Multidrug Analyses in Patients Distinguish Efficacious Cancer Agents Based on Both Tumor Cell Killing and Immunomodulation

Tumor response to cancer treatment is highly patient-specific and is mediated by complex interactions between drugs, genetically diverse tumor cells, and a heterogeneous tumor microenvironment (1, 2). The wide diversity of tumors observed in the clinic for any given solid tumor indication presents a challenge to development of novel effective drug therapies. Most preclinical translational models do not adequately capture intratumoral or interpatient heterogeneity. This issue is underscored by the 85% failure rate of new anticancer agents in the clinical setting (3). The era of genomics, highlighted by remarkable advances in high-throughput next-generation sequencing technology, has sparked hope that this issue can be solved through a precision medicine approach to cancer treatment: matching the right drugs to the right patients. However, despite some clear early successes such as the greater than 90% response rate to imatinib in patients with BCR-ABL–mutated chronic myeloid leukemia (CML; refs. 4, 5), genome-based precision medicine has been limited by an incomplete understanding of the relationship between cancer mutations and drug responsiveness (6). Furthermore, tumor mutational status likely represents only one factor, among many, with potential to influence solid tumor responsiveness to drug therapy (2, 7). Thus, opportunity exists to employ novel technologies that capture the many facets of solid tumor heterogeneity, to ultimately improve overall patient outcomes through more informed translational drug development.

Soft tissue sarcoma (STS) represents a highly heterogeneous disease that accounts for approximately 1% of all adult and 15% of pediatric cancers, with more than 75 subtypes currently recognized, and a median survival of 15 months in the metastatic setting (8). The anthracycline doxorubicin has been employed in the STS clinic since the 1970s and is still the most frequently used agent in the first-line treatment of metastatic STS, with a response rate of 12%–24% (9, 10).

As responses to doxorubicin are mostly partial responses, and durable responses to single-agent treatment are infrequent, combination therapies with doxorubicin are extensively being investigated. Given that responses are likely driven by a combination of direct drug effects on tumor cells, influences of the surrounding tumor microenvironment, and the extent of response by the patient immune system, assessments of drug combination efficacy should be pursued in contexts that capture this multidimensional activity.

As a precursor to investigation in humans, we directly tested multiple drugs simultaneously with Comparative In Vivo Oncology (CIVO)-arrayed microinjection technology (11) in canine patients with STS. This allowed us to study drug effects in a diverse set of patients each with their own unique tumor genomic profile, intact tumor microenvironment, and immune system functional status (12). The CIVO approach utilized here is designed to be toxicity-sparing to the patient, as each drug or drug combination is introduced directly into the tumor at microdose levels, thus avoiding systemic drug pharmacologic activity. Drug sets in this study included standard FDA-approved chemotherapy agents used to treat STS, but also included a novel preclinical lysosomal inhibitor, the dimeric quinoline, PS-1001 (formerly Lys-05; refs. 13, 14). Like most solid tumor trials, the responses elicited were observed in both a drug- and patient-specific manner, capturing complex responses including those of tumor and immune cells.

This study was conducted in privately owned dogs with naturally occurring STS that were recruited and enrolled through board-certified veterinary oncologists and surgeons at specialty clinics and remained under the custody and care of their owners and the participating veterinarians. The goal of the study was to assess the performance of the CIVO technology in a diverse set of patients with naturally occurring autochthonous tumors, with a focus on the ability to detect drug and patient-specific localized perturbations of tumor cells and the local microenvironment following CIVO microinjection of diverse agents. All procedures in this study were reviewed and approved by the Institutional Animal Care and Use Committee. Informed consent was obtained from owners prior to enrollment in the study. Clinical data was obtained from the participating veterinarian, owner, and medical records. Inclusion criteria for the study included a diagnosis of presumptive or confirmed STS based on cytology or histopathology, the presence of an easily accessible STS mass of 3 cm in diameter or more, and the absence of any contraindications to anesthesia or surgery. Thirty dogs with injectable STS were enrolled within a 22-month period (Supplementary Fig. S1). Up to 8 drugs or drug combinations were microinjected per patient along with a fluorescent tracking marker (FTM; fluorescent yellow, magenta, or orange dyes loaded in polymer microspheres and suspended in a liquid solvent containing glycerin, ethanol, propylene glycol, and witch hazel manufactured by Millennium Colors, Inc.) to denote the position of each injection site. Injections delivered by CIVO are set 4.5 mm apart. This distance was determined to be optimal for assessment of multiple drugs simultaneously while avoiding encroachment of one sample into another as previously established by analyzing the distribution of radiolabeled compounds in a tumor following CIVO microinjection (11). Board-certified veterinary pathologists confirmed all tumors included in this study were STSs. Only tumor samples that were intact with sufficient viable tumor tissue and identifiable FTM microinjection sites were included in the analysis. Of the tumors included, only microinjection sites that were clearly identifiable by the presence of FTM following histologic staining and fluorescent whole-slide imaging were included in the final analysis.

