Monitoring systemic immune responses to checkpoint inhibition in breast cancer reveals host responses and mechanisms of resistance

Abstract

Background

Immune checkpoint inhibitors (ICI) have improved survival in various cancers, but their success in breast cancer, specifically triple-negative breast cancer, remains limited, benefiting less than 10% of patients.

Methods

We modeled ICI response in vivo to unravel the mechanisms underlying immunotherapy efficacy, identify mechanisms of resistance in non-responsive tumors, and ascertain the therapeutic benefits of different chemotherapeutic combinations with ICI in breast cancer. We investigated the impact of ICI as monotherapy and in combination with other therapeutics in mouse models of mammary cancer, which we found robustly suppressed primary tumor growth and extended survival.

Results

Interestingly, even within a single model, responses to ICI were highly variable. Resistance was not reliably retained by transplantation into syngeneic hosts, suggesting a role for systemic host immunity rather than tumor-autonomous mechanisms. Transcriptomic analysis of the primary tumor landscape by fine-needle aspiration revealed that upregulated cytotoxic T-cell response and inflammatory interferon signaling (both at baseline and post anti-programmed death-ligand 1 administration) corresponded to favorable response to ICI. Longitudinal analysis of the peripheral blood uncovered enhanced myeloid cell recruitment in resistant mice, prior to therapy initiation. Similar effects were observed through longitudinal assessment of peripheral blood in patients with ICI-treated human breast cancer. Blocking myeloid cell recruitment with navaraxin (CXCR1/2 inhibitor) improved ICI responses, further suppressing tumor growth and improving survival.

Conclusions

These findings provide insight into resistance mechanisms and suggest the potential for minimally invasive strategies (sampling of systemic immune cells from peripheral blood) to identify patients likely to respond to ICI. This approach may help inform de-escalation strategies to mitigate therapeutic toxicities and limit unnecessary treatments.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Immune checkpoint inhibitors (ICIs) have demonstrated broad efficacy across multiple cancer types but have shown limited clinical benefit in breast cancer. While the addition of ICIs to standard chemotherapy has improved pathologic complete response rates in triple-negative breast cancer, most patients either respond to conventional therapy alone or exhibit resistance to ICIs. The biological mechanisms underlying this resistance and the heterogeneity in treatment response remain poorly characterized.

WHAT THIS STUDY ADDS

• This study uses a preclinical murine mammary tumor model to dissect the mechanisms of response and resistance to anti-programmed cell death protein-1 (PD-1) therapy, revealing variable therapeutic outcomes—including complete response, acquired resistance, and intrinsic resistance—even among genetically identical hosts. Notably, resistance is not tumor-intrinsic but is mediated by host-specific immune factors, as evidenced by the lack of resistance on tumor transplantation. Furthermore, the study identifies peripheral blood transcriptomic signatures enriched in myeloid cell recruitment as potential markers of resistance.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The findings herein underscore the critical role of host immunity in modulating ICI efficacy and highlight the need to integrate systemic immune profiling with tumor microenvironment analysis. The identification of peripheral biomarkers associated with resistance may inform the development of predictive diagnostics and guide patient stratification. Ultimately, these results support the rationale for combinatorial therapeutic strategies aimed at overcoming host-mediated resistance, with potential translational relevance across multiple cancer types.

Introduction

The implementation of immune checkpoint inhibition (ICI) has transformed therapeutic options for patients with cancer. ICI robustly activates the immune system to eradicate tumors. However, despite the widescale clinical success in treating several tumor types (eg, non-small cell lung cancer (NSCLC) and melanoma), it has shown more limited but clear benefit in some other solid tumors, particularly breast cancer.

Breast cancer is the most prevalent cancer among women worldwide.¹ Despite advances in diagnosis and therapeutic strategies, treatment of breast malignancies remains a significant challenge due to diverse molecular and clinical subtypes. Typically, treatment of breast cancer is dictated by the hormone receptor and human epidermal growth factor receptor-2 (HER2) status of the tumor.² In recent years, immunotherapy has emerged as a promising treatment modality in patients with breast cancer, particularly those with triple-negative breast cancer (TNBC). Despite the potential of harnessing the power of the immune system to target and eradicate tumor cells,³ many patients with breast cancer do not respond to treatment. Furthermore, among patients who are initially responsive, the effectiveness and durability of response can significantly vary, ranging from pathologic complete response (pCR) to resistance, and in some cases, recurrence. This heterogeneity in response to immunotherapy represents a major challenge in the effective treatment of breast cancer.

