An in situ depot for the sustained release of a TLR7/8 agonist in combination with a TGFβ inhibitor promotes anti-tumor immune responses

Abstract

Cancer curing immune responses against heterogeneous solid cancers require that a coordinated immune activation is initiated in the antigen avid but immunosuppressive tumor microenvironment (TME). The plastic TME, and the poor systemic tolerability of immune activating drugs are, however, fundamental barriers to generating curative anticancer immune responses. Here, we introduce the CarboCell technology to overcome these barriers by forming an intratumoral sustained drug release depot that provides high payloads of immune stimulatory drugs selectively within the TME. The CarboCell thereby induces a hot spot for immune cell training and polarization and further drives and maintains the tumor-draining lymph nodes in an anticancer and immune activated state. Mechanistically, this transforms cancerous tissues, consequently generating systemic anticancer immunoreactivity. CarboCell can be injected through standard thin-needle technologies and has inherent imaging contrast which secure accurate intratumoral positioning. In particular, here we report the therapeutic performance for a dual-drug CarboCell providing sustained release of a Toll-like receptor 7/8 agonist and a transforming growth factor-β inhibitor in preclinical tumor models in female mice.

Introduction

The anticancer immune response is generated in the antigen-rich cancerous tissue and in the associated lymph nodes¹. In most cancer patients, the anticancer immune response is, however, counteracted by an immunosuppressive tumor microenvironment (TME), resulting in a poorly organized and uncoordinated adaptive immune response^2,3. The immunosuppressive TME is a target that, if successfully transformed, can improve the outcome of cancer therapies^4,5,6. Immune activating compounds, which activate and link the innate and adaptive immune response, are among the most promising candidates to achieve this transformation^7,8. However, systemic administration of these agents is challenging and has a significant risk of dose-limiting toxicities at sub-therapeutic dose levels^6,8,9,10,11. As explored previously, direct intratumoral drug administration of these agents offers a tolerated and therapeutically effective alternative that can further generate systemic anticancer immune responses^{10,11,12,13,14}.

Tumors are central sources of antigens expressed across multiple cancer cell clones; and, intratumoral therapies thus potentially stimulate a polyclonal in situ vaccination directly in the antigen-rich cancer tissue¹. Therapeutic effects provided by free-drug injections are, however, often hampered by rapid loss of drug activity within the TME due to the lack of optimal formulation designs^1,15,16. This results in poor treatment outcomes as short-lived immune stimulation is insufficient to secure a coordinated immune response and especially when considering the reactive and plastic trafficking of intratumoral immune cells.

The TME is highly heterogeneous and plastic even for cancers of the same histology^17,18,19,20; and, several factors shape the TME and support cancer cell escape²¹. This has led to the consensus that TME polarizing therapies optimally should be multi-targeted to break the immunosuppressive barrier²¹. One example of such an approach is the combination of a Toll-like receptor 7/8 agonist (TLR7/8a) and a transforming growth factor-β inhibitor (TGFβi). TLR7/8a has been shown to link the innate and adaptive immune system^22,23,24; and TGFβi is known to redirect lymphoid and myeloid immune cells towards the anticancer phenotypes via pleiotropic effects^25,26. Co-delivery of TLR7/8a and TGFβi thus has the potential to generate effective anticancer immune responses and to overcome the immunosuppressive TME^{22,23,24,25,26}. To date, successful administration of TLR7/8a is limited²⁷, but sustained intratumoral delivery technologies based on e.g., hydrogels are being explored^28,29. Hydrogels, however, may suffer from rapid release of small molecules unless chemically linked to the hydrogel polymer, and sustained co-delivery of multiple agents with optimal dosing is inherently challenging. To overcome these obstacles, we designed the CarboCell platform to provide an immune cell hot spot directly in the cancerous tissue (Fig. 1a) and associated lymph nodes. Upon injection into the tumor, CarboCell transforms from a liquid into a semi-solid scaffold from which single or multiple small molecule drug(s) may be released - each with individually tailored release profiles. The therapeutic potential of controlled release has recently been demonstrated for CarboCell formulations releasing antibiotics in musculoskeletal infection models³⁰. Previously, the practicality of intratumoral injections has been subject for debate; and variation and operator dependent outcomes have been reported³¹. Some of the central challenges are associated with the rapid wash out of free drug from the TME and lack of methods for validating the positioning of injected drugs³¹. The CarboCell technology overcomes these limitations; it is further compatible with image-guided injection technologies and is effectively retained at the injection site, thereby alleviating inconsistencies related to rapid washout and insecure positioning. Clinical validation for injection across a range of malignancies and anatomical positions has already been demonstrated for imaging marker technologies with similar physical properties^32,33. In a previous publication, CarboCell was demonstrated to be intrinsically visible by ultrasound, magnetic resonance imaging, as well as by computed tomography (CT) upon co-formulation of iodine rich carbohydrate esters (Supplementary Fig. 1)³². This is attractive for image-guided injection and post-treatment follow-up. In comparison to alternative technologies that have been proposed for similar applications (e.g., those based on nanostructured materials or cross-linked polymer hydrogels^34,35) CarboCell is a liquid composition comprised of carbohydrate esters, triglycerides, biodegradable polymers, and ethanol as solvent (Fig. 1b). The carbohydrate esters and triglyceride excipients of CarboCell are biodegradable by hydrolysis and degradants are well-tolerated disaccharides and low molecular organic acids.

Fig. 1: CarboCell allows for flexible administration and provides sustained intratumoral release of drug combinations with high tumor retention and minimal systemic spillover.

a CarboCell is compatible with clinical injection technologies, including image-guided endoscopy, enabling treatment of most lesions (left). Upon contact with tissue or water it self-assembles into a depot (right). b The main constituents of CarboCell are the esterified carbohydrate sucrose octabenzoate (SuBen), the triglyceride glycerol trioctanoate (GTO), the TLR7/8a Resiquimod (R848) and the TGFβi RepSox. c R848 and RepSox release profiles from two CarboCell compositions (CC and CC-ER, Supplementary Table 1) were evaluated by s.c. injection of 50 μL CarboCell in mice [n = 3 per time point] and is presented as percent cumulative release of the total amount injected. Drug retention was quantified by HPLC of excised CarboCell. d Blood concentration [n = 4 per time point] and (e) intratumoral activity of R848 administered intratumorally (i.t.), intravenously (i.v.), or intratumorally via CarboCell (CarboCell TLR) was determined by liquid scintillation counting (LSC) using ³H-R848 [n = 4 per time point]. ³H-R848 was formulated in PBS or CarboCell (CC-4 (Supplementary Table 1)). The results are presented as mean ± SEM. Source data are provided as a Source Data file. Illustration created with BioRender.com, released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license.

In this work, we present data for CarboCells releasing TLR7/8a alone or in combination with TGFβi. We demonstrate a platform technology that can deliver sustained, high intratumoral drug doses with minimal off-target effects, and that drug release and dosing intervals can be controlled by modifying the CarboCell composition. The sustained intratumoral drug activity stimulates tumor directed immune activation, confirming the synergistic effect of TLR7/8 activation and TGFβ inhibition. The induction of a T cell inflamed TME and a coordinated adaptive anticancer response, including systemic anticancer activity, is furthermore demonstrated across preclinical in vivo cancer models.

Results

CarboCell provides sustained tumor drug activity and minimal systemic exposure

CarboCell is a liquid composition comprising mixtures of a carbohydrate ester, a triglyceride, a polymer additive, and ethanol as solvent (Fig. 1b). Prior to administration, the CarboCell liquid is 700-800-fold more viscous than water and is injectable using standard procedures (Supplementary Table 1, Fig. 1a). After administration, the semi-solid CarboCell depot is formed at the site of injection caused by efflux of the solvent. In this transition, the viscosity of CarboCell increases and it is immobilized in the tissue, from where high doses of a single or multiple drug(s) may be released. Several compositions were included in the optimization of the CarboCell TLR7/8a:TGFbi delivery system, which are abbreviated CC and sequentially described in Supplementary Table 1. CarboCell is intrinsically hydrophobic and can be utilized for delivery of a broad range of lipophilic drugs (logP > 0). Here, the TLR7/8a Resiquimod (R848, logP = 1.7, Fig. 1b) and the TGFβi RepSox (logP = 2.5, Fig. 1b) were co-formulated in CarboCell, as both compounds are lipophilic with comparable logP, rendering these drugs optimal for synchronized dual release. The drug release profiles are governed by the initial solvent efflux as well as the composition of the CarboCell. This was exemplified in vivo by subcutaneous injection (s.c.) of CC and CC extended release (CC-ER) CarboCell compositions (Fig. 1c). Here, the CC composition reached 92–95% drug release at day 7 in comparison to the CC-ER composition that only reached 51–57% drug release. A reduced release rate was obtained by lowering the triglyceride and polymer content (Supplementary Table 1). Importantly, such compositional change altered the release profile for both R848 and RepSox by the same magnitude, thereby preserving the synchronized release feature.

Intratumoral technologies are inherently dependent on the ability to support accurate and safe placement. The CarboCell imaging properties can assist the injection procedure and validate depot formation. To demonstrate the ability of CarboCell to form stable intratumoral depots, we incorporated the iodinated carbohydrate ester 6,6’-ditriidobenzene-isobuturic-sucrose (xSAIB) for radiographic contrast³². Micro-computed tomography (µCT) imaging confirmed that CarboCell formed well-defined depots after intratumoral injection, which were retained at the site of injection (Supplementary Fig. 1a–c). Next, we investigated in vivo pharmacokinetics and biodistribution (PK/BD) of R848 to validate the sustained intratumoral drug activity provided by the CarboCell. CarboCell formulated with the TLR7/8a R848 is referred to as CarboCell TLR. Tritium labelled R848 (³H-R848) was used for the PK/BD comparison of CarboCell TLR with intravenous and intratumoral injections of free ³H-R848. The blood activity curves were comparable for intravenous and intratumoral free ³H-R848, which illustrates the rapid wash out kinetics from tumors after intratumoral injections of free drug. On the contrary, we observed a very low blood activity of ³H-R848 in mice receiving the intratumoral CarboCell TLR (Fig. 1d). The total activity within the depot and tumor tissue is represented by the intratumoral CarboCell 3H-R848, as depicted in Fig. 1e. This, combined with the sustained release from CarboCell, supports the assertion that 3H-R848 offers continuous activity within the tumor tissue, accompanied by a low and decreasing systemic spill-over.