Following owner consent, a diagnostic work-up to confirm eligibility, and enrollment, dogs were briefly sedated or anesthetized for ultrasound imaging of the candidate tumor mass and for CIVO microinjection. FTM was mixed at a final concentration of 5% with each drug and/or vehicle solution prior to device loading. For tissue orientation purposes, a fluorescent magenta or orange FTM was typically included in needle 1 and 5 in addition to or in lieu of the fluorescent yellow FTM, which was included as the sole FTM in other needles. Up to 8 microliters of the suspension were then injected per needle into the patient's tumor using the CIVO microinjection device. Samples were microinjected along a 10-mm axis as single agent or in combination. Most microinjections delivered doxorubicin (Pfizer), docetaxel (Sagent Pharmaceuticals), and gemcitabine (Zydus Hospira), 3 well-characterized agents commonly used in the human STS clinic (8). The clinical-grade injectable forms of these drugs were diluted in sterile water for microdose injections. PS-1001 (formerly Lys-05, University of Pennsylvania; synthesized by Agile Discovery Partners; refs. 13, 14), a preclinical investigational compound diluted in water was also included to demonstrate the feasibility of obtaining clinically relevant information prior to drug optimization for systemic administration studies in a patient. In addition, some patient tumors received pilot drug microinjection sets that included other agents or drug combinations (Supplementary Table S1) to prioritize drugs that warranted further analysis in future studies. These included mafosfamide (Santa Cruz Biotechnology), everolimus (Afinitor, Novartis), palbociclib (Pfizer), and hydroxychloroquine (USP), or combinations of doxorubicin + mafosfamide, docetaxel + gemcitabine, and doxorubicin + hydroxychloroquine.

Drug microdose calculations were guided by FDA guidelines for Exploratory IND (Investigational New Drug) studies. For instance, based on standard intravenous dosing of doxorubicin in humans, 60 to 100 μg would be allowed per patient for this compound under Exploratory IND guidelines (15). In this study, we used 13.5 μg of doxorubicin per microinjection with no more than four injection sites per patient containing doxorubicin.

Forty-eight to 72 hours following the CIVO microinjection procedure, the injected tumors were surgically removed by participating veterinary surgeons. This timeframe was selected on the basis of optimal signal intensity of drug-specific tumor responses (e.g., DNA damage, mitotic arrest) used in this study in part as surrogate markers for drug distribution, and to avoid potential time-dependent loss of these markers due to drug clearance observed in previous preclinical studies (11). This timeframe was also selected to match that of the design of our current investigation of CIVO in the human sarcoma clinic where in most cases, especially with intermediate and high-grade cancer, significantly delaying surgical resection of STS is avoided (16). Upon resection, injection columns were identified macroscopically with a LED NIGHTSEA BlueStar Flashlight with an excitation wavelength of 440 to 460 nm, combined with the matched long-pass 500 nm filter glasses (Electron Microscopy Sciences system). Injected tumor tissue was then promptly sectioned into 2-mm thick cross-sections and placed in cold 10% neutral-buffered formalin containing protease and phosphatase inhibitors and processed for histologic staining and localized drug response evaluation.

To capture phenotypic responses to drug exposure, tumor sections were subjected to IHC staining with a battery of antibodies to detect effects on both cancer and immune cells. These included drug-specific pharmacodynamic biomarkers such as phospho-γH2AX (γH2AX) as a measure of DNA damage induction (17), and phospho-Histone H3 (pHH3) as a measure of mitotic arrest (18). Tumor cell death was measured by loss of vimentin-positive cells, and tumor cell apoptosis was tracked by cleaved caspase-3 (CC3) staining. T-cell and macrophage/monocyte accumulation was tracked by CD3 and S100A9 (MAC387) staining, respectively.

