen-free conditions at the Massachusetts General Hospital (MGH). Experiments were approved by the MGH Institutional Animal Care and Use Committee (IACUC) and were performed in accordance with MGH IACUC regulations. IL-12p40-eYFP mice (n = 17) were used for IL-12 induction experiments. C57BL/6J mice (n = 106) were utilized for MC38 and B16-F10 tumor implantations (Table S5).

Biodistribution experiments.

Mice (C57BL/6J; n = 6) received tail-vein injections of ~100 μCi ⁶⁴Cu-CANDI with HAMT under anesthesia (2% isoflurane with 2 L/min O₂). Whole-body biodistribution studies were performed 24 h after administration. Mice were euthanized and perfused with PBS through a left ventricle prior to organ harvesting. Excised organs were weighed and subjected to radioactivity measurement using a γ counter (1480 Wizard 3-in., PerkinElmer, Waltham, MA). Biodistribution data were obtained after corrections of radioisotope decay and residual activity at the injection site and expressed as percent injected dose per gram tissue (%IDGT)). Autoradiography of tissues was performed using a storage phosphor screen in a cassette (GE Healthcare) for ~ 60 h and read with a Sapphire Biomolecular Imager (Azure Biosystems).

Toxicity experiments.

Wild type C57BL/6J mice (n = 12) were injected with different drugs via the tail vein. Twenty-four hours later mice were sacrificed. Blood was drawn for comprehensive testing (serum analytes, blood counts, metabolic parameters). Mice were then perfused with PBS though a left ventricle before collection of the liver for histology. Livers were fixed in formalin solution (10%) overnight before washing in ethanol (70%), embedding in paraffin, and processing for hematoxylin-eosin staining.

Tumor cell implantation.

MC38 and B16-F10 cells were implanted at 2 × 10⁶ cells and 0.5 × 10⁶ cells, respectively, in the flank of C57BL/6J mice, and tumors were allowed to grow for at least 1 week before treatment. Tumor size was at least 50 mm³ before the initiation of therapy for the MC38 model. For the metastatic model, B16-F10 cells were injected intravenously at 0.2 × 10⁶ cells in sterile PBS via tail vein injection.

Drug treatment.

All CANDI preparations were administered by tail-vein injection (100 μL PBS 0.5x, pH = 7.4) containing 5 mg nanoparticle. Before injection, all solutions were sterilized by filtration through a 0.22 μm sterile centrifugal filter (VWR), vortexed, and used promptly or frozen at −20 °C.

Intravital microscopy.

Dorsal windows were implanted into IL-12 eYFP reporter mice.^{53, 54} All confocal images were collected using a customized Olympus FV1000 confocal microscope (Olympus America). A 2x (XLFluor, NA 0.14), a 4x (UPlanSApo, NA 0.16), and an XLUMPlanFL N 20x (NA 1.0) water immersion objective were used for imaging (Olympus America). MC38 H2B-apple tumor cells, HAMT^AF647, and vascular probes were excited sequentially using a 405 nm, a 473 nm, a 559 nm, and a 633 nm diode laser, respectively, in combination with a DM-405/488/559/635 nm dichroic beam splitter. Emitted light was further separated by beam splitters (SDM-473, SDM-560, and SDM-640) and emission filters BA430-455, BA490-540, BA575-620, and BA655-755 (Olympus America). Confocal laser power settings were carefully optimized to avoid photobleaching, phototoxicity, or damage to the tissues. FIJI (ImageJ, 2.9.0/1.53t) was used for image analysis. HAMT^AF647 was administered as a single injection containing a mixture of HAMT 1+2+3 (5 mg per injection, 100 μL) and CANDI^AF647 (5 mg, 100 μL).

Statistical analysis.

All statistical data analyses were performed using GraphPad Prism 9 software, and results are expressed as mean ± standard deviation. We used a 2-tailed Student’s t-test and one-way ANOVA followed by Bonferroni’s multiple comparison tests for normally-distributed datasets. We performed non-parametric Mann-Whitney or Kuskal-Wallis tests when variables were not normally distributed. For survival analysis, p values were computed using the Log Rank test. p values > 0.05 were considered insignificant (n.s.), and p values < 0.05 were considered significant. * p value < 0.05, ** p value < 0.01, *** p value < 0.001, **** p value < 0.0001.