In comparison, less than 19% of the intratumorally injected free ³H-R848 could be detected in the tumor after 30 min, and less than 1% were found at all time points in intravenously injected mice (Fig. 1e). This observation is comparable to previous reports on intratumoral drug retention of small molecules and underlines the importance of a depot technology to achieve sustained stimulation^36,37,38. Mice injected with CarboCell TLR also displayed much lower ³H-R848 organ activity compared to mice injected with free intratumoral and intravenous ³H-R848 (Supplementary Fig. 2a–e). This demonstrates that the CarboCell technology effectively maintains drug activity in the TME, which is of central importance in reducing the risk of adverse effects associated with innate immune system activating compounds.

CarboCell TLR:TGFb demonstrates the therapeutic advantage of combinatorial immunotherapy

For initial demonstration of the therapeutic performance of CarboCell, we used a CarboCell composition that provides sustained release of the TLR7/8a R848 or a combination of the TLR7/8a R848 and the TGFβi RepSox, referred to as CarboCell TLR:TGFb, and compared these in the CT26 tumor model (Fig. 2a). For therapeutic evaluation, the CarboCell was given as four weekly intratumoral injections unless otherwise stated. The CarboCell TLR:TGFb-treated group displayed potent anticancer activity with 89% of the mice experiencing a complete response (i.e., tumor eradication with no adverse effects on viability). Conversely, only 22% complete responses were evinced in the CarboCell TLR-treated group (Fig. 2b, c), confirming the synergistic effect of TLR7/8 activation and TGFβ inhibition over single agent activity. All therapies were well tolerated, and all mice that demonstrated responses exhibited features of immunological memory upon tumor cell rechallenge (Supplementary Fig. 3a–c).

Fig. 2: Combinatorial CarboCell TLR:TGFb improves therapeutic efficacy and links the innate and adaptive immune response.

a Treatment schedule (CT26 tumors, CC (Supplementary Table 1)). b Individual tumor growth curves and c survival plots of mice bearing CT26 tumors [start size ~150 mm³, n = 9/group]. d–h Mice bearing established CT26 tumors were injected intratumorally with CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg, CC (Supplementary Table 1)). Untreated mice were included as controls. One day (1d), three days (3d) and seven days (7d) later, tumors and tdLNs were harvested for analysis. d Representative images of anti-MLKL immunohistochemistry (IHC) analysis of tumors. For remaining images see Supplementary Fig. 13. [bar = 200 µm, n = 3 (UT and d3), n = 4 (d1 and d7)]. e Gene expression analysis (one-way ANOVA with Tukey post-test) of tumors; Ccl4 (F(2,10) = 3.798, P = 0.0593), Cxcl10 (F(2,10) = 10.87, P = 0.0031), [n = 5 (UT), n = 5 (d1), n = 3 (d7)] and tdLNs; Ccl4 (F(2,12) = 15.10, P = 0.0005), Cxcl10 (F(2, 12) = 115.4, P < 0.0001) [n = 4 (UT), n = 5 (d1), n = 6 (d7)]. f Analysis of intratumoral levels (one-way ANOVA with Tukey post-test) of CXCL10 (F(3,11) = 3.661, P = 0.0475) and IFN-β (F(2,9) = 9.200, P = 0.0067) determined by ELISA [all groups n = 4, except CXCL10 d7 n = 3]. g, h Flow cytometry analysis (two-tailed unpaired t test). For gating strategies see Supplementary Fig. 14 and Supplementary Fig. 15. g %cDC1s, CD86 expression (MFI) on cDC1s, and %Inflammatory monocytes in tdLNs [n = 6]. h CD8⁺/Treg ratio and %AH1⁺ (of CD8⁺) T cells in tumors [n = 5]. i Treatment schedule. BALB/c [start size ~135 mm³, n = 8] and BALB/c-nu [start size ~ 121 mm³, n = 9] mice bearing established CT26 tumors received two weekly intratumoral injections with CarboCell TLR:TGFb. Untreated BALB/c [start size ~144 mm³, n = 8] and BALB/c-nu [start size ~133 mm³, n = 8] mice were included as controls. j Individual tumor growth curves. The results are presented as mean ± SEM. Source data and exact P-values are provided as a Source Data file.

Next, the mechanistic background for the improved therapeutic efficacy of the CarboCell TLR:TGFb treatment was investigated. The ability of CarboCell TLR:TGFb to increase the expression of Mixed Lineage Kinase Domain-Like (MLKL) protein in tumor tissue was first explored. MLKL mediates necroptosis, an alternative mode of programmed cell death, which has the potential to amplify the anticancer immune responses³⁹. Anti-MLKL staining of CT26 tumors after CarboCell TLR:TGFb injection resulted in increased expression of MLKL in tumor sections as compared to untreated controls (Fig. 2d), indicating that CarboCell TLR:TGFb induce necroptosis in the cancerous tissue. Additionally, gene expression analysis of CT26 tumors and tumor-draining lymph nodes (tdLNs) showed significantly elevated levels of the inflammatory chemokines Ccl4 and Cxcl10 one day after CarboCell TLR:TGFb treatment. This effect was gradually reduced at day seven (Fig. 2e). The expression of Cxcl9, Irf7, and Tnf-α was also significantly increased in the tdLNs one day after treatment with CarboCell TLR:TGFb while the levels of Cxcl9 and Irf7 were unchanged in tumors at both time points (Supplementary Fig. 3d). The increased gene expression of inflammatory cytokines and chemokines was mirrored at the protein level in CT26 tumor-bearing mice. Here, a tendency towards increased intratumoral levels of CXCL10 and IFN-β was observed one day following CarboCell TLR:TGFb treatment (Fig. 2f).

The higher levels of inflammatory chemokines and cytokines were associated with significantly increased levels of conventional type 1 dendritic cells (cDC1s) and inflammatory monocytes in tdLNs seven days after CarboCell TLR:TGFb treatment of CT26 tumor-bearing mice; further, an increased expression of the activation marker CD86 was observed on cDC1s (Fig. 2g and Supplementary Fig. 14). In the CT26 tumors, a potent improvement of the T cell composition was demonstrated by a significantly increased CD8⁺ T cell/T regulatory cell (Treg) ratio and an increased fraction of intratumoral cytotoxic T cells with specificity towards the CT26 immunodominant gp70 epitope (Dex-AH1⁺) (Fig. 2h and Supplementary Fig. 15). The observed gene expression patterns (Fig. 2e and Supplementary Fig. 3d) support this observation as CCL4, CXCL9, and CXCL10 produced by activated dendritic cells (DCs) are central for trafficking and localization of CD8⁺ T cells^40,41,42,43. Furthermore, TLR7/8-mediated stimulation is the main activator of IRF7⁴⁴ and leads to the production of TNF-α^45,46.

To investigate the therapeutic significance of CD8 + T cells in the context of CarboCell TLR:TGFβ treatment, we conducted a comparative efficacy study between immunocompetent BALB/c mice and immunodeficient BALB/c-nu mice, which are devoid of T cells. CT26 tumors were established in both mouse strains, followed by two intratumoral administrations of CarboCell TLR:TGFb (Fig. 2i). In immunodeficient BALB/c-nu mice treated with CarboCell, a delay in tumor progression was noted when compared to their untreated controls. Conversely, the immunocompetent BALB/c mice exhibited sustained tumor suppression over study period (Fig. 2j). These findings underscore the critical role of T cells in mediating the antitumor efficacy of CarboCell TLR:TGFb intratumoral therapy.

Tailoring the CarboCell drug release kinetics can increase injection intervals, thereby reducing the total number of injections. Therefore, the influence of release rate on therapeutic efficacy was investigated by comparing the CarboCell TLR:TGFb standard formulation with a CarboCell TLR:TGFb extended drug release composition with double drug concentration in CT26 tumor-bearing mice (Supplementary Fig. 3e). For both formulations, a single intratumoral injection of CarboCell TLR:TGFb resulted in complete responses in the majority (50-75%) of the treated animals (Supplementary Fig. 3f, g). Thus, increasing the dose and reducing the burst effect can achieve comparable treatment outcomes, potentially allowing for fewer and less frequent doses, enhancing clinical relevance.

CarboCell TLR:TGFb shows therapeutic efficacy and immune activation across models

The observed therapeutic efficacy in the CT26 tumor model, a model known to be highly infiltrated by cytotoxic immune cells⁴⁷, encouraged further exploration of CarboCell performance across different immune infiltrated TME backgrounds. In the MC38 tumor model, a model dominated by immunosuppressive cell types⁴⁷, we compared the therapeutic performance of four weekly intratumoral injections of empty CarboCell or CarboCell TLR:TGFb with two intratumoral injections (with 14-day intervals) of the CarboCell TLR:TGFb composition that provides extended release (Fig. 3a). Both compositions of CarboCell TLR:TGFb effectively decreased tumor growth and improved animal survival (Fig. 3b, c) independent of their different drug release rates; all CarboCell compositions were well tolerated (Supplementary Fig. 4a, b).