The tumor biopsies were fixed for 48 hours for processing and embedding in paraffin wax. Four micron sections were cut, placed on glass slides, and stained with Harris hematoxylin and eosin Y reagents (H&E), Masson Trichrome reagents, or specific biomarkers via IHC. For fluorescent IHC, sections were incubated overnight in primary antibodies including rabbit anti-cleaved caspase-3 (1:150; Cell Signaling Technology), rabbit anti-Light Chain 3 isoforms A/B (LC3A/B, MAP1LC3A/MAP1LC3B; 1:200; Cell Signaling Technology), mouse anti-MAC387 (1:1,000; Dako), rabbit anti-phospho-γH2AX (1:500; Novus Biologicals), rabbit anti-phospho-Histone H3 (1:500; Millipore), rabbit anti-calreticulin (1:500; Abcam), and mouse anti-vimentin (1:100; Abcam). Tissues were incubated with secondary antibodies raised in goat and conjugated to AlexaFluor 647 (1:600; Jackson Immunoresearch) for 1 hour at room temperature. IHC costains were completed with the following pairs of antibodies: rat anti-CD3 (1:250; Abcam) and rabbit anti-Granzyme B (1:500; Abcam), mouse anti-MAC387 (1:1,000; Dako), and rabbit anti-phospho-STAT1 (1:100; Cell Signaling Technology), mouse anti-MAC387 (1:1,000; Dako) and rabbit anti-c-Maf (1:250, Santa Cruz Biotechnology), mouse anti-CD3 (1:200; Leica) and rabbit anti-phospho-STAT1 (1:100; Cell Signaling Technology). Detection for costains was completed using secondary antibodies raised in goat and conjugated to AlexaFluor 555 (1:400; Invitrogen) and AlexaFluor 647 (1:600; Jackson Immunoresearch). DAPI (2 μg/mL; Invitrogen) nuclear counterstain was applied for 10 minutes prior to addition of coverslips to slides with Prolong Gold mounting medium (Invitrogen). Slide images were acquired in brightfield (H&E) or fluorescent mode (fluorescent IHC stains) at ×20 magnification using a 3DHistech Pannoramic 250 Flash digital slide scanner (3DHistech).

Tumor and immune cell responses were quantified from each whole slide image using Presage custom CIVO Analyzer image analysis platform as previously described (11). Briefly, using Cell Profiler (19), whole-slide images were segmented into cellular objects using the nuclear (DAPI) channel and classified as biomarker-negative or biomarker-positive for each antibody or histology stain. Following cellular segmentation and classification, circular regions of interest (ROI) were identified around each microinjection site in each image. Cellular responses were then quantified as functions of radial distance from the injection site, which has been shown to be a surrogate for drug concentration (inverse relationship; refs. 11, 20). MAC387 was quantified as the local density of biomarker-positive cells (cells/mm²) and plotted versus radial distance from the center of the injection site. Other biomarkers including γH2AX and phospho-histone H3 were quantified as the fraction of cells that were biomarker-positive and plotted versus radial distance from the center of the injection site. To account for varying background levels of each biomarker, the response located 2,000 μm distance away from each injection site was measured and subtracted from each corresponding curve. Thus, all measurements represent the change in biomarker expression from background and agents that suppress a biomarker will exhibit negative values. Finally, to obtain overall metrics of response, regions were manually drawn at the outer edge of apparent alterations in biomarker expression and the effective radius of the region was calculated using the square root of the area divided by π (Figs. 3A and 4C). Responses were quantified from up to two tissue sections per tumor, representing multiple depths along the injection column approximately two millimeters apart from one another.

The RAW 264.7 mouse macrophage cell line was obtained from ATCC and cultured under adherent conditions using DMEM + 10% FBS (Thermo Fisher Scientific) in a standard humidified 37°C incubator. Cell line verification and mycoplasma testing was not performed as vials of RAW 264.7 cells received from ATCC were immediately thawed and used at passage 10 or less for all experiments. Lipopolysaccharide (LPS) + IFNγ (Sigma Aldrich; Shenendoah Biotechnology) was used to induce the M1 activation state in RAW 264.7 cells, whereas IL4 (Shenendoah Biotechnology) was used to induce the M2 activation state. RAW 264.7 cells (3 × 10⁵) were plated in 6-well 10-cm² plates in triplicate and allowed to attach overnight. To assess the impact of PS-1001 on M2 activation markers, 5 ng/mL IL4 was added to triplicate wells either alone (Control) or in combination with PS-1001 (3 μmol/L) and incubated for 24 hours. To assess the impact of PS-1001 on M1 activation markers, 5 ng LPS + 1 ng IFNγ were added to triplicate wells either alone (Control) or in combination with PS-1001 (3 μmol/L) and incubated for 24 hours. Cells were harvested for RNA and M1 and M2 activation state was monitored via qPCR TaqMan probes (Thermo Fisher Scientific) specific for mouse IL6, iNOS, Socs3, and Arg1, Ym2, Mrc1, Socs1, respectively.

