Engineered immunologic niche monitors checkpoint blockade response and probes mechanisms of resistance

Antibodies to programmed cell death protein 1 (anti‐PD‐1) have become a promising immunotherapy for triple negative breast cancer (TNBC), blocking PD‐L1 signaling from pro‐tumor cells through T cell PD‐1 receptor binding. Nevertheless, only 10%–20% of PD‐L1+ metastatic TNBC patients who meet criteria benefit from immune checkpoint blockade (ICB), and biomarkers to predict patient response have been elusive. We have previously developed an immunological niche, consisting of a microporous implant in the subcutaneous space, that supports tissue formation whose immune composition is consistent with that within vital organs. Herein, we investigated dynamic gene expression within this immunological niche to provide biomarkers of response to anti‐PD‐1. In a 4T1 model of metastatic TNBC, we observed sensitivity and resistance to anti‐PD‐1 based on primary tumor growth and survival. The niche was biopsied before, during, and after anti‐PD‐1 therapy, and analyzed for cell types and gene expression indicative of treatment refractivity. Myeloid cell‐to‐lymphocyte ratios were altered between ICB‐sensitivity and resistance. Longitudinal analysis of gene expression implicated dynamic myeloid cell function that stratified sensitivity from resistance. A niche‐derived gene signature predicted sensitivity or resistance prior to therapy. Analysis of the niche to monitor immunotherapy response presents a new opportunity to personalize care and investigate mechanisms underlying treatment resistance.


INTRODUCTION
Immunotherapy has emerged as a promising treatment for cancer, 1 with more than 60% of these therapeutic approaches targeting T cells by checkpoint inhibition. 2 T cells are powerful mediators of anti-tumor immune surveillance through their ability to detect and eliminate cancerous cells. 3,44][15][16] Nevertheless, only 10%-20% of PD-L1 + metastatic TNBC patients who meet selection criteria benefit from ICB. 9,17 Biomarkers to stratify patients by likelihood of response are needed for effective treatment outcomes. 18,19The most widely used biomarker for stratifying sensitivity to anti-PD-1 is PD-L1 expression on tumor infiltrating lymphocytes or the tumor stroma, yet surprisingly, some PD-L1 -patients are sensitive to ICB, demonstrating the limitations of PD-L1 alone to stratify patients. 20,21Tumor mutational burden (TMB) has also been utilized as an ICB biomarker in solid tumors, 22 yet the majority of even these biomarker-positive patients are ICB-resistant. 23As such, TMB remains controversial for identifying benefit from immunotherapy in TNBC where mutational load is relatively low. 9,24Additionally, tumor-derived gene signatures, such as Oncotype DX or FoundationOne, have been pursued for predicting ICB response, but similarly fail to effectively stratify patients with TNBC. 25,26PD-L1 status, TMB, and tumor-derived gene signatures do not accurately predict a clinical response to ICB, possibly because they are derived from the primary tumor (PT), which reflects the local environment at the time of biopsy yet provides insufficient information regarding the systemic immune response or the environment within distant tissues that may be sites for recurrence, often after an extended period.A single tumor biopsy may not adequately reflect the heterogeneity of PD-L1 and other biomarker expression, 27 and changes in PD-L1 tumor expression may occur before or under therapy with PD-1/PD-L1 inhibitors, leading to differential sensitivity missed by a snapshot biopsy. 284][35][36][37][38] The biopsy of metastatic sites has been proposed as having the appropriate characteristics, yet the biopsy of a vital organ, such as the lung, has associated risks and cannot be routinely performed for biomarker assessment.We have previously reported an implant that supports the formation of a surrogate tissue in the subcutaneous space, which recruits tumor cells, [39][40][41] and the immune cells within this engineered tissue are characteristic of those within an inflamed vital organ such as the metastatic lung. 42A microporous scaffold is implanted subcutaneously, with the porous structure supporting persistent intravasation of host cells and vascularization.