Fig. 3: Combinatorial CarboCell TLR:TGFb provides single agent therapeutic activity in MC38 and EMT-6 tumor bearing mice.

a Treatment schedule (MC38 tumors, CC (top) and CC-ER (bottom) (Supplementary Table 1)). b Individual tumor growth curves and (c) survival plots of mice bearing MC38 tumors [start size ~109 mm³, n = 9 (UT), n = 9 (Empty CarboCell), n = 8 (CarboCell TLR:TGFb), n = 8 (CarboCell TLR:TGFb extended release)]. d–f Mice bearing established MC38 tumors were injected intratumorally with CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg, (CC (Supplementary Table 1)). Untreated mice were included as controls. One day (1d) and seven days (7d) later, tumors and tdLNs were harvested for analysis. d Gene expression analysis (one-way ANOVA with Tukey post-test) of tumors; Ccl4 (F(2,12) = 10.87, P = 0.0020) and Cxcl9 (F(2,12) = 19.44, P = 0.0002) [n = 5 (UT), n = 5 (d1), n = 4 (d7)] and tdLNs; Ccl4 (F(2,12) = 10.87, P = 0.0020), Cxcl9 (F(2,12) = 19.44, P = 0.0002) and Irf7 (F(2,12) = 7.059, P = 0.0094) [all groups n = 5]. e Analysis of intratumoral levels (one-way ANOVA with Tukey post-test) of IFN-β determined by ELISA assay [n = 4]. f Representative images of anti-CD8 immunohistochemistry (IHC) analysis of tumors [bar=100 µm, n = 3]. g Treatment schedule EMT-6 tumors (CC (Supplementary Table 1)). h Individual tumor growth curves and (i) survival plots of mice bearing EMT-6 tumors [start size ~95 mm³, n = 8 (UT), n = 6 (CarboCell TLR), n = 5 (CarboCell TLR:TGFb]. j, k Flow cytometry analysis (two-tailed unpaired t test). For gating strategies see Supplementary Fig. 16 and Supplementary Fig. 17. Mice bearing established EMT-6 tumors were injected intratumorally with CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg, CC (Supplementary Table 1)) and untreated mice were included as controls [n = 6/group]. Analysis was performed seven days after treatment. j %cDC1s (of CD45⁺), CD86 expression (MFI) on cDC1s, and %inflammatory monocytes (of CD45⁺) in tdLNs. k %TAMs (of CD45⁺) in tumors. The results are presented as mean ± SEM. Source data and exact P values are provided as a Source Data file.

To decipher the mechanisms for the observed therapeutic effect, we performed gene expression analysis in MC38 tumor-bearing mice. In these tumors, Ccl4 was significantly increased, whereas Cxcl9, Irf7, and Cxcl10 were not significantly elevated following treatment (Fig. 3d and Supplementary Fig. 4c). In the tdLNs, a more pronounced effect was observed as Ccl4, Cxcl9, and Irf7 were all significantly elevated one day after treatment (Fig. 3d). Cxcl10 levels were also initially elevated whereas Tnf-α was unchanged at both evaluated time points (Supplementary Fig. 4c). Additionally, we evaluated the cytokine production in MC38 tumor-bearing mice following treatment with CarboCell TLR:TGFb. IFN-β production was slightly elevated one day after treatment but similarly decreased at day 7 (Fig. 3e). CXCL10 production was unchanged over time (Supplementary Fig. 4d). The elevated levels of inflammatory chemokines and cytokines were mirrored in increased infiltration of CD8⁺ cells in the MC38 tumors seven days after treatment with CarboCell TLR:TGFb (Fig. 3f).

Next, we compared the therapeutic efficacy of CarboCell TLR with CarboCell TLR:TGFb in the EMT-6 tumor model (Fig. 3g). In this model, CarboCell TLR:TGFb again showed potent anticancer activity with 80% of the mice demonstrating complete responses to treatment as compared to 50% and 0% in the CarboCell TLR and untreated groups, respectively (Fig. 3h, i). All treatments were well tolerated, and long-lasting immunological anticancer memory imparted by CarboCell TLR:TGFb was confirmed by cancer cell rechallenge in the opposite flank (Supplementary Fig. 4e–g). As in the CT26 tumor model (Fig. 2g, h), CarboCell TLR:TGFb significantly increased the infiltration of cDC1s and inflammatory monocytes in tdLNs as well as the expression of CD86 on cDC1s, and significantly decreased infiltration of tumor-associated macrophages (TAMs) in the tumors seven days after treatment (Fig. 3j, k and Supplementary Figs. 16, 17).

Radiation therapy (RT) is a central part of the therapeutic regimen for more than 50% of patients with solid cancer and is a potent immunogenic cell death inducer. RT alone, however, is rarely capable of generating curative immune responses⁴⁸. Accordingly, we evaluated the synergistic activity of CarboCell and RT in MC38 (Supplementary Fig. 5a), B16.F10 (Supplementary Fig. 5d), and CT26 (Supplementary Fig. 5g) tumor-bearing mice. For all radio-immunotherapeutic studies, we used CarboCells formulated with 2.5 times lower TLR7/8a dose relative to the CarboCell monotherapy dose. This was done to reduce CarboCell-mediated necroptosis and minimize complete anticancer responses in the CarboCell only treatment groups. This should allow the radiation therapy to mediate the release of cancer antigens and demonstrate potential synergy between the two treatment modalities⁴⁹. Across all tumor models, we observed therapeutic efficacy and synergistic activity (Supplementary Fig. 5b, c, Supplementary Fig. 5e, f, Supplementary Fig. 5h, i).

CarboCell induces systemic anticancer activity

To investigate the systemic anticancer properties of CarboCell TLR:TGFb, we used mice bearing two established CT26 tumors (one on each flank), and injected CarboCell TLR:TGFb in only one tumor (Fig. 4a). This generated an effective systemic anticancer response as the treatment reduced tumor growth in both the injected and uninjected tumors and led to 60% complete responders (Fig. 4b, c). The individual tumor growth curves support that systemic anticancer immune reactivity is generated over time towards the growing tumors leading to elimination of these by the systemic adaptive response. Interestingly, some tumors grew to very large sizes ( ~ 500 mm³) before a therapeutic effect was observed, which underlines the potency of the technology. Systemic anticancer reactivity was also demonstrated in large A20 tumors using a bilateral tumor model (Supplementary Fig. 6a–c).

Fig. 4: Intratumoral CarboCell TLR:TGFb induces systemic anticancer immune activity and immunological memory.

a Treatment schedule (CT26 tumors, CC Supplementary Table 1). Mice bearing two established CT26 tumors were injected intratumorally in one tumor with CarboCell. b Individual tumor growth curves and (c) survival plots of mice bearing two CT26 tumors [start size ~100 mm³ (injected tumor), start size ~87 mm³ (uninjected tumor), n = 9 (UT) and n = 15 (CarboCell TLR:TGFb)]. d Flow cytometry analysis (one-way ANOVA with Tukey post-test) of %CD36 (of Tregs) (F(2,19) = 20.99, P < 0.0001) and %Tregs (of CD45 + ) (F(2,19) = 0.6498, P = 0.5334) in tumors 3 days [n = 6 (UT)] or 14 days [n = 5 (CarboCell TLR:TGFb)] after injection of CarboCell. Mice bearing two established CT26 tumors were injected intratumorally in one tumor with CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg, CC (Supplementary Table 1)). Untreated mice were included as controls. For gating strategy see Supplementary Fig. 18. e TCRβ CDR3 nucleotide sequence overlap (two-tailed unpaired t-test) between intratumoral TCRβ CarboCell injected tumors and uninjected tumors from mouse bilateral tumor model [n = 6 (UT), n = 4 (CarboCell TLR:TGFb)]. f Morisita overlap index of intratumor TCRβ in CarboCell injected (R) and contralateral uninjected tumor (L) from CarboCell CarboCell TLR:TGFb treated mice [n = 4 (M01-M04)]. g, h Lung metastases were evaluated in mice bearing s.c. 4T1 flank tumors after three weekly intratumoral injections of CarboCell TLR (7.5 mg/kg), CarboCell TGFb (20 mg/kg), or CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg) starting at day 7 post inoculation (CC (Supplementary Table 1)), or no treatment (UT) [n = 8 (CarboCell groups), n = 7 (UT)]. 24 days post inoculation, lung metastases were quantified by 6-thioguanine clonogenic assay (Supplementary Fig. 19). g Percent of mice displaying complete clearance of pulmonary metastases. h Analysis of lung colonies (one-way ANOVA with Tukey post-test) (F(3,27) = 5.765, P = 0.0035) i Flow cytometry evaluation of tumor arginase-1 expression (MFI) in 4T1 tumors injected with CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg, CC (Supplementary Table 1)) analyzed (one-way ANOVA with Tukey post-test) one (1d), three (3d), and seven days (7d) after treatment [n = 6 (UT), n = 5 (d1) n = 5 (d3) n = 6 (d7)] on Mo-MDSCs (F(3,18) = 8.852, P = 0.0008), PMos (F(3,18) = 22.24, P < 0.0001), and TAMs (F(3,18) = 7.347, P = 0.0020). For gating strategy see Supplementary Fig. 20. j Treatment schedule (CT26 tumors, splenocyte transfer). Mice bearing established CT26 tumors were treated with intraperitoneal injection (i.p.) of cyclophosphamide (CTX) one day prior to splenocyte transfer. The mice received either no splenocytes, splenocytes from naïve mice, splenocytes from mice bearing established CT26 tumors, or splenocytes from mice rejecting two CT26 tumors (data shown in a–c). k Mean tumor growth curves and (l) survival plots of mice bearing CT26 tumors [palpable tumors, n = 8/group]. The results are presented as mean ± SEM. Source data and exact P values are provided as a Source Data file. Illustrations created with BioRender.com, released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license.

The ability to generate optimally polarized effector T cells is central for the CarboCell induced immune hotspot. However, Tregs in the TME are known to suppress effector T cell function and it has been shown that CD36 expression is involved in this process⁵⁰. We, therefore, investigated CD36 expression on tumor infiltrating Tregs from mice bearing two established CT26 tumors (one on each flank). Again, the mice were only treated with CarboCell TLR:TGFb in one tumor. The percentage of Tregs was unchanged following treatment, however, the percentage of CD36⁺ Tregs was significantly decreased in both the injected and the uninjected tumors (Fig. 4d and Supplementary Fig. 18). The decreased CD36 expression was equally pronounced in the injected and uninjected tumors, further indicating that CarboCell TLR:TGFb is able to generate a systemic response.