RAW 264.7 mouse macrophage cells were plated at a density of 2 × 10⁶ cells in T-25 tissue culture flasks and incubated overnight to allow for attachment. Following attachment, the media were exchanged with fresh media containing 10% FBS with or without 3 μmol/L PS-1001. Twenty-four hours following media exchange, the overlying media was harvested and spun at 1,000 × g for five minutes to remove cell debris. Fifty microliters of control supernatant or PS-1001–containing supernatants were analyzed using the Multi-Analyte ELISArray Kit (Qiagen, catalog no. MEM-003A) according to the manufacturer's protocol. Supernatant samples were run in triplicate and read for absorbance as per manufacturer's protocol on an Envision microplate reader (Perkin Elmer).

The final analysis was performed on complete and verified datasets obtained from each center in accordance with internal procedures. Mean radial response curves for each measurement were generated and presented with SE bars. Statistical comparisons were made by computing the areas under the individual curves and modeling the results with a linear mixed effects model with the injected agent as a fixed effect and the tumor as a random effect. Statistical significance was set at P < 0.05 via Wald test. Cluster analysis was performed using K-means clustering. The SD of fold induction was calculated using triplicate experimental samples for all qPCR assays. The SD of fold change over control was calculated using triplicate wells for all cytokines assayed via ELISArray kit.

Of the 30 CIVO microinjected patient tumors, 18 were sufficiently viable (nonnecrotic) to evaluate responses to microinjected drugs and were included for analysis in this study (Supplementary Table S2). One hundred twenty-three out of 144 potential injection sites (85%) resulted in identification of microinjected FTM (Fig. 1A) and were subsequently analyzed. Of the sites eliminated from analysis, 7 (5%) were due to injections into necrotic tissue, 3 (2%) were due to lack of detectable FTM delivery, and 11 (8%) were due to mixing of FTM and hence drug from adjacent injection sites.

Initial assessment of tumor response to microdosed drugs focused on doxorubicin, docetaxel, and gemcitabine. As observed previously (11, 20), CIVO-microinjected drugs resulted in spatially localized tumor response zones (generally < 3.0 mm in diameter) with induction of histologic phenotypes specific to the known mechanism(s) of each introduced agent (Fig. 1B–F). In many cases, tumor responses were sufficiently robust to observe by standard hematoxylin and eosin (H&E) staining. Notable regions of tumor cell clearance surrounded by a periphery that included scattered cells exhibiting small dense nuclei were often observed at sites of doxorubicin injection (Fig. 1B and C), whereas regions enriched with rounded cells exhibiting abnormal mitotic spindles were commonly observed at sites of docetaxel injection (Fig. 1D). Gemcitabine did not induce any clear phenotypic response compared with uninjected regions of the tumor (Fig. 1E and F).

Biomarker analysis was also in line with expected pharmacologic activity of each agent. Doxorubicin is a topoisomerase II inhibitor, which intercalates into DNA, leading to DNA damage and production of phosphorylated γH2AX (21). Consistent with this activity, significant (P < 0.001) induction of γH2AX was observed at doxorubicin sites (Fig. 1G–I). Docetaxel binds to microtubules preventing their disassembly and turnover, leading to cell-cycle arrest at G₂–M phase of the cell cycle and to cell death (22). Consistent with this established activity, docetaxel was the most prominent inducer of localized enrichment of phospho-histone H3 (pHH3)-positive cells (P < 0.001; Fig. 1J–L). In contrast to doxorubicin or docetaxel, gemcitabine rarely resulted in any biomarker changes with only one patient exhibiting significant induction of γH2AX. Like previous studies, the intensity of observed responses decreased with increasing radial distance from the center of the injection site, reaching background levels by 1,700 μm (Fig. 1I and L). As induction of γH2AX by doxorubicin and pHH3 by docetaxel were highly penetrant across the study population, these signals were used as surrogate biomarkers of drug distribution in downstream analyses of tumor responsiveness to drug exposure.

No CIVO-related serious adverse device effects and two nonserious possible adverse device effects were reported. Both nonserious effects were reported as mild to moderate inflammation of the injected tumor mass, which was rectified upon surgical removal of the tumor with no further complications reported. These data support that CIVO microinjection of FTM and FTM plus multiple distinct anticancer agents at microdose quantities into a single tumor mass do not pose significant risk to patients on study.

Of the drugs investigated in our study, doxorubicin is the most frequently employed single-agent first-line therapy for STS. Consistent with this, doxorubicin was the only agent that induced regional loss of vimentin-positive sarcoma cells across multiple tumors (41% of doxorubicin injection sites, Fig. 2A and B). Most of the tumor cell death appeared to be due to necrosis with few cleaved caspase-3 (CC3)-positive cells observed within the regions of doxorubicin exposure (Fig. 2C). The superior cytocidal activity of doxorubicin was unlikely due to greater tumor biodistribution as regions of biomarker-positive responses around sites of drug microinjection were generally comparable between agents when compared within each tumor (Fig. 1, compare G to J, and I to L).