Immune cells within the vasculature are recruited to this newly formed tissue environment due to the foreign body response, and thus the local microenvironment is dynamically modified through disease initiation and progression as an immunologic niche (IN). 40, 43 In a murine model of TNBC, the longitudinal analysis of gene expression effectively monitored disease progression and indicated responsiveness to a primary tumor resection. 39,44 this work, we investigated the dynamic gene expression within the IN to provide biomarkers that correlate with the response to anti-PD-1 in TNBC.We employed the 4T1 model of metastatic TNBC, which was treated with anti-PD-1 and resulted in cohorts that were either

Divergent disease progression and survival in response to immune checkpoint blockade therapy
We initially investigated the divergent responses to ICB in a murine model of advanced TNBC.BALB/c mice were orthotopically inoculated (day 0, D0) with triple negative 4T1 tumor cells and administered anti-PD-1 or isotype control intraperitoneally starting on day 9 post-tumor cell inoculation, for a total of four doses (D9, D11, D13, D15) (Figure 1A).No difference was found between the isotype control cohort and the full anti-PD-1 cohort (Figure S1A); however, the anti-PD-1 cohort could be stratified into sensitive and resistant groups based on the fold change in PT volume on D19 post-inoculation compared to the D7 baseline and the isotype-treated control group.Indeed, ICB-sensitive mice were defined as those which had significantly reduced PT growth compared to the ICB-resistant mice and isotype control on day 11 post-inoculation and later times (Figure 1B), including tumor volume fold change ≤2.5 on day 19 from day 7. ICB-resistant mice had indistinguishable progression of disease from the isotype control (Figure 1B) in the first 19 days post-inoculation, while ICB-sensitive mice had significantly improved survival compared to the ICB-resistant cohort (Figure 1C).Prior to treatment, all tumors were PD-L1 positive (Figure 1D, G), and PD-L1 positive cell proportions remained at similar levels or higher following treatment (Figure 1G).While PD-L1 expression in the tumor 1 week after ICB treatment was completed is significantly different between sensitive and resistant mice (Figure 1G, p = 0.0048), PD-L1 expression in the primary tumor before treatment could not distinguish between ICB-sensitive or resistant mice after treatment (Figure 1E-G).

Ratio of innate to adaptive cell populations is skewed at the IN in ICB-sensitivity
Microporous poly(ε-caprolactone) (PCL) scaffolds (5-mm diameter, 2mm height) were implanted into the dorsal subcutaneous space 14 days prior to tumor cell inoculation (Figure 2A).These implants did not impact the primary tumor growth (Figure S1B), consistent with previous studies. 40 S4).We have previously shown that the IN implant contains disease relevant cell recruitment 40,44 and accumulates predominantly innate leukocytes, with vasculature, stromal cells, and a smaller population of lymphocytes and cancer cells. 40,45Notably, the IN had fewer leukocytes (CD45 + ) after therapy in sensitive mice (Figure S3A), while also showing enrichment of both myeloid cells (monocytes and macrophages) (Figure 2B, Figure S3B-D) and lymphocytes (T cells and NK cells) (Figure 2C, Figure S4), as compared to the PT.Also, the PT from ICB-sensitive mice had significantly increased infiltration of CD8 T cells (CD3 + CD8 + ) compared to ICB-resistant PT (Figure 2C, Figure S4A).No significant difference was observed between the proportion of monocytes (CD11b + Ly-6C + Ly-6G − ), neutrophils (CD11b + Ly-6G + Ly-6C − ), and macrophages (CD11b + F4/80 + ) at the PT of ICB-sensitive and ICB-resistant mice (Figure 2B), however.