To explore anti-cancer T-cell responses in CarboCell TLR:TGFb-treated and untreated tumors, TCRβ CDR3 immune sequencing was conducted on whole tumor tissue from mice with bilateral CT26 tumors. Encouragingly, despite no increase in overall TCR clone count (Supplementary Fig 8) the treated tumor and contralateral tumors showed sequence overlap of 731, compared to only 41 sequences for controls (Fig. 4e). Notably, while individual CarboCell TLR:TGFb-treated mice displayed high overlap, minimal overlap was observed between mice as illustrated by the Morisita index map (Fig. 4f, Supplementary Fig. 9).

The murine 4T1 breast cancer model can spontaneously metastasize to other sites of the body including to the lungs⁵¹. Considering that TGFβ is a central mediator of the formation of the immunosuppressive TME and intricately linked to epithelial–mesenchymal cancer cell transition and metastasis⁵², we used this model to investigate if CarboCell TLR:TGFb was able to reduce lung metastases. This was investigated using a 6-thioguanine-clonogenic assay performed on lungs from mice bearing 4T1 flank tumors. CarboCell formulated with the TGFβi RepSox is referred to as CarboCell TGFb. In lungs harvested 24 days after cancer cell inoculation, we identified 6-thioguanine-resistant 4T1 cells across all treatment groups; however, six out of eight mice treated with CarboCell TLR:TGFb did not show any signs of metastases compared to only one out of eight mice in the CarboCell TLR and CarboCell TGFb treated groups (Fig. 4g, h). Using ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸FDG-PET)/ computed tomography (CT) it was furthermore shown that CarboCell TLR:TGFb and RT significantly lowered the ¹⁸FDG positive lung volume compared to RT alone in the 4T1 model (Supplementary Fig. 7a, b). To confirm that sustained and effective intratumoral inhibition of TGFβ was achieved, we investigated intratumoral arginase-1 (Arg-1) expression in myeloid populations in 4T1 tumors. In the monocytic myeloid-derived suppressor cells (Mo-MDSC), patrolling monocytes (PMo), and TAM populations we observed that Arg-1 expression (MFI) levels were significantly reduced three and seven days after intratumoral CarboCell TLR:TGFb treatment (Fig. 4i and Supplementary Fig. 20).

It has been suggested that the formation of long-lasting immunological anticancer memory often develops independently of treatment for murine syngeneic tumor models, questioning the utility of rechallenge experiments⁵³. To confirm that immunological memory was formed in response to CarboCell treatment (Supplementary Fig. 3c, Supplementary Fig. 4g), we transferred splenocytes, recovered more than 100 days after initial inoculation, from the mice that had previously rejected both the injected and uninjected CT26 tumors in response to CarboCell TLR:TGFb (Fig. 4a–c). Here, we preconditioned mice bearing established CT26 tumors with cyclophosphamide one day prior to splenocyte transfer from naïve mice, CT26 tumor-bearing mice, or from the mice rejecting two CT26 tumors (Fig. 4j). Splenocyte transfer from mice rejecting two CT26 tumors following treatment with CarboCell TLR:TGFb led to 100% complete responders whereas mice receiving splenocytes from naïve or CT26 tumor-bearing mice resulted in 12.5% and 25% complete responders, respectively (Fig. 4k, l). This demonstrates that CarboCell TLR:TGFb generates a sustained central immunological anticancer memory response capable of eliminating established tumors when transferred into tumor-bearing recipients.

CarboCell improves the therapeutic performance of checkpoint inhibition

The observed innate and adaptive immune activation induced by CarboCell TLR:TGFb is highly attractive for improving the therapeutic activity of immune checkpoint inhibitors (ICIs). Therefore, we investigated the influence of CarboCell treatment on ICI extravasation and its effects on tumor vasculature. For this, we employed optical tissue clearing allowing for 3D imaging of large tumor volumes using confocal microscopy. The data was processed using machine learning-based image analysis as previously reported⁵⁴. Tumor-bearing mice were treated with CarboCell and five days later the mice received AF555-labelled lectin to visualize the tumor vasculature and AF647-labelled programmed cell death ligand 1 antibody (aPD-L1) to evaluate ICI extravasation. Our analysis demonstrated that CarboCell TLR and CarboCell TLR:TGFb stimulated a dramatic reduction in viable tumor tissue volume (Fig. 5a–c), an increased targeting of the remaining viable tissue by the aPD-L1 (Fig. 5d), and its more extensive distribution outside of the viable tissue (Fig. 5e).

Fig. 5: Intratumoral CarboCell TLR:TGFb decreases viable tumor tissue volume and augments conventional aPD-L1 immunotherapy.

a–f Mice bearing established CT26 tumors were injected intratumorally with empty CarboCell, CarboCell TLR (7.5 mg/kg), CarboCell TGFb (20 mg/kg), or CarboCell TLR:TGFb (7.5 mg/kg:20 mg/kg) (CC (Supplementary Table 1)). Untreated mice were included as reference. Five days after treatment, tumors were harvested for analysis. a 3D rendering of the representative preprocessed images of AF555 lectin-labelled tumor vasculature (red) and AF647-labelled anti-PD-L1 (turquoise) extravasation in optically cleared tumors. b 3D isosurface rendering of representative masks of total tumor tissue (semi-transparent grey) and viable tumor tissue (cyan). Scale bar, 1 mm. c–e Results of image analysis describing effects of the treatments (one-way ANOVA with Tukey post-test) [n = 4/group] on (c). the volume of viable tumor tissue (F(3, 12) = 4.769, P = 0.0206), d targeting of viable extravascular space (EVS) (F(3,12) = 2.658, P = 0.0958), e fraction of anti-PD-L1 extravasation volume ending up in viable EVS from total extravasation volume (F(3,12) = 11.07, P = 0.0009), and (f). total extravasation volume per total vessel surface (vessel permeability) (Kruskal-Wallis test (F) P = 0.0248, Dunn’s post-test). g Treatment schedule (MC38 tumors, CC-ER (Supplementary Table 1)). h Individual tumor growth curves and (i) survival plots of mice bearing MC38 tumors [start size ~129 mm³, n = 8 (UT), n = 7 (aPD1), n = 8 (CarboCell TLR:TGFb), n = 6 (CarboCell TLR:TGFb + aPD1)]. The results are presented as mean ± SEM. Source data and exact P values are provided as a Source Data file.

Analysis of vessel permeability showed a dramatic increase in the CarboCell TLR:TGFb group (Fig. 5f), both with regards to the amount of aPD-L1 leaving the vasculature, as well as in number of extravasation foci, which can be a surrogate measure for a number of breaches in vasculature (Supplementary Fig. 10a). Comparing permeable vasculature between the groups yielded similar differences (Supplementary Fig. 10b, c), suggesting that the treatments increase the sizes of existing breaches and the number of breaches in the already permeable vessels. Interestingly, at the evaluated time point, the treatments did not lead to changes in the tumor vessel architecture (Supplementary Fig. 8d). While the increase in vessel permeability seemed to be the main driver of the antibody extravasation, tumor decompression could also contribute to the improved antibody distribution. This is suggested by a combination of a more extensive distribution of the antibody beyond the primary source of extravasation (Fig. 5a, e) and a slight but significant decrease in the fluorescent intensity within extravasation foci (Supplementary Fig. 11a, b), indicating a decrease in local antibody concentration due to facilitated diffusion within tumor tissue.

Next, we hypothesized that CarboCell could augment the therapeutic response to ICIs. One intratumoral injection of extended release CarboCell TLR:TGFb was therefore combined with a systemically administered aPD-L1 in the MC38 tumor model (Fig. 5g). aPD-L1 monotherapy yielded no complete responders; however, a single injection of CarboCell TLR:TGFb combined with aPD-L1 did result in 50% complete responders compared to 25% for CarboCell TLR:TGFb monotherapy (Fig. 5h, i). The treatment was well tolerated, which supports the potential of CarboCell as part of a multi-targeted therapeutic approach (Supplementary Fig. 12a, b).

The combination of liposomal oxaliplatin (lip OxPt) and systemic R848 was previously demonstrated as a potent chemo-immunotherapeutic combination⁵⁵. We, therefore, decided to test the CarboCell system in combination with lip OxPt with or without the addition of a systemically administered programmed cell death 1 (PD-1) antibody. As in the radio-immunotherapy studies, we used CarboCells formulated with 2.5 times lower TLR7/8a dose. The combined CarboCell TLR and lip OxPt treated group displayed potent anticancer activity with 67% complete responders in the CT26 model (Supplementary Fig. 13a–c). Encouragingly, addition of PD-1 antibody to the combination of CarboCell TLR and lip OxPt in the CT26 tumor model yielded a 100% complete response rate (Supplementary Fig. 13d–f).

CarboCell is tolerated in experimental canine model

Lastly, to validate that the release kinetics and tolerability observed in mice could be translated into larger animals, we investigated the effects of s.c. injection of large volume CarboCell TLR:TGFb depots (Supplementary Fig. 14a) in laboratory beagle dogs. CarboCell depots were surgically collected at day 7 or day 14 as well-confined viscous depots and analysis validated that both TLR7/8a and TGFβi were continuously released beyond the 14 days period (Supplementary Fig. 14b). None of the dogs displayed adverse reactions or failures to thrive throughout the study period (Supplementary Fig. 14c–h). The tolerability of CarboCell TLR:TGFb was therefore found to be highly promising for further advancement of the CarboCell system.

Discussion

Intratumoral immunotherapy has been subject for debate. Irrespective of this, solid evidence supports that the anticancer immune response is generated in the tumor, peritumoral tissue, and tumor associated lymphatic tissue¹. This makes the tumor a strategic target for cancer immunotherapy, which is underlined by the high number of technologies and strategies that have been pursued to achieve tumor-directed immune activation^{1,10,11,12,14,56}. Notwithstanding these efforts, no technology has yet successfully progressed into a clinical product. Studies on systemic tumor-targeting technologies have demonstrated poor tolerability. Recent clinical data for HER2 targeting TLR conjugates reported a narrower-than-anticipated therapeutic window and high-grade adverse effects during dose escalation in addition to disappointing therapeutic activity²⁷.