Many mechanisms give rise to resistance to immunotherapy,^{4 5} including tumor intrinsic factors such as low tumor mutational burden and dysregulated expression of antigen presentation machinery and processing genes.⁶ Moreover, extrinsic factors such as the immunological landscape of the tumor microenvironment can impact disease progression and response to therapy.⁷ The extensive interplay between tumor-intrinsic and tumor-extrinsic factors, for example, stromal and immune cells, likely contributes to heterogeneity in response to immunotherapy in breast cancer.^{8 9}

The objective of our study was to identify in vivo preclinical models to evaluate therapeutic resistance and heterogeneity in response, ultimately leading to discovery of combinations with enhanced therapeutic efficacy of ICI. The effective use of such models can drive insights into the dynamic breast cancer-specific mechanisms of therapeutic resistance or factors mediating favorable outcomes that can subsequently be implemented clinically to identify patient populations likely to respond.

This study presents a heterogeneously ICI-responsive in vivo model that emulates variable patient with TNBC response to ICI despite very tightly controlled experimental conditions. We describe the efficacy of single-agent ICI in upregulating cytotoxic immune cell infiltration and expansion within the primary tumor microenvironment, thereby diminishing tumor growth and enhancing survival. We also highlight differential responses in genetically similar hosts and tumors, reflecting heterogeneous patient intrinsic immunologic response to the tumor and ICI observed in clinical settings, allowing us to uncouple tumor-specific factors and elucidate a role for the host systemic immune response in therapy outcome.

Methods

Cell culture: EMT6, 4T1, and E0771 murine mammary carcinoma cells were cultured in Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12) or Roswell Park Memorial Institute (RPMI) Medium 1640 (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific).

Viral transduction: the pSHUTTLE-mCxcl1 or mCxcl2 (GeneCopoeia) was incorporated into the pINDUCER-20 plasmid¹⁰ using LR Clonase (Invitrogen), resulting in DOX-inducible mouse Cxcl1 or Cxcl2 transgene expression. Lentiviral particles were produced by co-transfecting pINDUCER-mCxcl1 or pINDUCER-mCxcl2 plasmid with helper plasmids pMD2G and psPAX into 293FT cells. Target cells were transduced in the presence of polybrene and selected by neomycin resistance.

Orthotopic animal experiments: for murine models, cell lines (5×10⁴ cells) were injected in 100 µL of phosphate-buffered saline into the #4 mammary fat pad of 6-week-old BALB/c or C57BL/6 female mice. Tumor formation and growth was followed for up to 100 days. Tumors were measured two times weekly with calipers, and volume was calculated in mm³ using the formula (width²×length/2). Mice were humanely euthanized when their tumor burden reached 2,000 mm³, at 100 days for survival studies, or at earlier time points for analysis. All animal studies were reviewed by the Vanderbilt University Medical Center (VUMC) Institutional Animal Care and Use Committee within the Animal Care and Use Program to monitor research integrity and compliance (protocol M2300055).

In vivo therapy regimens. Anti-programmed death-ligand 1 (PD-L1) (clone 6E11; Genentech) or anti-programmed cell death protein-1 (PD-1) (clone RMP1-14; Bio X Cell) was administered at an initial dose of 200 µg intraperitoneally when individual mouse tumors reached 100 mm³ and subsequently dropped to 100 µg weekly for up to 4 weeks of total administration time. Navarixin (SelleckChem) was administered daily via oral gavage at 200 µg/mouse until the end of the study.

Fine-needle aspiration (FNA): following previously published techniques,¹¹ FNA was performed on mouse tumors using a sterile, beveled 25G needle attached to a 10 mL syringe with a syringe holder (Belpro Medical). The needle was inserted into tumors in a saw-like motion collecting multiple representative areas of the tumor. The FNA-acquired cells were processed for flow cytometry analysis or RNA isolation for bulk RNA sequencing.

Flow cytometry: samples were run on an Attune NxT Acoustic Focusing Cytometer (Life Technologies). Analysis was performed in FlowJo. Gating was first done on forward scatter and side scatter to exclude debris. Doublets were excluded by gating on FSC area versus FSC height. Zombie Violet (BioLegend) was used to exclude dead cells from analyses. Antibodies: CD3- AF488 (BioLegend), CD8a- PE-CY7 (BioLegend), CD45- PerCP/Cy5.5 (BioLegend), NKp46- APC (BioLegend), CD11b- PE/Cy7 (BioLegend), CD11c- AF488 (BioLegend), IA/IE- AF700 (BioLegend M5/114.15.2), IFNG- PE (BioLegend), Ly-6G- PE (BioLegend), Ly-6C- APC-CY7 (BioLegend), CXCL1- APC (eBioscience), CXCL2- PE-CY7 (BioLegend). The gating strategy for flow cytometry plots is shown in online supplemental figure 5.