A potential reason that doxorubicin has historically outperformed other agents in the STS clinic is that exposure to doxorubicin induces changes in the immune microenvironment that result in multiple, potentially coordinated, tumoricidal activities in addition to direct cancer cell–autonomous effects. For instance, doxorubicin has been shown to recruit myeloid cells and CTLs into solid tumors (23), the latter triggered through the process of immunogenic cell death (ICD; ref. 24). We therefore assessed whether CIVO captured localized changes in the immune microenvironment specifically induced by doxorubicin. Standard H&E staining showed infiltration of multiple immune cell populations that included lymphocytes, neutrophils, and macrophages directly into regions of cell death induced by doxorubicin exposure (Supplementary Fig. S2). Further characterization with the macrophage-selective antibody MAC387 showed that macrophages represented a substantial portion of the infiltrating cells (Fig. 2D).

Furthermore, consistent with well-documented activity as a potent inducer of ICD, robust accumulation of CD3-positive T cells was observed surrounding the perimeter of the tumor cell death zone (Fig. 2E). Costaining for CD3 and granzyme B (GZMB) often showed polarization of GZMB within the CD3-positive cell when adjacent to a nearby sarcoma cell, suggesting that this double-positive cell population represents active CTLs (Fig. 2F; ref. 25). ICD is characterized by tumor cell necrosis and subsequent release of immunogenic signals called damage-associated molecular patterns (DAMP) that are recognized by T cells, which in turn initiate antitumorigenic immune responses (24, 26). Consistent with an ICD-triggered effect, the necrotic cells surrounded by the CD3-positive perimeter exhibited translocation of calreticulin (CRT), a hallmark DAMP that mediates ICD (Fig. 2H; ref. 27). The observed recruitment of macrophages and T cells was specific to doxorubicin, but was not penetrant across all doxorubicin sites, which is to be expected for a drug effect when tested in a diverse patient population consisting of tumors with highly variable tumor microenvironments. These observations demonstrated that when employed in the context of native tumors, CIVO captures immune responses consistent with the known biology of a drug such as doxorubicin and may be used more comprehensively to characterize the biology of novel anticancer agents during the translational phase of drug development.

Doxorubicin failed to induce tumor cell killing in over half of the tumors in our study. This apparent resistance to doxorubicin may be mediated by intrinsic (cancer cell autonomous) mechanisms, or to extrinsic factors such as stromal barriers to drug penetration (7, 28). We therefore assessed whether CIVO could differentiate among these two possibilities. To do so, we used γH2AX positivity as a surrogate biomarker of doxorubicin distribution and loss of vimentin-positive cells as a measure of tumor responsiveness. By plotting tumor response versus the maximum doxorubicin distribution and applying K-means clustering analysis, patient tumors were classified into three distinct groups: (i) drug sensitive: characterized by high γH2AX distribution (mean ± SD: 1.044 ± 0.152 mm) and high vimentin clearance (0.645 ± 0.137 mm), (ii) poor distribution: characterized by low γH2AX distribution (0.498 ± 0.154 mm) and low vimentin clearance (0.210 ± 0.092 mm), and (iii) drug resistant: characterized by high γH2AX distribution (0.978 ± 0.097 mm) and low vimentin clearance (0.137 ± 0.090 mm; Fig. 3A). Plotting of the first two groups showed a clear correlation between doxorubicin distribution and tumor response. Consistent with structural features of the tumor microenvironment playing a role in preventing doxorubicin distribution and efficacy, Masson Trichrome staining revealed increased tumor cell density and thickened collagen matrix in some tumors within the poor distribution subgroup when compared with tumors from the other two subgroups (Supplementary Figs. S3 and S4). In contrast, the lack of substantial tumor cell loss despite extensive γH2AX distribution observed in the drug-resistant tumors suggests a sarcoma cell–autonomous mechanism of drug resistance to doxorubicin. Therefore, with the hypothesis that cell-autonomous resistance may be overcome by a drug combination approach, we focused the rest of our studies on this group of resistant tumors.

We next investigated whether autophagy inhibition could reverse resistance to doxorubicin. Autophagy is the major lysosomal pathway that maintains cellular homeostasis by recycling damaged intracellular components into amino acids and has emerged as a key process modulating response to cancer therapies such as doxorubicin (29–31). However, the impact of autophagy inhibition in the context of cancer treatment and progression has been controversial, with data pointing to both pro and antitumor effects depending on tumor context (32). Thus, use of CIVO technology in this diverse set of patients provides a unique opportunity to explore whether autophagy inhibition provides benefit over conventional agents alone in a context-specific manner.