The CarboCell system was, therefore, developed as a flexible, tolerated, and broadly applicable platform technology for sustained delivery of a single drug or drug combinations directly in the antigen-rich tumor. The CarboCell technology achieves three key therapeutic goals: (i) uses the tumor as a source of cancer associated antigens, (ii) continuously activates the TME, and (iii) links the innate and adaptive immune response to maintain a proinflammatory milieu in the tumor. The plasticity and reactivity of intratumoral immune cells and the time required to establish an adaptive immune response makes continuous activation central to avoiding rapid reversal of the TME to an immunosuppressive status. This may be essential for establishing a systemic polyclonal antitumor response and could explain the clinical failure of a free drug intratumoral TLR9 technology, in which the primary endpoint of objective response rate was not met⁵⁷.

Multi-targeted immunotherapy improves the therapeutic response rates, but it has also been associated with increased adverse reactions^58,59. Remarkably, we demonstrated improved therapeutic efficacy for our combinatorial CarboCell TLR:TGFb without compromising tolerability. Increased activation of both the innate and adaptive immune system upon CarboCell injection was observed across a range of heterogenous TMEs in all evaluated models. Importantly, we observed a substantial benefit of intratumoral CarboCell TLR:TGFb in the tdLNs, on the systemic anticancer response, and on the pulmonary metastasis formation. This observation makes CarboCell highly attractive as monotherapy as well as in combination regimens with neoadjuvant radiotherapy.

The observed systemic anticancer response after injection with CarboCell indicates that well-coordinated adaptive immune responses have been established.

The TCR sequence overlap between the injected and uninjected tumors in the same mouse is an important observation. This suggests a systemic immune response induced by the local immunotherapy. The high degree of overlap implies that T cells in the treated tumor are also recognizing and responding to the untreated tumor following circulation of activated T cells.

This ability to link the innate and adaptive immune response and improve T cell activation is highly attractive for combination with immune checkpoint inhibitors, as they are inherently dependent on an adaptive response^60,61,62. The value of CarboCell-mediated immune activation was shown by the observed therapeutic potentiation of both aPD-L1 and aPD-1 therapy.

The core of the CarboCell technology is comparable to a multimodal fiducial marker system for which we previously demonstrated injectability and depot-forming characteristics in both translational models and cancer patients^32,33,63. Contrary to oil-based depots, the current carbohydrate ester-based technology displays an approximate 1000-fold increase in viscosity upon injection, which provides depot stability and prevents fragmentation³². In comparison to alternative technologies, CarboCell is very adaptable in terms of drug incorporation and release rates. These parameters can be adjusted, even for multi-drug combinations, by simple ratio variations of the CarboCell base components. Achieving controlled, sustained release of small molecular drugs has been reported as highly challenging for competing technologies³⁴. TLR7/8 loaded hydrogels are being advanced both clinically and preclinically for intratumoral injection^28,29. Hydrogels, however, display rapid release of small molecules or necessitate polymer-drug conjugates with increased chemical complexity, resulting in higher production cost. Whereas the integrity of hydrogels is governed by polymer-polymer and polymer-water interactions, CarboCell is non-polymeric and self-coalescent due to its hydrophobic nature. Hydrogels in addition contain water, posing the risk of premature drug hydrolysis and imposing constraints on shelf-life. CarboCell is a non-aqueous liquid and has a completely scalable production setup which underlines the value of the current system. All steps in the CarboCell manufacturing process can further be automated; and, all base components, including several small molecular drugs, are available in GMP quality. Long-term stability has further been demonstrated for CarboCell formulations and stability following product sterilization by steam has been established (Supplementary Fig. 23, Supplementary Tables 8–11). These features are major advantages when compared to existing technologies that require advanced formulation, fabrication, and synthesis (e.g., nanostructures or drug conjugates with challenging scalability^34,64).

The therapeutic value of the CarboCell technology remains to be explored in human cancer patients where the clinical potential may be limited to cancers expressing antigens applicable for presentation and recognition by the immune system. We do, however, believe that the single-agent therapeutic performance of combinatorial CarboCell as well as the synergistic effect of CarboCell with clinically approved therapies underlines the translational value of the technology. In this regard, the easy adaptation of imaging contrast in the CarboCell system provides not only image guidance during the injection but also enables clinicians with the possibility to validate the CarboCell location for safety and performance evaluation during clinical trials. The ability to delineate CarboCell on combined anatomical and functional imaging features further allows clinicians to plan injections accordingly (e.g., to evaluate PET based metabolic activity^31,65). Injectability through thin needle technologies is a further attractive feature as clinical data supports that decreasing needle size reduces the risk of morbidity associated with intratumoral procedures⁶⁶.

Drug release from CarboCell injections in mice was demonstrated to last beyond one week and more than two weeks for the CC-ER composition. Extending the period of release lowers the number of required injections, which is important to reduce the risks of cancer cell seeding and dissemination from needle punctures^67,68. Higher compliance of patients and medical personnel is an expected benefit of fewer administrations. Prolonged release and reduced number of injection procedures are furthermore a prerequisite if more advanced injection technologies and clinical procedures are to be feasible with respect to their complexity and costs.

In summary, the flexible CarboCell system provides a solution to achieve sustained intratumoral release of multiple small molecular drugs in therapeutically relevant doses with minimal systemic spillover.

Methods

Study design

The study objective was to demonstrate the anticancer activity of the intratumoral CarboCell technology. This was investigated for a TLR7/8a with or without the addition of a TGFβi. Release kinetics and biodistribution was determined using direct truly quantitative methods. Efficacy studies were conducted in syngeneic murine cancer models alone or in combination with chemotherapy, immune checkpoint inhibitors, or RT. Tolerability was evaluated in mouse models and laboratory beagle dogs. Mechanistic evaluation of the therapeutic response was conducted using flow cytometry, ELISA, qPCR, and IHC. With one exception, the mice in all efficacy studies were randomized into treatment groups based on tumor size. The mice in the study with splenocyte transfer were randomized according to the presence of palpable tumors. Power analysis was not performed to predetermine sample size, but sample size was chosen based on previously published studies. Researchers were not blinded during the studies. The studies, study design, results and finding are not specific towards any sex or gender. Sample size for each experiment and statistical methods are described in figure legends. All studies and procedures involving animals were performed under license from the Danish National Animal Inspectorate (mice) or Hungarian Animal Research license (Beagles) and all study procedures were approved by the Institutional Ethical Administrative Committee at the Technical University of Denmark or Copenhagen University. The maximum tumor burden in studies involving mice was according to license 2 g, or 10% of the total bodyweight.

Preparation of CarboCell CC and CC-ER

CarboCell was formulated with the TLR7/8a Resiquimod, R848, (Ark Pharm or AmBeed) and the TGFβi RepSox (Selleckchem) unless stated otherwise. CarboCell was prepared by weighing each component (sucrose octabenzoate (SuBen), glycerol trioctanoate (GTO), ethanol (EtOH), polylactic acid (PLA, Mw 10000-18000Da)) in the amount corresponding to the weight percentage (or weight ratio) of each composition (Supplementary Table 1). Each mixture was placed in an ultrasonication bath at 70–80 °C for 1.5–2 h and occasionally vortexed until obtaining a clear, homogeneous solution. R848 and/or RepSox were incorporated in the composition by proportionally adding CarboCell on top of pre-aliquoted drug. CarboCell was subjected to magnetic stirring at 40–50 °C until the drugs were completely dissolved. CarboCell was stored in sealed glass vials at 4 °C until use. SuBen, GTO, and PLA was obtained from Sigma Aldrich and ethanol from VWR Chemicals.

CarboCell TLR:TGFb in vivo release kinetics

Female BALB/c mice were injected s.c. with 50 μL of CarboCell TLR:TGFb compositions. R848 and RepSox content was measured 24 h, 3 days, 7 days, and 10 days after injection. Depots were immediately dissolved in 1 mL acetonitrile and incubated at 25 °C overnight. The samples were filtered using 0.45 μm pore nylon syringe filters and diluted five times. As reference material, 10 µL of each CarboCell TLR:TGFb composition were dissolved in 1 mL acetonitrile. Samples were analyzed on a Shimadzu Nexera-X UHPLC. Samples were injected (5 μL) onto a Waters Terra XBridge BEH C8 column (2.5 μm, 4.6 × 75 mm) at 25 °C, 0.8 mL/min. The solvent system consisted of mobile phase A (5% acetonitrile, 0.1% TFA in water) and mobile phase B (0.1% TFA in acetonitrile). The gradient was 0% B for 1 min, 0 to 100% B in 5 min, 100% B for 2 min, 100% B to 0% B in 0.5 min, 0% B for 1.5 min. UV detection was used to measure the content of R848 (320 nm), RepSox (320 nm), and SuBen (280 nm). The release kinetics of R848 and RepSox was determined by ratiometric analysis. The AUC values were measured to calculate ratios of R848/SuBen and RepSox/SuBen in all samples. Released drug was calculated by comparing AUC ratio in collected CarboCell with the corresponding ratios in the reference material.

Formulation of CarboCell and free drug solutions containing tritium-labelled TLR7/8a R848

CarboCell with ³H-labelled TLR7/8a was prepared by pipetting the [³H]-R848 stock (Hartmann Analytic, 330 µL, 12 MBq) in a glass vial and removing the EtOH under a stream of argon for 10 min. CarboCell TLR (1.2 mg/mL R848, CC-4, 1.71 g, 1.62 mL) was then added to the dry residue and the mixture was magnetically stirred at 30 °C for 30 min. After complete stirring, the entire mixture was transferred to a glass vial, resulting in a transfer of 1.68 g (1.59 mL). The homogeneity and radioactivity concentration of the solution was tested in two samples of 36.2 mg (34.5 µL) and 28.2 mg (26.9 µL), removed from the vial. These samples were diluted with EtOH to 5 mL and from each diluted solution, 200 µL was added to a scintillation vial (Perkin Elmer) containing 10 mL Ultima Gold (Perkin Elmer), before analysis by liquid scintillation.