Similarly, no difference in the proportions of monocytes, neutrophils, and macrophages was observed at the IN based on ICB response, while reduced proportions of both B cells (CD19 + ) and NK cells (CD49b + ) were observed at the IN of ICB-sensitive mice (Figure 2C), though not cell quantities (Figure S4C, D).As B cells and NK cells are influenced by myeloid-derived immune cells in ICB, 46,47 the ratios of innate to adaptive immune cells were also assessed, which has been a useful prognostic indicator in other reports. 48,49Interestingly, the macrophage: NK cell, neutrophil: NK cell, neutrophil: B cell, and neutrophil: T cell ratios were all significantly increased at the IN of ICB-sensitive mice relative to ICB-resistant mice (Figure 2D, E).Collectively, these results indicate the immunological niche captures a high density of both innate and adaptive cells for analysis, yet few differences are observed between sensitive and resistant mice, suggesting that broad immune population dynamics alone cannot distinguish ICB treatment refractivity.

Gene expression and immune pathways are differentially regulated after therapy between ICB sensitivity and resistance
Under the same treatment ICB schema as above, we next performed bulk RNA sequencing (RNA-seq) of the IN to investigate the potential for longitudinally monitoring ICB response, as changes in gene expression observed at the IN corresponding with autoimmune activity are likely due to alterations in cell phenotypes within the IN microenvironment. 44The mice were similarly stratified following ICB treatment with IN implants as ICB-sensitive or ICB-resistant (Figure S5A, B).We initially focused on the determination of a response to therapy, and thus D21 IN gene expression was analyzed for the genes and pathways that underlie ICB sensitivity and resistance shortly after treatment.Gene set enrichment analysis (GSEA) identified 143 differentially regulated pathways between IN implants from ICB-resistant and ICB-sensitive mice.The pathways involve many leukocyte functions including proliferation, migration, differentiation, chemotaxis, inflammation, and cytotoxicity (Figure S6A).Although ICB has been viewed as primarily targeting T cells, checkpoint blockade efficacy and resistance are also associated with myeloid cells through direct and indirect mechanisms. 50,51GSEA indicates multiple myeloid cell types are impacted including neutrophils, monocytes, and macrophages (Figure 3B).Myeloid cell phenotypes have been shown to accumulate in immune niches and develop immunosuppressive activity induced by systemic immune alterations from the solid tumor. 52 identified by fold change in expression between ICB-resistant and sensitive mice (Figure S7A) differentiate between IN biopsies from these groups (Figure S7B, C).This sparse pathway set that is most strongly correlated with ICB refractivity includes regulation of the well-known oncogene Myc, which has been associated with metastatic breast cancer aggressiveness 53 ; again IFNγ and IFNα signaling, which are also influenced by Myc and its impact on myeloid cells 54 ; PI3K-AKT network signaling, dysregulation of which is related to therapy resistance 55 ; and retinoic acid signaling, which is implicated in reprogramming suppressive macrophages and dendritic cells toward greater anti-tumor activity. 56,57Collectively, these pathway analyses suggest that myeloid cell phenotype and function con-tribute to many of the differences in ICB response assessed at the IN implant.
A total of 243 differentially expressed genes (DEGs) were then identified between the ICB-sensitive and ICB-resistant cohorts (Table S1).response after treatment (Figure 3D, E, Table S2).Thirteen of these 21 genes were upregulated in the ICB-sensitive mice, whereas nine were upregulated in the ICB-resistant mice.Ripply3, Fdps, Pagr1a, and Stfa2 account for the most variable importance of the panel in distinguishing ICB response (Figure S9A).Ripply3 has been associated with colon cancer through methylation-regulated control of cell proliferation. 58ps is known to be highly expressed in several cancers and plays an oncogenic role through the GTPase/AKT axis. 59Mouse Pagr1a modulates the expression of BMP2, 60 which is known to promote breast cancer cell invasion. 61Lastly, Stfa2 has been described as an anti-cancer gene and promising candidate marker for carcinoma of the head and neck 62 and laryngeal cancer, 63 and our group previously found that Stfa2 steadily increases at the metastatic lung throughout disease progression. 64This signature of ICB response after therapy indicates that the IN recapitulates aspects of disease.A permutation test of the sensitive and resistant profiles confirmed that these groups differ from one another according to the variance of this gene signature (Figure S9B) and partial least squares discriminant analysis (PLS-DA) cross-validation error (Figure S9C).Together, these results indicate that gene expression at the IN implant immediately after completion of ICB therapy could monitor treatment response in both pathway regulation and gene expression.