Immunohistochemistry and image analysis: formalin-fixed paraffin-tissue sections were cut at 4 µm and deparaffinized. Antigen retrieval was performed with citrate buffer pH 6 (Agilent, S2369). Endogenous peroxidases were inhibited with 3% hydrogen peroxide (Fisher Scientific, BP2633), and protein block solution (Agilent, K3468) was applied. Sections were then incubated with the primary antibody (CD8 4SM16, eBioscience, 14-0195-82 at 1:100) overnight at 4°C. Visualization was performed with the Envision system (Agilent Technologies) with DAB (Agilent, X0909) as the chromogen and hematoxylin (Fisher Scientific, 220–100) as the counterstain. Optimization of assay conditions was performed on murine lymph node and small bowel. Whole-slide images were digitally acquired using an AxioScan Z1 slide scanner (Carl Zeiss) at 20×. Automated quantification was performed by a pathologist using QuPath¹² software.

Tumor dissociation: EMT6 tumors were resected from mice (serum-free Dulbecco’s modified eagle medium (DMEM), Gibco) and dissociated using the Murine Tumor Dissociation Kit (Miltenyi Biotec) according to the manufacturer’s specifications with the gentleMACS Octo dissociator (Miltenyi Biotec) default tumor protocol (40 min at 37°C under constant agitation). The tumor cell dissociate was then passed through a 70 µm filter and red blood cells were lysed using ACK buffer.

Clinical samples: 83 samples from patients enrolled in the I-SPY-2 clinical trial¹³ were included in this study. Patients were treated weekly with paclitaxel for 12 weeks with or without four rounds of pembrolizumab administered every 3 weeks. Longitudinal peripheral blood samples were collected at baseline and after one treatment cycle (3 weeks of paclitaxel±pembrolizumab). At the time of blood collection, whole blood was centrifuged at room temperature at 1,100–1,300 g for 20 min and the buffy coat layer was stored at −80°C prior to RNA isolation and sequencing.

RNA isolation: RNA was harvested from cells or peripheral blood mononuclear cells (PBMC) using the LEV simplyRNA Tissue or Blood purification kit (Promega), respectively, using the Maxwell 16 automated workstation (Promega). RNA was analyzed for concentration by NanoDrop 2000 (Thermo Fisher).

Quantitative real-time PCR: complementary DNA (cDNA) from RNA was synthesized using the SensiFAST cDNA synthesis kit (Bioline) with 1 µg of RNA per sample. cDNA and advanced SYBR green universal supermix (Bio-Rad) were then combined with target specific primers on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Oligo sequences used for quantitative real-time polymerase chain reaction (qRT-PCR): Cxcl1 forward 5’-CCGCTCGCTTCTCTGTG-3’ reverse 5’- GCCTCGCGACCATTCTT −3’; Cxcl2 forward 5’- GTGAACTGCGCTGTCAATG-3’ reverse 5’- GCCTTGCCTTTGTTCAGTATC −3’; and Gapdh forward 5’- AGGTCGGTGTGAACGGATTTG −3’ reverse 5’- TGTAGACCATGTAGTTGAGGTCA-3’.

RNA sequencing and analysis: a quality control analysis was performed on the RNA samples using the Agilent Bioanalyzer to assess quality and an RNA Qubit assay was used to determine quantity. Library preparation was completed using the NEBNext rRNA Depletion kit and protocol (New England Biolabs) per manufacturer recommendations. The kit employs an RNase H-based method to deplete both cytoplasmic and mitochondrial rRNA. After ribosomal depletion, the RNA samples underwent messenger RNA enrichment using oligo(dT) beads, followed by thermal fragmentation using divalent cations under elevated temperature. First-strand cDNA synthesis was performed using random hexamer primers, while second-strand cDNA synthesis was carried out using DNA Polymerase I and RNase H. End repair, A-tailing, and adapter ligation were performed to generate the final cDNA library. The library quality was assessed using a Bioanalyzer and quantified using a quantitative PCR-based method with the KAPA Library Quantification Kit (Roche) and the QuantStudio 12K instrument.

Prepared libraries were pooled in equimolar ratios, and the resulting pool was subjected to cluster generation using the NovaSeq 6000 System, following the manufacturer’s protocols. 150 bp paired-end sequencing was performed on the NovaSeq 6000 platform targeting 50M reads per sample. Raw sequencing data (FASTQ files) obtained from the NovaSeq 6000 was subjected to quality control analysis, including read quality assessment. Real-Time Analysis Software and NovaSeq Control Software (V.1.8.0; Illumina) were used for base calling. MultiQC (V.1.7; Illumina) was used for data quality assessments. Reads were trimmed to remove adapter sequences using Cutadapt V.2.10¹⁴ and aligned to the Gencode GRCm38.p6 genome using STAR (V.2.7.8a).¹⁵ Gencode vM24 gene annotations were provided to STAR to improve the accuracy of mapping. Quality control on both raw reads and adaptor-trimmed reads was performed using FastQC (V.0.11.9) (www.bioinformatics.babraham.ac.uk/projects/fastqc), featureCounts (V.2.0.2)¹⁶ was used to count the number of mapped reads to each gene. Heatmap3¹⁷ was used for cluster analysis and visualization. Significantly differentially expressed genes with absolute fold change ≥2 and false discovery rate (FDR) adjusted p value≤0.05 were detected by DESeq2 (V.1.30.1).¹⁸ Genome Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway over-representation analysis was performed on differentially expressed genes using the WebGestaltR package.¹⁹ Gene Set Enrichment Analysis (GSEA) was performed using the GSEA package (V.4.2.3)²⁰ on database (V.7.5.1).