The free drug solutions for i.t. and i.v. administration were prepared by drying down the [³H]-R848 stock (459 µL, 18 MBq) as above. A solution of R848 in PBS (1.2 mg/mL, 2.40 mL) was then added to the dry residue. The mixture was stirred for 1 h at 40 °C. From the resulting mixture, 1.22 mL was collected for i.t. injections (1.2 mg/mL [³H]-R848 in PBS). From the same mixture, 1.10 mL was collected and mixed with 1.10 mL PBS, to generate the i.v. injection sample (0.6 mg/mL [³H]-R848 in PBS). From each of the PBS solutions, 2 × 10 µL was removed to test for homogeneity and radioactivity concentration. These were diluted to 1 mL with EtOH and 200 µL was added to a scintillation vial containing 10 mL Ultima Gold.

Pharmacokinetics and biodistribution

Mice were injected with [³H]-R848 administered i.v. in PBS (3.75 MBq/mL [³H]-R848, 3 mg/kg R848), intratumoral in PBS (7.5 MBq/mL [³H]-R848, 3 mg/kg R848) or intratumoral with CarboCell composition as previously described (7.4 MBq/mL [³H]-R848, 3 mg/kg R848, CC-4, Supplementary Table 1). The stated activity concentrations are nominal but correspond to measured activity values. Collected blood and tissue samples were dissolved in Soluene 350 (Perkin Elmer) and/or EtOH (1-6 mL) and heated overnight (60 °C). Hydrogen peroxide (30%, 0.2 mL) was added followed by heating (30 min. at 60 °C). 200 µL sample solution was transferred to liquid scintillation containing 10 mL Ultima Gold (Perkin Elmer). Specific tritium content was determined by liquid scintillation counting.

Liquid scintillation [³H]-R848

Tritium content was measured by liquid scintillation counting on a Hidex 300 SL, using the double coincidence detection tritium setting (Hidex, Finland). Samples in scintillation liquid were allowed to settle for 24 h and counted until a minimum of 1600 counts or for a maximum of 10 min. After initial counting, aliquots of tritiated water with known activity levels (270000 CPM/aliquot) were spiked into each sample and the samples recounted to quantify the sample matrix quench. The counting efficiency (CE) for each sample was determined as CE = [A(CPM of spiked sample) - A(CPM of sample)] / A(CPM of spike in PBS). Subsequently, the matrix quench corrected sample activity (A_QC) was obtained as A_QC = A(CPM of sample) / CE, where CE is the sample specific counting efficiency. The results were reported as percent injected dose per gram (%ID/g) or percent injected dose per organ/tumor (%ID/tumor).

In vivo CarboCell volume monitoring by CT

BALB/c mice carrying s.c. CT26 tumors (90 mm³) were treated with three doses of oxaliplatin loaded liposomes (8 mg/kg) followed by four intratumoral injections of xSAIB containing CarboCell TLR. Under general anesthesia ( ~ 4% sevoflurane) mice were CT scanned (nanoScan Mediso Medical Imaging Systems) 24 h after each CarboCell administration. Image analysis was performed using Inveon software (Siemens Medical Systems). A manual ROI covering the tumor volume were constructed and volume of the CarboCell determined by segmentation using a threshold of > 400 Hounsfield units.

Mice and cell lines

Female mice were obtained at 5–7 weeks of age (BALB/c, BALB/c-nu (BALB/cAnN-Foxn1nu/nu/Rj) or C57BL/6 from Janvier and Charles River). Mice housed in IVC2 or IVC3 cages with bedding material, and environmental enrichment provided running a 12-hour light cycle and ad lib standard chow and water. CT26, EMT-6, A20, and 4T1. CT26, EMT-6, A20, and 4T1 cells were obtained from the American Type Culture Collection (ATCC) with the following catalog numbers: CT26 (CRL-2638), EMT-6 (CRL-2755), A20 (TIB-208), and 4T1 (CRL-2539™). MC38 cells were obtained from Kerafast (Catalog #: ENH204). All cells were maintained in RPMI 1640 medium (Gibco^TM or Sigma-Aldrich), except EMT-6 cells which were maintained in DMEM (Gibco^TM). Media were supplemented with 10% fetal bovine serum (Gibco^TM or Biowest) and 1% penicillin-streptomycin (Gibco^TM or Sigma-Aldrich). A20 cells were further supplemented with 0.05 mM 2-mercaptoethanol (Gibco^TM). Cell lines were cultured in a humidified tissue culture incubator at 37 °C with 5% CO₂. Cells were subcultured when confluent and Trypsin-EDTA (0.25%) (Thermo Fisher Scientific) was used to remove adherent cells.

In vivo mouse therapy studies

Tumors were established in 6–11 weeks old female BALB/c (3 × 10⁵ CT26 cells/tumor, 5 × 10⁵ EMT-6 cells/tumor, 5 × 10⁵ A20 cells/tumor, 0.5 × 10⁵ 4T1 cells/tumor (6-thioguanine clonogenic assay), or 1 × 10⁵ 4T1 cells/tumor) or C57BL/6 (3 × 10⁵ MC38 cells/tumor) mice. The mice were inoculated by s.c. injection of cells in 100 μL serum-free medium. Tumors were allowed to reach a mean size between 70 and 199 mm³ before randomization. In the splenocyte transfer study, the mice were randomized when tumors were palpable. CarboCell injection volume was 50 μL and was injected intratumorally unless stated otherwise. Mice were anesthetized using ~3–5% sevoflurane or isoflurane during injections. RT was delivered using a small animal irradiator (1 Gy/min., 320 kV/12.5 mAs, X-RAD 320, Precision X-Ray, Inc.). Oxaliplatin loaded liposomes were formulated as previously described in ref. ⁵⁵. aPD-1 (anti-mouse PD-1 (CD279), clone RMP1-14) and aPD-L1 (anti-mouse PD-L1 (B7-H1), clone 10 F.9G2) were administrated by i.p. injection.

Tumor size (volume = length x width²/2) and bodyweight was measured 2-3 times/week. Mice were sacrificed once tumors reached between 1000 mm³ and 1500 mm³ depending on the study. Mice were terminated from studies in case of ulcerations, failure to thrive, respiratory distress or weight loss >15%. Tumor-free mice were analyzed for protective immunity by contralateral injection of cancer cells (rechallenge) as described above.

Immunohistochemistry

a-MLKL IHC

Tumor sections (4 μm section FFPE) were heated to 40 °C overnight and to 60 °C for 1 h the following day before deparaffinization in Histo-Clear II (BioNordika) and rehydration. Preconditioning was performed in citrate buffer (pH 6.0) with heating until the boiling point for 15 min. The tissue slides were blocked with peroxidase-blocking solution (DAKO), pre-incubated with a 2% BSA (Sigma) solution in PBS, and incubated overnight with the primary antibody, phospho-MLKL Ser345 recombinant rabbit antibody (Invitrogen, MA5-32752) diluted in 2% BSA solution to a concentration of 1/100. The following day, the slides were incubated with the secondary antibody EnVision+ System-HRP Labelled Polymer (Anti-rabbit) (DAKO) followed by incubation with the DAB+ system (Agilent). Counter-staining was performed with Haematoxylin solution and slides were dehydrated and cover glass was mounted. Stained slides were scanned using the automated slide scanner for brightfield and fluorescence microscopy, Axio Scan.Z1 (ZEISS) and processed using the microscope software ZEN Lite 3.2 (ZEISS).

a-CD8 IHC

Tumor sections were cut (4 μm section FFPE), mounted onto adhesion slides, deparaffinized, and rehydrated to water in decreasing concentrations of ethanol. Preconditioning was performed in Tris/EDTA buffer (pH 9.0) with heating for 3 × 5 min. The tissue slides were pre-incubated with 10% normal serum (goat serum) and incubated for 1 h at 25 °C with the primary antibody, anti-CD8 alpha recombinant rabbit antibody (Abcam, ab217344) diluted in 1% BSA solution to a concentration of 1/2000. Slides were blocked with peroxidase-blocking solution (DAKO) and incubated with the secondary antibody, Goat Anti-Rabbit IgG H&L (HRP) (Abcam) followed by incubation with DAB substrate kit (Abcam) and counter staining with Haematoxylin solution. Dehydrated slides had cover glass mounted using Pertex and were imaged (LEITZ DMRB microscope, MC170 HC camera, Leica Application Suite V4.9 (Leica)).

Gene expression analysis

Quantitative polymerase chain reaction (qPCR) was performed to evaluate the expression of IRF7, CXCL9, CXCL10, CCL4, and TNF-α in tumors and tdLNs. Tumors and tdLNs were excised, placed in cryo-tubes, snap-frozen in liquid nitrogen, and stored at −80 °C until further processing.

Tumors samples ( > 40 mg) were transferred to extreme-temperature-fluctuation-resistant bags, tissueTUBE TT05 Extra Thick (Covaris), immersed in liquid nitrogen, and pulverized using the CryoPREP Automated Dry Pulverizer (Covaris). 20 mg of smashed tumor tissue was transferred to CKmix bead beating tubes together with 350 μL RLT lysis buffer (QIAGEN) and 1% β-mercaptoethanol. Tumor samples ( < 40 mg) and tdLNs were added directly to bead beating tubes. The samples were homogenized (Precellys Evolution, Bertin Instruments, settings: 25 s, 0 °C, 6000 RPM, 2 cycles). Afterwards samples were spun down at ≥ 12.000 g, 4 °C, 3 min. A maximum of 350 μL of supernatant was transferred to 2 mL RNase-free Eppendorf-tubes. RNA extraction was carried out by column-based separation on the QIAcube connect instrument (QIAGEN) following the manufacturers protocol: “RNease miniprep with DNase digest”. The concentration of RNA was determined on the NanoDrop^TM One Microvolume UV-Vis Spectrophotometer (Thermo Fisher Scientific) and the samples were stored at −80 °C until further processing.

cDNA was synthesized in strip-tubes from 300 ng extracted RNA by reverse transcription (RT)- PCR using the High-Capacity RNA-to-cDNA^TM Kit (Thermo Fisher Scientific) by following the manufacturer’s instructions. To check for genomic contamination, a no template control sample was run without RNA. The RT-PCR was performed on the Eppendorf Mastercycler Gradient PCR instrument, under the thermal conditions (37 °C for 60 min., 95 °C for 5 min.) and stored at −20 °C for qPCR analysis.

qPCR was performed by mixing 2 μL cDNA with 18 μL PCR-grade water containing 50% TaqMan Fast Advanced Master Mix (ThermoFischer Scientific) and 5% FAM-MGB fluorophore-conjugated primer and probes mixtures (TaqMan Gene Expression Assay primers and probes, Thermo Fisher Scientific) for each gene of interest (see Supplementary Table 12 for an overview of primers and probes) and housekeeping genes (Gapdh and Ldha). A control sample without cDNA were included to ensure the purity of the samples and reagents. The samples were run in non-skirted 96-well plates sealed with optically clear flat 8-cap strips (ThermoFischer Scientific) on the Mx3005p qPCR instrument (Agilent Technologies) under thermal conditions (95 °C for 3 min., 40 cycles of 95 °C for 30 s, 60 °C for 60 s).