For this analysis, we investigated whether exposure to a novel lysosomal autophagy inhibitor, the bivalent quinoline, PS-1001 would influence the antitumor responses exerted by doxorubicin. PS-1001 was microinjected as a single agent and in combination with doxorubicin into 16 of the 18 subjects that were analyzed in our study, 5 of which were classified as doxorubicin resistant. Consistent with anticipated activity as an autophagy inhibitor, localized microinjection of PS-1001 induced the appearance of LC3-positive puncta, indicative of immature autophagosomes that are unable to fuse with the lysosome (Fig. 4A and B). Localized exposure to PS-1001 alone was not sufficient to induce observable tumor cell killing (Supplementary Fig. S5). However, following exposure to the combination of PS-1001 and doxorubicin, 3 of 5 doxorubicin-resistant tumors exhibited measurable clearing of vimentin-positive cells, similar in extent to that observed in tumors that were classified as sensitive to doxorubicin alone (Fig. 4C and D). In contrast, the addition of PS-1001 to doxorubicin had no observable effect in tumors where resistance was mediated by poor drug distribution (Supplementary Fig. S6). Thus, PS-1001 does not appear to affect barriers to doxorubicin penetration, but enhances the antitumor effect of doxorubicin by either neutralizing a cell-autonomous mechanism of STS tumor cell resistance to chemotherapy exposure, or by enhancing a doxorubicin-induced antitumor immune response, or both. Indeed, a growing body of literature suggests that autophagy regulates multiple components of the immune system including that of the tumor microenvironment (33, 34). To elucidate how PS-1001 contributes to the antitumor activity of doxorubicin in the context of tumors exhibiting resistance to doxorubicin alone, we next investigated whether an enhanced immune response contributes to the effect of PS-1001.

Examination of localized responses to the combination of PS-1001 and doxorubicin revealed prominent infiltration of MAC387-positive cells over that observed at sites exposed to doxorubicin alone (P = 0.04; Fig. 5). This enhancement of macrophage recruitment was dependent upon our classification of doxorubicin responsiveness. All three resistant tumors that were converted to sensitive by addition of PS-1001, exhibited the response, whereas none of the poor distribution tumors exhibited enhanced macrophage infiltration following exposure to doxorubicin and PS-1001 (Fig. 5B compare left and middle columns). It is possible that the effect on macrophages is a secondary response to increased STS cell death. However, 2 of 6 doxorubicin-sensitive tumors that received the PS-1001 + doxorubicin combination also exhibited increased macrophage infiltration at sites of drug combination exposure, suggesting that increased cell death is not the only mechanism underlying the impact of PS-1001 on macrophage recruitment.

Macrophages play a complex role in cancer and can exert either pro- or antitumor activities depending on polarization state (35). The tumor microenvironment appears to direct this polarization of macrophages into two extreme subtypes, namely the antitumor M1 macrophages and protumor M2 macrophages (36). In the M1 state, tumor-associated macrophages (TAM) stimulate a Th1 response against tumor cells by activating an immune response, whereas in the M2 state, TAMs are immunosuppressive and protumorigenic. To assess whether PS-1001 influenced the polarization state of the infiltrating macrophage population, we costained tumor sections for MAC387 and the M1 biomarker pSTAT1 or the M2 biomarker c-Maf (MAF; ref. 37). Consistent with PS-1001 promoting M1 polarization, the localized MAC387-positive infiltrate was highly positive for the M1 biomarker pSTAT1 but devoid of the M2 biomarker c-Maf (Fig. 6A; Supplementary Fig. S7). To investigate whether PS-1001 directly influences the polarization state of macrophages, RAW 264.7 cells were stimulated with agents that induce either M1 (LPS/IFNγ) or M2 (IL4) polarization in the presence or absence of PS-1001, and were subjected to quantitative real-time PCR (qPCR) for biomarkers indicative of the two different polarization states. In agreement with the CIVO results, and consistent with a direct M1-promoting effect of PS-1001, expression of the M1 markers IL6, iNOS (Nos2), and Socs3 (38) were markedly enhanced, while expression of the M2 markers, Arg1, Ym2, Mrc1, and Socs1 (38) were significantly suppressed in RAW264.7 cells exposed to PS-1001 (Fig. 6B). Furthermore, treatment of cells with PS-1001 alone induced increased secretion of the proinflammatory cytokine TNFα, further suggesting a direct effect of PS-1001 on guiding macrophages toward a proinflammatory state (Fig. 6C; ref. 39).