Statistical analysis: statistical analyses were performed in GraphPad Prism. In data with two groups, two-sample t-tests were used. For multiple analyses with >2 groups, significant differences were determined by analysis of variance with a Tukey’s post hoc test adjustment for multiple comparisons. For all multiple comparisons, statistical significance is noted by *p<0.05; **p<0.01, ***p<0.001, and ****p<0.0001. The cut-off for statistical significance was set to p≤0.05.

Results

Credentialing in vivo models of anti-PD-L1 response

Since ICI has been evaluated in combination with other treatment modalities, we sought to identify murine breast cancer cell lines that are responsive to anti-PD-L1 monotherapy, to uncouple the effects of other agents. We evaluated anti-PD-L1 efficacy in immunocompetent orthotopic mammary tumor syngeneic models: E0771-C57BL/6, 4T1-BALB/c, and EMT6-BALB/c (experimental schema figure 1A). The E0771 and 4T1 models showed a modest response to therapy, while in contrast, EMT6 tumor growth was substantially suppressed, including complete responses (figure 1B,C, and online supplemental figure 1D). Although tumor-specific PD-L1 expression is an element of the clinical biomarker(s) for patient ICI treatment eligibility,^{21 22} clinical trials in patients with early-stage TNBC showed favorable benefit irrespective of PD-L1 expression.²³ In line with these findings, when we implanted CRISPR-KO PD-L1-null EMT6 cells (silenced PD-L1 expression evaluated by flow cytometry online supplemental figure 1A), we found that anti-PD-L1 retained efficacy in suppressing primary tumor growth (online supplemental figure 1B). Similarly, blocking the PD-L1 receptor using anti-PD-1 likewise suppressed primary tumor growth (online supplemental figure 1C). We continued to observe mice to assess their survival following anti-PD-L1 administration and found that anti-PD-L1 monotherapy increases the survival of some mice compared with those that do not respond whose tumors progressed to endpoint (2,000 mm³). When responder mice were re-challenged with tumor cells in the opposing mammary fat pad, they failed to develop new primary tumors (data not shown) suggesting a robust memory immune response. Characterizing local immunological alterations, anti-PD-L1 demonstrated expected outcomes in the tumor microenvironment including increased CD8+T and dendritic cells, and reduced tumor-associated macrophages (figure 1D,E). Overall, the EMT6 mammary tumor model is sensitive to PD-1/PD-L1 blockade with similar correlates of response as seen in both mouse and human studies.24,26

Figure 1. Credentialing models of single-agent anti-PD-L1 response. (A) Schematic representing in vivo strategy; mice were orthotopically inoculated with murine mammary carcinoma cells and administered i.p. IgG antibody or anti-PD-L1 at an initial dose of 200 µg then subsequently 100 µg once weekly until tumors reached terminal end volume of 1–2,000 mm³. Graphs show tumor growth rate (B) and survival (C) of orthotopically transplanted E0771, 4T1, and EMT6 (p<0.0001) mammary tumor cells in C57BL/6 or BALB/c mice (n=10 mice per group), dotted lines represent treatment times. CR. Responder mice were re-challenged with a tumor cell injection (one mouse with 4T1 and six mice with EMT6) in the opposing mammary fat pad 12 weeks after initial tumor cell inoculation and clearance of tumor cells and failed to establish new tumors. Phenotypic assessment of EMT6 primary tumors by (D) immunohistochemistry (n=8–10) for CD8+T cells and (E) flow cytometry (n=4) reveals that anti-PD-L1 treatment increases activated CD8 T cells (CD45+, CD3+, CD8+, and GZMB+) and dendritic cells (CD45+, CD11c+) and reduces macrophages (CD45+, CD11b+, Ly-6C−). Data in B is depicted in mean±SEM and was analyzed by ANOVA and Tukey post hoc test and survival data in C was analyzed by log-rank (Mantel-Cox) test. Unpaired t-test used to compare treatments in D and E. Statistical significance is noted by *p<0.05; **p<0.01, ***p<0.001, and ****p<0.0001. ANOVA, analysis of variance; CR, complete responders, mice that have cleared their tumor burden; GZMB, granzyme-B; i.p., intraperitoneal; PD-L1, programmed death-ligand 1.