Principal component analysis (PCA) demonstrates that the expression
We next investigated whether the 21 gene signature could identify sensitivity or resistance mid-therapy.The gene signature could not distinguish sensitivity from resistance at D14 (Figure S10).The analysis of a signature at individual time points selects for the biggest differences at that time and does not capture the inherent dynamics and may be more susceptible to heterogeneity among individuals.This result motivated an alternative biomarker derivation approach.

Serial analysis of IN gene expression captures dynamics of ICB resistance
A "serial" analysis of gene expression for each mouse was obtained by normalizing expression at a particular time point to the baseline (D7) by taking the difference in expression for each gene (Figure 4A).As samples are collected, they can be analyzed relative to the initial time and to previous time points to identify differences.We initially analyzed the differences between D14 and D7, and EN regression with 2000 iterations of cross validation identified a panel of 22 genes (Figure 4B, Table S2).We implemented a more rigorous scoring system involving an unsupervised clustering of samples through singular value decomposition (SVD), and a second supervised method using the bagged-tree Random Forest™ (RF) machine learning algorithm (Figure 4C, D).The SVD and RF each assign a score to the gene panel to assess the ability to distinguish sensitivity and resistance.These scoring metrics separate the two groups well, and the receiver operating characteristic (ROC) curve sensitivity and specificity for identifying ICB response with this serial gene panel indicated only a 10% error rate (Figure S11A).The SVD and RF algorithm exhibited strong separation of the ICB-response cohorts at D21 (Figure 4D), with the ROC curve being error-free (Figure S11B, C).Of these 22 genes, 14 increased in expression from baseline (D7) to after treatment (D21) among the ICB-sensitive cohort, whereas the remaining eight genes decreased (Figure 4E).Differences between conditions were observed between each time point; how-ever, the greatest differences emerged between days 7 and 21 (Figure S11B, C).Among the panel, Crlf1 and, to a lesser extent, Gm18362, Sprtn, and Ptgs2 have the greatest variable importance in separating the cohorts (Figure S11D).Crlf1 has been shown to increase cell migration, invasion, and tumor growth by activating the ERK1/2 and AKT pathways and may be a potential therapeutic target. 65While little is known regarding Gm18362, mutations in Sprtn, an essential protease in DNA damage repair, 66 are known to cause liver cancer in mice and humans. 67Ptgs2 plays a key role in promoting cancer growth and metastasis, 68 and knockdown by siRNA inhibits MDA-MB-231 breast cancer cell proliferation. 69Collectively, longitudinal IN-derived gene expression provides the ability to distinguish sensitivity and resistance both during and after ICB treatment.
The myeloid cell enrichment and associated pathway enrichment identified at the IN after ICB motivated an analysis of the contribution of myeloid cells in driving the differential gene expression between ICB-resistance and sensitivity at the IN.The expression profiles from myeloid (CD11b+) and non-myeloid (CD11b−) cell fractions were isolated from ICB-sensitive and ICB-resistant cohorts (Figure S12A) and applied to associate a given transcript with myeloid cells (Equations 1, 2), positive myeloid score, or non-myeloid cells, negative myeloid score (Figure S12B).For the 22 predictor genes comprising the niche-derived serial panel, the genes in the panel that are most divergent between sensitive and resistant are associated with myeloid cells (Figure 4F).Genes strongly associated with myeloid cells are closely aligned with the vertical axis, 11 genes total, whereas only two genes are weakly associated with non-myeloid cells, along the horizontal axis.