Finally, serial sectioning and biomarker analysis was performed to examine the spatial relationship and potential functional interplay between various components of the microenvironment following exposure to doxorubicin and PS-1001. Most notable from this examination was a ring of pSTAT1⁺/MAC387⁻ cells that approximated the distribution pattern of autophagy inhibition and that surrounded a core of putative (pSTAT1⁺/MAC387⁺) M1-polarized macrophages (Fig. 7A and B). Consistent with proimmunogenic paracrine signaling between M1 macrophages and T cells, CD3 staining also showed the same distribution pattern as the pSTAT1 staining pattern and costaining revealed significant CD3/pSTAT1 positivity (Fig. 7C). Taken together, these data indicate that in concert with doxorubicin, PS-1001 increases macrophage recruitment, skews this population of immune cells toward the antitumorigenic M1 polarization state, and that this proimmunogenic influence correlates with enhanced antitumor activity in a subset of patient tumors identified by CIVO.

CIVO was developed with the premise that a shift in setting from rodent models to fully intact tumors in human patients will ultimately improve translational oncology studies and the efficiency of cancer drug development. Toward that shift, CIVO enabled toxicity-sparing investigation of multiple drugs, drug combinations, and preclinical investigational agents in canine patients with STS. Evaluation of tumor response to drug exposure using CIVO captured patient diversity, perturbations of both tumor cells and immune cells, and the impact of the structural microenvironment on drug response. The combination of PS-1001 and doxorubicin induced antitumor responses in a subset of patients that did not respond to microinjected doxorubicin alone. These responses may be due in part to modulation of the immune microenvironment as the addition of PS-1001 to doxorubicin increased recruitment of myeloid cells and skewed polarization of macrophage toward the proimmunogenic/antitumorigenic M1 state. The ability to investigate drug efficacy directly in patients without inducing toxic side effects holds promise for improving both cancer drug development and treatment for cancer patients by providing a potential next-generation functional test for precision medicine in the oncology clinic.

Accurate preclinical modeling of complex solid cancers has proven difficult, rendering interpretations of drug efficacy based on data generated in the preclinical setting challenging. The complexity of the tumor microenvironment and remarkable patient-to-patient diversity in the oncology clinic is rarely modeled appropriately prior to clinical trials. As a consequence, results from typical preclinical studies rarely forecast clinical outcomes (40). The use of a highly diverse set of patient-derived xenograft (PDX) models, many of which retain the heterogeneity of the original tumor sample, legitimately addresses issues of heterogeneity and patient diversity (41). However, limitations shared with many preclinical models remain, including lack of an intact immune system and human tumor stroma. Thus, the utility of PDX models toward investigation of agents that modulate the host immune system or target nonmalignant components of the tumor microenvironment may be limited.

Investigation of PS-1001 highlights the unique ability afforded by CIVO to evaluate an early-stage preclinical molecule in a clinical setting without significant risk of drug-induced side effects. Early toxicity-sparing evaluation of preclinical agents could enable cost/benefit analysis and prioritization of the most promising programs for full clinical development. This may be particularly important for molecules such as autophagy inhibitors given that both pro- and antitumorigenic functions have been ascribed to autophagy based largely on the context in which the study was performed (32). Few anticancer agents exhibit efficacy across all patients and CIVO analysis of the effects of PS-1001 with and without doxorubicin follows that pattern. Interestingly, CIVO identified a potential subgroup of tumors where exposure to PS-1001 may provide the most benefit. In the current cancer drug development paradigm, enormous resources are spent on optimizing a drug candidate's bioavailability, safety, and stability before its efficacy can be tested in clinical trials (42, 43). The ability to stratify potential responders based on functional testing in the context of intact tumors could mitigate this issue.

CIVO identified tumors where doxorubicin distribution was limited and these were often composed of densely packed cells and thickened collagenous matrix. While this study focused on tumor and immune cells, investigation of additional cell populations that regulate the structural microenvironment such as cancer-associated fibroblasts (CAF), is warranted. Given that CAFs mediate deposition of extracellular matrix proteins, it will be interesting to assess whether increased infiltration of CAFs coincides with increased barriers to drug distribution in both our current and future patient samples. Furthermore, CIVO may represent an efficient technology to investigate whether drug combination strategies designed to overcome CAF-mediated drug resistance could potentially loosen structural barriers and "normalize" the tumor stroma (44–46).

While CIVO detected perturbations of the immune microenvironment, it is currently unclear whether CIVO is well-suited to investigation of several new cancer drugs such as immune checkpoint inhibitors, particularly those that do not directly target tumor cells (anti-CTLA4 and anti-PD-1 agents). Accurate assessment of such agents may require development of next-generation versions of CIVO that incorporate slow-release degradable microimplants for sustained delivery of slower acting immune modulators so that changes occurring over the course of weeks are observable. We expect that such a device would at least allow observation of microenvironment changes induced by inhibition of tumor cell–expressed immune checkpoint targets (e.g., PDL1). Such development is currently ongoing in our laboratory.