Response to anti-PD-L1 is heterogeneous in mammary tumor models

Consistent with clinical trial findings indicating that patients with TNBC exhibit variable responses to ICI²⁷ and in alignment with previously published in vivo data²⁸ we observed substantial heterogeneity in response to anti-PD-L1 despite the identical genetic background of the mice and polyclonal nature of the tumor cells implanted therein. The anti-PD-L1-treated EMT6 tumor-bearing mice exhibited three distinct response patterns: approximately 28% of tumors showed initial marked reduction in tumor volume followed by disease progression, 33% showed complete response (total elimination of tumor), and 39% demonstrated intrinsic resistance to therapy (no response whatsoever) (figure 2A). These phenotypes permitted us to uncouple tumor and patient heterogeneity observed clinically with other potential stochastic or biological effects in a tightly controlled experimental environment.

Figure 2. Response to PD-L1 blockade in EMT6 mammary tumors is heterogeneous and is mediated by host-intrinsic factors. (A) BALB/c mice (n=5 for IgG and 15 for anti-PD-L1) were orthotopically inoculated with parent EMT6 cells and administered intraperitoneal IgG antibody or anti-PD-L1 once weekly until tumors reached terminal end volume. Anti-PD-L1 induces differential responses in mice. Response patterns were stratified into three categories: responders, acquired resistance, and intrinsic resistance depending on tumor growth or elimination rates. (B, C) Tumors resistant to anti-PD-L1 treatment were excised, dissociated, and surgically implanted into the mammary fat pad of a new cohort of BALB/c mice (passage 1; n=10–15). Mice were treated with anti-PD-L1 and tumors resistant to anti-PD-L1 were dissociated and surgically re-implanted into new cohorts of mice (passage 2; n=10–15). Propagated EMT6 tumors retained response heterogeneity to anti-PD-L1. aPD-L1, anti-programmed death-ligand 1.

Interestingly, when individual anti-PD-L1 resistant EMT6 tumors were excised and re-implanted into new cohorts of mice, recipient mice maintained heterogeneous outcomes to anti-PD-L1, where some tumors exhibited sensitivity to PD-L1 blockade and others retained resistance, similar to de novo tumor implantation (figure 2B,C). These findings suggest that response and resistance to therapy can be dictated, at least in part, by host characteristics. Thus, we focused on identifying these factors as they may explain why some patients with breast cancer respond more favorably to ICI than others.

Co-administration of ICI with chemotherapy is the current standard-of-care practice in patients with newly diagnosed early-stage and locally advanced breast cancer. As such, there is a paucity of clinical data on the effect of primary ICI monotherapy in breast cancer, and thus we sought to evaluate whether the EMT6 model reflected clinical findings where combinations of ICI with chemotherapy incurred a therapeutic advantage in patients, both in the early and metastatic settings. We co-administered anti-PD-L1 in combination with two different classes of chemotherapy, paclitaxel or doxorubicin, and neither demonstrated benefit beyond the therapeutic efficacy of anti-PD-L1 alone (online supplemental figure 2A). When the course of treatment was extended to evaluate in vivo survival, neither paclitaxel nor doxorubicin combinations bestowed an added survival benefit in mice (online supplemental figure 2B). Phenotypic assessment of the tumor immune landscape revealed an activated immunological response characterized by reduction in tumor-associated macrophages and enhanced infiltration of natural killer and CD8+T cells (online supplemental figure 2C), which was likely due to the immune-activating function of anti-PD-L1, rather than the combination with chemotherapy.

Upregulated local interferon signaling is associated with anti-PD-L1 response

Due to the robust antitumor response elicited by anti-PD-L1 monotherapy, we focused on elucidating mechanisms underlying the differential response to ICI in the EMT6 model. To identify longitudinal changes in the tumor-immune milieu, we repeatedly sampled the primary tumor microenvironment prior to and following initiation of ICI by FNA and subsequent RNA sequencing (figure 3A). Within the tumor microenvironment, we found upregulated cytotoxic T-cell response and activation signatures (figure 3B). Performing GSEA on the tumors prior to systemic treatment with anti-PD-L1 revealed robust downregulation of oncogenic pathways, both at baseline and 7 days post anti-PD-L1, in responder-versus-intrinsically resistant mice (figure 3C). In untreated tumors, the majority of differentially expressed genes upregulated in mouse tumors that subsequently became complete responders were inflammatory interferon-related genes (figure 3D). The upregulated expression of these genes persisted after the initiation of therapy in responding tumors (figure 3E). Thus, there is a pre-existing priming in some tumor microenvironments leading to proclivity to ICI response, which becomes more obvious early in treatment. This phenomenon has also been observed in patients.29,31