Gene expression and pathways at the IN before therapy predicts sensitivity to checkpoint blockade
We subsequently investigated the most critical challenge in ICB therapy, identifying susceptibility to checkpoint inhibition prior to therapy.Analysis of the IN at D7 (prior to anti-PD-1) identified 322 DEGs between ICB-sensitive and ICB-resistant cohorts (Table S1), with differential clustering of sensitive and resistant mice (Figure S13A, B).The pathways associated with this differential expression between sensitive and resistant response prior to therapy were then investigated, identifying 207 immune-related pathways highly differentially regulated by ICB response.These immune pathways were again associated with leukocyte proliferation, migration, differentiation, chemotaxis, inflammation, and cytotoxicity (Figure S14A).The most differentially enriched adaptive pathways for ICB response before treatment were associated with T cells (differentiation, activation, proliferation, Th1/Th17 response, aberrant function), B cell activation, NK cell function, and adaptive immune cell responses (Figure 5A).Pathways enriched at the IN in ICB-sensitive mice notably include modulators of T cell receptor signaling and somatic diversification of immunoglobulins, including IFN and TNF signaling (Figure S14B).Pathways within myeloid cells, including neutrophils, macrophages, mast cells, and monocytes, were also differentially regulated (Figure 5B), with myeloid cell homeostasis, myeloid cell differentiation, and neutrophil-mediated immunity enriched within the IN before ICB.Conversely, macrophage tolerance and macrophage M2 pathways were depleted at the IN before treatment for ICBsensitive mice.Pathways were then converted to GSVA scores and then selected for those that were most strongly correlated with ICB refractivity using an EN regression to specify the fewest pathways needed for differentiation (Figure S15A, B).These differentiating pathways include CD28 co-stimulation, the TLR9 cascade, TP53 regulation, and P2Y purinoceptor 1 ADP signaling (Figure S15C).[72][73] We next employed a bootstrapping aggregation method in which we partitioned the IN samples into a training dataset (70%) and validation dataset (30%).One-hundred-fold bagging iterations of the training dataset with an EN regression identified a sparse 15-gene panel which produced a clear categorization of the response groups prior to the initiation of ICB when tested with the validation dataset (Figure 5C, D, Table S2), 12 of which are upregulated at the IN in sensitive mice.
This gene panel has high sensitivity and specificity in predicting ICB response (Figure S16A, B).Slc2a12, Slc5a7, Aldh1a2, and Id3 show the greatest variable importance among this predictive panel (Figure S16B, C).Slc2a12 is known to be upregulated in exosomes derived from patients with gastric cancer and has been correlated with tumor size, stage, lymph node metastasis, and degree of differentiation. 74c5a7 has shown potential as a biomarker in personalizing therapy for lung adenocarcinoma and lung squamous cell carcinoma. 757][78][79][80] Id3 has been noted to govern colon cancer-initiating cell self-renewal through cell-cycle restriction driven by the cell-cycle inhibitor p21, 81 and suppression reduces proliferation rate, invasiveness, and anchorage-independent growth. 82These results suggest the potential for analysis of the IN to identify sensitivity or resistance to ICB prior to the initiation of treatment.
After validating the 15-gene IN signature of ICB sensitivity before treatment, we further compared the signature to potential biomarkers identified in the field (Figure 5E).These comparison transcripts include the murine orthologs of the Oncotype DX panel, orthologs of transcripts identified as prognostic T cell markers of TNBC immune modulation and myeloid activity, 83 and orthologs identified as prognostic of TNBC survival based on PD-1 expression and tumor immune infiltration. 84The IN signature outperforms each of these tumor-based gene panels (via sensitivity and specificity) in identifying ICB sensitive and resistant mice.Additionally, the 15-gene panel from the IN derived by partitioning and rigorous cross-validation outperforms a distinct 19-gene panel of pre-treatment ICB sensitivity at the IN without partitioning samples for independent validation (Figure 5E).
Cross-validation error and test error of a PLS-DA model of this signature were both zero across 1000 sampling permutations, compared to 0.296 cross-validation error and 0.252 test error means from the lowest-error gene panel of 100,000 random gene panel permutations of up to 15 genes from the 322 DEG set (Figure 5f, Figure S16D,   E).These sensitivity and specificity results and distributions collectively indicate that we have derived a gene signature of ICB sensitivity through the IN which provides unique information from biomarkers derived at the primary tumor.