Before running the qPCR samples, all primers and probes were examined to optimize the settings of the Mx30005p qPCR instrument. The primer and probes were tested by creating a fivefold serial dilution of the cDNA and the standard curve was run under the same thermal conditions as the samples. Subsequently, the settings for the raw fluorescence signal were adjusted so that the signal was detected within the optimal range of detection (6000-35000 relative fluorescence unit). Additionally, the efficiency value (R2) was calculated from the standard curve to ensure the quality of the primers and probes. R2 in the range of 85%–110 % was accepted.

All qPCR data was processed using the MxPro qPCR software (Agilent Technologies) and normalized to the two housekeeping genes. Relative gene expression was calculated as previously described in ref. ⁶⁷.

Cytokine levels in murine tumor samples

Excised tumors were snap-frozen in liquid nitrogen and stored in cryotubes at -80 °C. Tumors were pulverized (CryoPREP automated Dry Pulverizer, Covaris) and resuspended in Irvine buffer (5 μL/mg tumor) consisting of 50 mM Tris-HCl, 150 mM NaCl, 10% glycerol, 1% NP-40 Surfact-Amps^TM Detergent solution (10%) (Thermo Fisher Scientific) and MilliQ water. The buffer was adjusted to pH 7.5, and 10 μL Halt^TM protease and phosphatase single-use inhibitor cocktail (100X) (Thermo Fisher Scientific) was added per mL buffer before use. Samples were shaken at 300 rpm for 1 h at 4 °C and centrifuged for 15 min at 15000 rcf at 4 °C, and the supernatant was transferred to new tubes. The DNase Pierce universal nuclease for cell lysis (ThermoFisher Scientific) was added to the supernatants (0.1 μL/mL Irvine buffer). After addition of DNase, the samples were shaken at 25 °C for 1 h at 300 rpm.

CXCL10 and IFN-β concentrations were measured by ELISA using mouse DuoSet® kits (R&D systems) according to the manufactures instructions (Substrate Solution and Stop Solution, R&D Systems). Absorbance was measured at 450 and 540 nm (FLUOstar®Omega microplate reader, BMG LABTECH) using the Omega software (BMG LABTECH).

Flow cytometry

Tumors and tdLNs were placed in MACS tissue storage solution (Miltenyi Biotec) or PBS. Weighed tumors were minced and enzymatically digested (Tumor Dissociation Kit, mouse, Miltenyi Biotec) (37 °C water bath, 40 min. with shaking (4T1 tumors, 60 min.)). TdLNs and digested tumors were passed through 70 μm strainers and diluted in PBS. Cell counts were determined (MUSE® Cell Count and Viability Assay, MUSE® Cell Analyzer, Merck Millipore). 1–10 × 10⁶ cells were Fc-blocked (5–10 min. on ice) with murine anti CD16/32 monoclonal antibodies (BD Biosciences) followed by dead cell labelling with fixable viability dye (VD)-efluor780 (eBioscience) and surface staining (30 min. on ice) with specific antibodies (see Supplementary Tables 3–8). For intracellular staining of IFN-γ, FoxP3, and GrzB cells were fixed (30 min., 25 °C) and permeabilized using the eBioscience^TM Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific). For intracellular staining of Arg-1 cells were fixed (30 min., 25 °C) using 1% formaldehyde in PBS (Sigma-Aldrich) and permeabilized using the permeabilization buffer from the eBioscience^TM Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific). See supplementary tables (Supplementary Tables 4, 7, 8) for information regarding antibodies used for intracellular staining.

The samples were acquired on a LSRfortessa X20 Flow Cytometer using FACSDiva software (BD Biosciences) and data analyzed using FlowJo software (Treestar). Spectral spillover compensation controls were included using UltraComp eBeads^TM Plus Compensation Beads (Invitrogen) and ArC^TM Amine Reactive Compensation Bead Kit (Invitrogen). Debris and doublets were excluded based on forward and side scatter. Fluorescence minus one (FMO) samples were used for identification of positive and negative populations. See Supplementary Figs. (Supplementary Figs. 16–20, Supplementary Fig. 22, Supplementary Tables 2–7) for individual gating strategies.

6-thioguanine clonogenic assay in lung cells from 4T1 tumor bearing mice

Female BALB/c mice were treated every seventh day with intratumoral injection of CarboCell for a total of three injections starting seven days after inoculation with 4T1 cancer cells. Three days after last CarboCell injection (24 days after inoculation) lungs were harvested for detection of 6-thioguanine resistant cells (4T1 cancer cells). The lungs were minced, enzymatically digested (Tumor Dissociation Kit, mouse, Miltenyi Biotec) (37 °C water bath, 60 min. with shaking) and passed through 70 µm cell strainers and washed twice in Hank’s Balanced Salt Solution (Gibco^TM). Cell suspensions were centrifuged at 500 g (5 min., 4 °C) and pellets resuspended in RPMI-1640 medium containing 60 µM 6-thioguanine (Sigma-Aldrich) supplemented with 10% fetal bovine serum (Gibco^TM) and 1% penicillin-streptomycin (Gibco^TM). The samples were serially diluted (0x, 10x, and 100x) in a 6 well tissue culture plate and placed in a humidified tissue culture incubator (37 °C., 5% CO₂). After 10 days the 6-thioguanine resistant colonies were fixated with methanol (5 min., 25 °C), washed in MilliQ water, and stained with 0.025% methylene blue (Sigma Aldrich) (5 min., 25 °C) followed by a final wash in MilliQ water. Clonogenic colonies were verified by microscopy (ZEISS Axiocam ERc5s) and counted manually.

¹⁸F-FDG PET/CT

¹⁸F-FDG PET/CT imaging was performed using a small animal PET/CT system (Inveon®, Siemens Medical Systems, PA, USA). Fasted mice (12 h) were injected intravenously with 7.07 ± 0.42 MBq ¹⁸F-FDG in 0.2 mL 0.9% isotonic saline solution. PET/CT imaging (5 min. PET acquisition) was performed after ¹⁸F-FDG injection. During injection, distribution and scanning, the mice were kept anesthetized by inhalation anesthesia ( ~ 3.5% sevoflurane). Body temperature was kept stable by heating devices. PET scans were reconstructed using a maximum posteriori reconstruction algorithm (0.815×0.815×0.796 mm). Image analysis was performed using commercial software (Inveon, Siemens Medical Systems). 3D lung ROIs were manually constructed and the ¹⁸F-FDG positive lung volume determined using a lower threshold of > 1 standardized uptake value (SUV).

Evaluation of protective memory by splenocyte transfer

BALB/c mice carrying palpable s.c. CT26 tumors were preconditioned with cyclophosphamide (Sendoxan, Baxter) i.p. The following day, mice received splenocyte transfer from mice that had rejected bilateral CT26 tumors after unilateral CarboCell TLR:TGFb injection, from naïve mice, or from CT26 tumor-bearing mice. Before splenocyte transfer, spleens were passed through 70 μm strainers, diluted in PBS, and red blood cell lysis (VersaLyse Lysing Solution, Beckman Coulter) was performed. Spleens were again passed through 70 μm strainers and diluted in 200 μL PBS. Splenocytes from one spleen were transferred by i.v. injection into one mouse. Anticancer activity was evaluated as described above.

TCRβ CDR3 immune sequencing

Mice bearing bilateral CT26 tumors (established as above) received three unilateral CarboCell TLR:TGFb injections. Tumors were harvested one week after CarboCell treatment (31 days after inoculation). Untreated tumors were harvested 17 days after inoculation. Genomic DNA was extracted and purified from tumors using the DNeasy Blood & Tissue Kit (QIAGEN) according to the manufacturer’s instructions. RNA-free genomic DNA was obtained by adding 4 μl RNase A (100 mg/ml, QIAGEN). Immunosequencing of TCRβ CDR3 regions was performed using the survey ImmunoSeq Assay (Adaptive Biotechnologies, Seattle, WA).

Anti-PD-L1 antibody labelling

Anti-PD-L1 antibody (anti-mouse PD-L1 (B7-H1), clone 10 F.9G2, BioXcell) was labelled with Alexa Fluor (AF)647 fluorophore using the Rapid Conjugation System (Lightning-Link®, Expedeon). The antibody was first mixed with the Lightning-Link® Rapid Modifier, and then, with lyophilized AF647. Before mixing with the Lightning-Link® Rapid Quencher, the obtained solution was incubated for 15 min at 20 °C. For washing, the labelled antibody was loaded onto an Amicon® Ultra 4 mL spin filter (30 kDa cut-off) and mixed with PBS. Concentration of the antibody was adjusted using centrifugation at 4000 g and NanoDrop™ spectrophotometer was used to determine antibody concentration in the final solution. The antibody was stored at 4 °C until use.