Figure 3. Upregulated cytotoxic signature in the primary microenvironment of untreated tumors correlates with favorable response to ICI. Core-needle biopsies of the primary tumor microenvironment from heterogeneously responding mice treated with anti-PD-L1 (A) were queried by bulk RNA sequencing revealing signatures of enhanced baseline T-cell activation based on curated T-cell cytotoxicity genes (B), top differentially regulated signaling pathways as analyzed by GSEA (C), upregulated interferon signaling (D), interferon composite score calculated by sum all FPKM expression values of GSEA interferon pathway genes in each response category (E) in responder mice both at baseline and 7 days post anti-PD-L1 administration. Paired t-test used to compare across time points, regular t-test used for comparisons between intrinsic-resistant and responder mice. Statistical significance is noted by *p<0.05; **p<0.01, ***p<0.001, and ****p<0.0001. aPD-L1, anti-programmed death-ligand 1; FC, fold change; FPKM, Fragments Per Kilobase of transcript per Million mapped reads; GSEA, Gene Set Enrichment Analysis; ICI, immune checkpoint inhibition; IFN, interferon; NS, nonsignificant.

Peripheral myeloid signatures are enriched in vivo and in patients treated with chemoimmunotherapy who do not respond to checkpoint inhibition

Given the varied responses in our experimentally controlled murine model and the pronounced outcome differences when considering individual dynamic responses rather than baseline samples, we aimed to evaluate systemic peripheral indicators as a non-invasive alternative. Thus, we collected blood from mice via the submandibular vein, before and after tumor inoculation and after every round of treatment. RNA was isolated and sequenced from PBMCs (figure 4A). Although we initially hypothesized that there would be substantive baseline differences in host blood transcriptomic patterns contributing to resistance or response to anti-PD-L1 treatment, we found that the genetic profiles of both anti-PD-L1 resistant and responder mice were similar prior to tumor inoculation, with deviations only appearing after anti-PD-L1 administration (figure 4B). Immunologic composition was inferred from bulk RNA sequencing using CIBERSORT. Myeloid signatures were enriched in the peripheral blood of anti-PD-L1 resistant mammary tumor-bearing mice (online supplemental figure 3A and B). In particular, genes associated with myeloid cell homing and recruitment were transcriptionally upregulated in the peripheral blood of mice resistant to anti-PD-L1, specifically the receptors Cxcr1 and Cxcr2 (figure 4C,D). Chemokine receptors CXCR1 and CXCR2 bind to the CXCL family of chemokines, play a pivotal role in mediating immune cell trafficking during infections and other pathologic conditions, for example, cancer.^{32 33} This signaling axis is implicated in breast cancer progression through promoting tumor growth, metastasis, and immunotherapy response. In the evaluation of pertinent clinical specimens, we conducted an analysis of peripheral blood samples obtained from patients with TNBC enrolled in the I-SPY-2 trial, who were treated with either chemotherapy alone or in combination with pembrolizumab at baseline and at an early treatment time point (figure 4E). Corroborating findings from murine models, our data revealed a significant downregulation of peripheral myeloid gene expression in patients who achieved a pCR to pembrolizumab compared with those with residual disease (RD), and no differences were observed in the chemotherapy-only group (figure 4F). Inferring the myeloid fraction using CIBERSORT, patients with pCR responsive to pembrolizumab harbored the lowest abundance of monocytes within their peripheral blood post-therapy (figure 4G).

Figure 4. Peripheral blood signatures only cluster post anti-PD-L1 treatment in responder compared with intrinsic resistant mice. (A) Serial sampling of the peripheral blood from mice treated with anti-PD-L1 before (NT) and after tumor inoculation (T) and after every round of treatment (T1–T4) was performed to track changes during the experiments. RNA was isolated and bulk RNA sequencing was performed. (B) Mice based on response began depicting differential genetic profiles after administration of anti-PD-L1, specifically attenuated myeloid homing and recruitment transcriptomic profile based on curated genes of myeloid mobilization and recruitment (C). Example of the longitudinal changes in myeloid recruitment receptor Cxcr1 (D). (E) Summary of peripheral blood samples from I-SPY2 clinical trial where patients with TNBC were treated with paclitaxel±pembrolizumab. Paired analysis of patients with TNBC treated with paclitaxel±pembrolizumab with RD or pCR depicting (F) myeloid composite score based on the leading-edge genes of the human myeloid GSEA and (G) monocyte fraction as deconvoluted by CibersortX. Paired t-test used to compare across time point, regular t-test used for comparisons between pCR and RD patients. Statistical significance is noted by *p<0.05; **p<0.01, ***p<0.001, and ****p<0.0001. aPD-L1, anti-programmed death-ligand 1; FC, fold change; GSEA, Gene Set Enrichment Analysis; IFN, interferon; NS, not significant; pCR, pathologic complete response; RD, residual disease; TNBC, triple-negative breast cancer.