While ultimately designed and intended for use in the human cancer clinic, the investigation performed here in the canine cancer clinic demonstrates the feasibility and utility of in-patient multidrug analysis in a clinically relevant setting. Previous reports have demonstrated high histologic, immunohistochemical, and genetic similarity between human and canine STS (12, 47, 48). One drawback to the canine setting, particularly for our evaluation of the prognostic potential of CIVO, is that almost all patients are treated by surgery alone with very few receiving adjunct chemotherapy. This precluded evaluation of a correlation between the localized responses induced by CIVO to responses induced by systemic delivery of drug. Therefore, to date, while such correlations have been established in vivo, these have been limited to rodent models (11, 20). Of note, our results are in line with those observed in a canine clinical trial in lymphoma where the combination of doxorubicin and hydroxychloroquine, used in this case as an autophagy inhibitor, resulted in a superior overall response rate to single-agent doxorubicin (49).

As suggested above, CIVO technology holds promise as a next-generation functional test to guide precision oncology. CIVO is agnostic to genomics, can capture influences of the structural microenvironment that are not currently evaluated by NGS, and allows investigation of multiple therapy options per patient. We therefore view CIVO as a potential complement to genomics-based approaches to guide treatment choices in the human oncology clinic. Toward that vision, we have now initiated a trial in the human STS clinic where, unlike in the canine setting, we will have the opportunity to assess whether the localized tumor responses induced by CIVO will correlate with patient outcomes when treated with the same drugs. Together, a combination of genomics and in-patient next-generation functional tests such as CIVO may enable comprehensive diagnostic approaches to accurately predict responses to cancer drugs in the context that matters—the cancer patient.

J. Frazier has ownership interest (including patents) in Presage Biosciences, Inc. A. Moreno-Gonzalez is a director of clinical technology development at Presage Biosciences. R.K. Amaravadi has ownership interest (including patents) in Presage Biosciences, has a patent licensed to Presage Biosciences, and is a consultant/advisory board member for Presage Biosciences. J.M. Olson is a director and has ownership interest (including patents) in Presage Biosciences. R. Klinghoffer has ownership interest (including patents) in Presage Biosciences. No potential conflicts of interest were disclosed.

Conception and design: J. Frazier, M.O. Grenley, K.L. Watts, A. Keener, J.M. Olson, R.A. Klinghoffer

Development of methodology: J. Frazier, J.A. Bertout, W.S. Kerwin, A. Moreno-Gonzalez, J.R. Casalini, M.O. Grenley, E. Beirne, K.L. Watts, A. Keener, D.J. Thirstrup, I. Tretyak, R.A. Klinghoffer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Frazier, J.A. Bertout, A. Moreno-Gonzalez, M.O. Grenley, E. Beirne, K.L. Watts, A. Keener, D.J. Thirstrup, I. Tretyak, S. Ditzler, C.D. Tripp, K. Choy, S. Gillings, M.N. Breit, K.A. Meleo, V. Rizzo, C.L. Herrera, J.A. Perry

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Frazier, J.A. Bertout, W.S. Kerwin, A. Moreno-Gonzalez, E. Beirne, D.J. Thirstrup, R.K. Amaravadi, J.M. Olson

Writing, review, and/or revision of the manuscript: J. Frazier, J.A. Bertout, W.S. Kerwin, A. Moreno-Gonzalez, E. Beirne, K. Choy, J.A. Perry, R.K. Amaravadi, J.M. Olson, R.A. Klinghoffer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Frazier, J.A. Bertout, J.R. Casalini, M.O. Grenley, A. Keener, J.A. Perry

Study supervision: J. Frazier, J.A. Bertout, A. Moreno-Gonzalez, R.A. Klinghoffer

Other (design and development of microinjection device used in the study):A. Moreno-Gonzalez

We would like to thank the surgeons, radiologists, and staff at VCSC, AMC, ASCS, SVS, Blue Pearl, and Summit for their contributions in making this study possible, and for the thoughtful and compassionate care they provide to their canine cancer patients. We thank Kate Gillespie and Connor Burns for their excellent work in immunohistochemistry and whole-slide imaging; Micah Ellison for his study database and image processing data management; Chantel Dixon and Angela Merrell for their technical support; Ben Weigler and Andrew Burich for their guidance and leadership regarding Animal Welfare. We thank Audrey Baldessari at Black Sheep Veterinary Pathology as well as Jennifer Ward and her staff at Specialty Vet Path for their expert pathologic review of study samples and guidance on histologic staining, and a special thanks to the owners of canine cancer patients that participated in this translational study.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.