Experimental procedure and sample preparation and 3D deep fluorescent imaging

Tumor-bearing mice received two i.v. injections, first, ~7.5 µg/g AF647-labelled anti-PD-L1 antibody and after two hours of the antibody circulation, the second one being 200 µl of AF555-conjugated wheat germ agglutinin lectin (1.5 µg/µl in PBS, Thermo Fisher Scientific, W32466). The lectin circulated for one hour, after which mice underwent transcardial perfusion with 30 mL of PBS followed by 20 mL of 4% methanol-free formaldehyde solution (Thermo Fisher Scientific). Afterwards, tumors were excised and postfixed in the same solution overnight at 4 °C. 2 mm-thick sections were obtained from the postfixed tumors for further processing. These samples underwent optical clearing procedure as follows: first, step-by-step dehydration in tetrahydrofuran (THF) in MilliQ water solutions (30% - 4.5 h, 50% THF – 4.5 h, 70% THF – overnight, 90% THF – 4.5 h, 100% THF – 4.5 h, 100% THF – overnight). For refractive index matching, dehydrated samples were incubated in dibenzyl ether (Merck) for at least 6 h. During all the steps of clearing procedure, samples were undergoing gentle agitation at 20 °C. After the optical clearing procedure was completed, the samples were imaged with laser scanning confocal microscope LSM 980 with Airyscan2 detector (Zeiss Microscopy). Images were acquired using 10x (with numerical aperture equals 0.3) objective in airyscan multiplex mode. For exciting AF647, we used 639 nm diode laser, and for AF555 excitation, we used 561 DPSS laser. Images from these two channels were acquired sequentially to avoid the crosstalk. Optical sectioning was done with oversampling ( ~ 2x Nyquist criteria) to achieve smoother signal in Z-plane for the lectin (vasculature) channel. Acquisition parameters including Z-correction were kept the same for all the samples.

Image processing and analysis

Acquired images underwent Airyscan reconstructioning (processing type: 3D, with fixed value of 4.2) and stitching in Zen Blue software (Zeiss Microscopy). After, images were exported to tif format and converted to 8 bit in Fiji^69,70.

For obtaining binary masks of a whole tumor (including both viable and necrotic regions), images from the extravasation channel underwent 2D Gaussian filtering (σ = 30) with subsequent thresholding (pixel intensity: 1–255). For obtaining masks of what can be considered viable tumor regions based on the close proximity to vasculature, images from the vessel channel underwent 2D Gaussian filtering (σ = 70) with subsequent thresholding (pixel intensity: 11–255).

For vasculature geometry analysis, images from AF555 channel (further referred to as the vessel channel) first underwent preprocessing, which consisted of the following operations executed in Fiji. Before pseudo flat field correction, to avoid erroneous corrections of intensity in the necrotic tumor regions characterized by very low or absent signal in the vessel channel, signal level in those regions was equalized. More specifically, first, such regions were separated by creating a mask through applying 2D Gaussian blur (σ = 70) with following thresholding (pixel intensity: 3–255), resulting in all regions with no structured signal from AF555 lectin as a foreground. This mask was subtracted from the initial lectin image. Then, we subtracted a value of 242 from the mask, and added it to the image resulting from the previously described subtraction operation. In such way, all regions with low non-structured, or absent signal from AF555 lectin gained an intensity value of 3. After, pseudo flat field correction⁷¹ was applied (blurring value = 300). Then, the initial binary mask was subtracted from the corrected image of the vessel channel. The resulting image underwent pseudo deconvolution as follows: a copy of the image underwent 3D Gaussian filtering (σ(x,y)=40, σ(z)=12) and was divided by 1.25, after which it was subtracted from the corrected vessel image. At the final stage of the preprocessing, masks of the viable tumor tissue were subtracted from the vessel images. Preprocessed images were used to train a deep learning model in Zen Intellesis software⁷² as described earlier⁵⁴. In brief, 10 different image stacks from six different tumors were used as training dataset, and 16 images, each from different tumor, were included into a validation dataset. The training algorithm was set to extract 256 image features and conditional random field⁷³ was applied for the postprocessing. Before segmentation, the preprocessed vessel images underwent subtraction of the masks of viable tumor tissue (tissue as a background) to avoid contribution of any unstructured lectin signal from the necrotic regions into a resulting spatial graph of the vasculature. The resulting binary masks underwent postprocessing. First, the masks underwent size exclusion operation in Amira 3D (Thermo Fisher Scientific) (remove small spots module, removal threshold = <20 000 µm³ in volume). For obtaining final binary masks of the vasculature, these masks underwent 3D Gaussian blurring (σ(x,y,z)=3) and following thresholding (pixel intensity: 160–255). The resulting masks underwent 3D skeletonization in Fiji⁷⁴ and were imported to Amira, where they were further refined by distance-ordered thinning with decreased sensitivity to short (potentially spurious) branches (resistance value = 50). After this, spatial graphs of vasculature were extracted. Diameter underestimation was corrected by multiplying the original segment diameter by the correction factor quantified as described below:

$$ {CF}=\left(Y0-{Plateau}\right)\times \exp \left(-K\times {MD}\right)+{Plateau} \\ \qquad\quad\qquad\quad{{{\rm{where}}}}: {{{\rm{Y0}}}}=2.811 \\ \qquad\quad\qquad\quad\quad{{{\rm{Plateau}}}}=1.22 \\ \qquad\quad\qquad\qquad\quad{{{\rm{K}}}}=0.8472$$

MD – initially measured diameter of a vessels segment

To avoid contribution from segments traced potentially erroneously (which could have arisen from high irregularity of tumor vasculature morphology), segments shorter than 3 µm and segments with more than 10 branches were excluded from the final quantification of vascular features.

For quantifying the antibody extravasation, images from AF647 channel (further referred to as AB channel) underwent 2D Gaussian filtering (σ = 2). For acquiring binary masks of extravasation, the preprocessed images were thresholded (pixel intensity: 10–255). To avoid contribution of signal coming from the antibody trapped inside the vessels, we subtracted final vessel mask from the extravasation mask. Obtained masks were further quantified as described earlier⁵⁴. More specifically, the masks were imported to Amira software, where label field was created by the label analysis module recognizing separate 3D objects (extravasation foci). Due to the technical limitations, for generation of label fields, extravasation masks were each divided into two subsets (left and right). For quantification of the extravasation part in the viable tumor tissue, we subtracted masks of the viable tumor tissue (viable tissue mask as a background) from masks of total extravasation.

For defining presumably permeable vessel segments (the ones in close proximity (20 µm) to extravasation), binary masks of extravasation underwent 3D dilation through 3D maximum filtering in sphere (x, y = 44, z = 3) using CLIJ2 plugin⁷⁵ in Fiji. Obtained masks were imported to Amira and after undergoing multithresholding, they were used for labelling of spatial graphs of the vasculature.

Tolerability evaluation in experimental beagle dogs

Studies in dogs were performed under Hungarian Animal Experiments License. Under general anesthesia dogs were s.c. injected in the dorsal neck region with CarboCell (CC-4, Supplementary Table 1). For injection of CarboCell dogs had a venous catheter placed in the cephalic vein and were anesthetized by i.v. injection (5 μg/kg fentanyl, 0.25 mg/kg midazolam and 0.5 mg/kg ketamine). The area between the shoulder blades was clipped and aseptically prepared for injection (23 Ga needle and 1 mL syringe) with CarboCell TLR:TGFb. Clinical examinations were performed by experienced veterinary oncologists on specified time points. Rectal temperature was measured, and blood samples collected (vacutainer system, two 2 mL EDTA and two 2 mL clot-activating serum tubes) immediately before injection of CarboCell and 4 h, 24 h, 3 days, 5 days, 7 days, 9 days, and 14 days after. Serum biochemistry (Normachem 200 (Norma Instruments, Budapest, Hungary)) and hematology (BC-2800, Mindray, Shenzhen, China) was performed on all blood samples except the 4-hour samples. Serum samples were aliquoted in two 1.5 mL Eppendorf tubes and stored at −80 °C until further analysis. On day 7 (group A) or day 14 (group B and C) dogs had the CarboCell depots surgically recovered under general anesthesia and analyzed for remaining concentration of TLR7/8a and TGFβi.

TLR7/8a and TGFβi concentration in CarboCell depots recovered from beagles

Recovered CarboCell samples were heated to 50 °C and stirred at 600 rpm to promote dissolution. The samples were subsequently filtered through 0.45 μm pore syringe filters and analyzed by RP-HPLC by Shimadzu Nexera-I with a PDA detector using a Waters XTerra RP-C8 column (5 μm, 4.6 × 50 mm). The solvent system consisted of solvent A (5% MeCN, 0.1% TFA in water) and solvent B (0.1% TFA in MeCN). The gradient was 5-100% B over 15 min. with a flow of 1.0 mL/min. Column temperature: 40 °C. Sample temperature: 21 °C. Injection volume: 10 µL. The quantification of the individual compounds R848, RepSox, and LOIB was measured at 300, 300 and 220 nm respectively. HPLC results were correlated to standard curves and the CarboCell release calculated based on the RepSox/LOIB or R848/LOIB ratio relative to the reference material ratio.

Cytokine and chemokine measurements

Serum from CarboCell injected dogs was analyzed with a canine cytokine multiplex assay (Millipore, CCYTOMAG-90K-08C premixed) on Luminex (Invitrogen). Serum samples were vortexed and centrifuged before being diluted 1:3 in assay buffer (20 µL serum and 40 µL assay buffer) and added to the plate. Premixed beads were diluted 1:2 and 25 µL was added to each well. The plate was shaken overnight at 4 ˚C at 800 RPM. Following the washing procedure with magnetic plate, the detection antibody was diluted 1:2 and 25 µL added to each well and incubated for 1 h at 25 °C and 800 RPM. Subsequently, 12.5 µL Streptavidin-phycoerythrin was added to each well and the plate was incubated for 30 min. before being washed twice. The samples were acquired by Luminex according to manufacturer’s protocol.

Statistics

Statistical analysis was performed in Prism 8 and 9 (GraphPad Software). Flow data was analyzed using FlowJo (Treestar) and analyzed in Prism. Comparison between two groups was performed with unpaired t test and comparison between multiple groups with one-way analysis of variance (ANOVA) with Tukey post hoc test for multiple comparisons unless otherwise stated in figure legends. Sequencing data were analyzed using immunoSEQ ANALYZER 3.0 (Adaptive Biotechnologies). All data are presented as the means ± SEM, unless otherwise stated. A p-value < 0.05 was considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Source data

Source Data