Combined myeloid blockade with anti-PD-L1 enhances response to immunotherapy in vivo

CXCL1/2 ligands (which bind CXCR1/2) promote neutrophil and myeloid-derived suppressor cell trafficking into the tumor microenvironment, creating an immunosuppressive milieu that fosters tumor progression.^{34 35} Thus, we combined myeloid cell blockade (CXCR1/2) with anti-PD-L1 to assess its utility in sensitizing tumors with intrinsic resistance to ICI. As such, we investigated the efficacy of the orally available small molecule inhibitor navarixin to block CXCR1/2 (online supplemental figure 4A and B) in efforts to further inhibit myeloid cell recruitment into the tumor with anti-PD-L1. We found that the combination synergistically increased in vivo response by blunting tumor growth (figure 5A) and enhanced survival (figure 5B). To further validate the involvement of CXCL1/2 in mediating resistance to anti-PD-L1 in mice, we employed the reverse strategy using a doxycycline-inducible vector to force expression of murine Cxcl1 and Cxcl2 in EMT6 tumor cells (figure 5C,D and online supplemental figure 4C). Inducing CXCL1 and CXCL2 production significantly increased Ly-6G+cells, thereby confirming neutrophil mobilization into the primary tumor site (online supplemental figure 4D). The enriched tumor myeloid cell presence due to induction of Cxcl1/2 was significantly associated with reduced complete response rates to anti-PD-L1 treatment (figure 5E,F). Survival analysis between the groups was not statistically significant, possibly reflecting the heterogeneity of the immune response in a rapidly growing model, making the primary differences observed not in time-to-endpoint, but instead on the proportion of mice demonstrating a complete response.

Figure 5. Targeting myeloid recruitment with CXCR1/2 inhibitors reduces tumor growth and synergistically extends survival in the EMT6 mammary tumor model in combination with anti-PD-L1. (A) Tumor growth and (B) survival in EMT6 tumor-bearing mice treated with anti-PD-L1 (i.p., weekly), a small molecule inhibitor navarixin 30 mg/kg (p.o, q.d.), or a combination of both. (C) qRT-PCR validation for murine Cxcl1 and Cxcl2 expression in EMT6 cells after doxycycline induction at 1 µg/mL for 48 hours. (D) Strategy to evaluate effect of Cxcl1/2 induced myeloid recruitment in vivo. Mice (n=15) were injected with pInducer-Cxcl1/2 EMT6 cells and given oral 5% sucrose solution with or without doxycycline (2 mg/mL), daily. As tumors reached 100 mm³, mice were treated with anti-PD-L1, and survival (E, F) was recorded; Fisher’s exact test p value=0.005. CR. Data in A is depicted in mean±SEM and was analyzed by ANOVA and Tukey post hoc test and survival data in B and E is analyzed by log-rank (Mantel-Cox) test. Statistical significance is noted by *p<0.05; **p<0.01, ***p<0.001, and ****p<0.0001. ANOVA, analysis of variance; aPD-L1, anti-programmed death-ligand 1; CR, complete responders; i.p, intraperitoneal; p.o, oral administration; q.d, once a day; qRT-PCR, quantitative real-time polymerase chain reaction.

Discussion

Immunotherapy, and particularly ICI (eg, targeting the PD-1/L1 axis), has become a mainstay of cancer therapy across a rapidly growing number of clinical indications. In 2020 and 2021, the anti-PD-1 monoclonal antibody pembrolizumab was approved for treatment of metastatic TNBC and early TNBC, in combination with chemotherapy.^{23 36} However, only a fraction of these patients achieves clinical benefit. In early-stage TNBC, where benefit can be directly assessed by pCR, phase III studies have reported a significant benefit of adding ICI in only ~15% of patients^{23 37} and <8% receive a durable recurrence-free survival.³⁸ Thus, an improved understanding of why ICI fails to benefit most patients with TNBC is needed to more appropriately target inherent mechanisms of resistance and to expand benefit, while limiting severe immune-related toxicities in patients who will not benefit.³⁹

It is unclear if the truly ICI-responsive early-stage TNBC population requires coadministration of chemotherapy, and provocative findings to the contrary have been demonstrated in several studies that used ICI monotherapy in this setting.^{29 40} Indeed, increasing data suggests a small population of patients may be capable of complete response to the absence of cytotoxic chemotherapy regimens. A recent report (BELLINI) in patients with early-stage TNBC treated with anti-PD-1 alone or co-administered with anti-cytotoxic T-lymphocyte associated protein 4 (CTLA-4) showed increased immune activation (higher CD8 T cells and increased interferon gamma) corresponding to complete pathological response,⁴⁰ but these patients represented a low-risk group with historically a >85% long-term survival without any systemic therapy whatsoever.⁴¹ Other detailed studies²⁹ in paired pre-anti-PD-1 and post-anti-PD-1 patients receiving neoadjuvant chemotherapy attributed response to ICI to the clonal expansion of cytotoxic CD8 T cells (expressing high granzyme B (GZMB) and perforin-1 (PRF1)) and Interferon gamma (IFNG)+effector CD4 T cells. Correspondingly, DCs and PD-L1/2^hi macrophages were enriched in ICI-responsive patients, suggesting a dynamically cooperative tumor microenvironment that facilitates tumor clearance on ICI treatment. Such findings underscore the importance of identifying and using in vivo preclinical models that accurately recapitulate patient outcomes to better understand response mechanisms and inform patient care.

Findings from our study indicate that blocking the PD-1/PD-L1 axis alone immunologically activates murine tumors sensitive to anti-PD-L1, without significant synergistic benefit from chemotherapy modalities, leading to complete responses with memory/recall. The data generated herein provides rationale for potentially de-escalating therapeutic regimens to reduce risk of developing toxicities in patients with breast cancer. Additionally, our study identifies elevated baseline local inflammatory signaling in anti-PD-L1 responsive mice, which has been observed in a variety of studies using baseline tissue samples in patients.42,45

Another deterrent for ICI success in breast cancer is the lack of reliable biomarkers for predicting patients that will benefit from ICI,⁴⁶ especially since tumor-specific PD-L1 expression has proven to be largely ineffective.⁴⁷ Indeed, most studies have shown that patients who benefit do not have higher tumor-infiltrating immune cells, tumor mutation burden or PD-L1 expression.^{23 37 48} Thus, these markers have proven to be of limited clinical utility in breast cancer.

The detailed analyses of the range of ICI responses observed within isogenic mammary cancer models performed in our study are pivotally informative for understanding heterogeneity in outcomes, as well as local and systemic changes that occur with ICI administration. Our study sheds light on the dynamic transcriptomic alterations pretreatment and post-treatment that may influence patient response to ICI, particularly those occurring in peripheral blood, a largely unexplored resource for ICI biomarkers. While our study identifies local inflammatory signaling as robust markers of responsiveness to anti-PD-L1, even at baseline, peripheral blood-based biomarkers could be a potentially useful, minimally invasive approach for monitoring the dynamic patient-specific changes occurring in response to ICI in order to optimize and tailor neoadjuvant therapy.

Tumor-specific expression of CXCR1/2 is a reported negative prognostic marker for relapse in patient with breast cancer.⁴⁹ Data from murine peripheral blood reported herein supports this observation; anti-PD-L1 resistant mice harbored elevated CXCR1/2+myeloid cells after treatment. Further characterization of patient with TNBC peripheral blood in the I-SPY-2 trial responding to pembrolizumab supports the role of circulating myeloid cells in modulating resistance to ICI. Therefore, sampling myeloid abundance in PBMC, specifically markers of recruitment like CXCR1/2, may provide a less invasive peripheral biomarker for predicting ICI response in patients. Furthermore, therapeutic targeting of enhanced myeloid cell trafficking in both murine models and human subjects has the potential to augment responsiveness to ICIs by fostering a tumor-suppressive microenvironment conducive to malignant cell clearance. Although this approach has been explored in a limited number of clinical trials involving NSCLC, prostate cancer, and colorectal cancer,⁵⁰ the resulting combination therapies demonstrated modest clinical benefit and were frequently associated with treatment-related adverse events. As such, limited data are currently available regarding the efficacy of navarixin-based combination therapies in breast cancer. Furthermore, the timing of therapeutic intervention may critically influence treatment outcomes, underscoring the importance of delineating an optimal window for inhibiting myeloid cell recruitment as opposed to implementing prolonged blockade strategies.

The longitudinal sampling of tumor tissues and peripheral blood presented in this study enables us to identify therapy-specific alterations on an individual level. This approach is particularly useful due to immunological adaptations during tumor progression and treatment. Additionally, this study is limited to using a single responsive tumor model (EMT6 to PD-L1 blockade), which may constrain the generalizability of the findings. Moreover, the methodologies employed do not allow for precise resolution of the cellular sources contributing to cytokine signaling and the exclusion of specific immunophenotyping markers further limits the ability to delineate key immune subsets involved in the observed responses. In our murine model and patient cohort, we only evaluated the impact of blocking the PD-1/PD-L1 axis. Consequently, future assessments are necessary to evaluate additional longitudinal changes to uncover signatures linked to other ICI modalities (eg, CTLA-4 blockade) and to characterize potential markers of response and resistance.

Overall, the information generated in this study provides dynamic, mechanistic insights into resistance mechanisms for immunotherapy, and supports the rationale for combinatorial immunotherapy approaches with potential application to various cancer types. This knowledge can be used to potentially predict ICI-resistant patients and instruct therapies to mitigate resistance, thus improving rates of their progression-free survival.

Data availability statement

Data are available upon reasonable request.