DISCUSSION
Checkpoint inhibitors have been approved for early, locally advanced, and metastatic TNBC, and while profound responses have been observed in ICB-sensitive patients, many are resistant or develop resistance to ICB.Additionally, patients on ICB can suffer significant unintended consequences, such as development of autoimmunity. 85is combination of resistance and unintended consequences motivates the development of biomarkers to identify patients that will respond to therapy, and to monitor the response to ICB.ICB may be administered in the neoadjuvant setting, which is analogous to the studies herein, and in the adjuvant setting after PT resection.The primary biomarker employed currently is expression of PD-L1 in a PT biopsy or at surgical resection, which provides only a snapshot of the local PT microenvironment, and biomarker expression at the PT is often discordant with biomarker expression at distant sites. 86This provides significant complications when making clinical decisions based on PT-derived biomarkers, especially when treatments for metastatic breast cancers are motivated by a PT biopsy that may have been collected months or years prior. 87The inability to monitor response or predict responsiveness prior to treatment highlights the clinical need for a technology to profile the immune response to ICB and was the basis for investigating the IN implant as a vascularized lymphatic structure dynamic with the status of the immune system and prognostic for ICB response.
9][90] We have demonstrated that IN-derived gene expression can be probed to monitor ICB response.Measuring changes in size of a tumor, often by radiographic imaging, is the clinical gold standard for monitoring response.
For most chemo-and radiotherapies, change in lesion size indicates response to treatment.Pseudo-progression and hyper-progression following ICB, however, uniquely confound the ability to correlate lesion size with immunotherapy response. 91Pseudo-progression is characterized by the radiographic appearance of lesion growth, as a result of immune cell infiltration into a tumor, in ICB-sensitive patients; this initial growth is followed by tumor regression as a result of an anti-tumor immune response. 91A small proportion of patients experience hyperprogression, or a rapid progression of disease following the initiation of ICB. 91When lesion enlargement is observed, radiographic imaging alone cannot stratify patients experiencing pseudo-progression or hyper-progression during administration of therapy.4][35][36][37][38] This inability to monitor ICB efficacy may result, in part, from differences in the phenotype and function of immune cells in the peripheral blood relative to those within a tissue. 29The IN provides a platform to analyze immune cells that have exited systemic vasculature to a tissue-like environment, which can be longitudinally biopsied without significant risk to a vital organ.Though previous work by our group has identified that cancer cells also accumulate at the IN, [39][40][41] the characterization of the scaffold gene expression is predominantly associated with innate immune cells.Indeed, myeloid cell subtypes express PD-1 and/or PD-L1 and are direct targets of checkpoint blockade.PD-1 inhibition on myeloid cells has been shown to affect infiltration, activation, and metabolism.ICB has previously been found to have an equal, or greater, gene expression impact on the phenotype of tumor associated macrophages than T cells. 92Similarly, dendritic cell activity within the tumor microenvironment is also regulated by PD-1, 93 and improving APC-T cell crosstalk through NF-κB signaling may help convert cold immunological niches to checkpoint responsive. 51We identified multivariate gene signa-  52 The phase 2 TONIC trial of nivolumab (anti-PD-1) in metastatic TNBC patients investigated induction strategies for modulating the tumor microenvironment prior to ICB. 97 Interestingly, that study found the highest overall response rate among the doxorubicin (35%) and cisplatin (23%) induction cohorts and detected an upregulation of genes associated with PD-1/PD-L1, Th1 cell, myeloid cell, T cell cytotoxicity, and IFNg pathways, as a result of pre-ICB induction.In pre-clinical studies, doxorubicin and cisplatin are immunomodulatory, in part through their targeting of suppressive myeloid cells. 98,99The IN implant collects enriched populations of monocytes, macrophages, T cells, B cells, and NK cells versus both the PT and spleen, which functions as a surrogate for the blood, likely due to the foreign body response of the IN.
Isolating tissue-specific immune phenotypes has been an immense limitation in the clinical utility of liquid biopsies as a cancer immunotherapy diagnostic. 29Probing the gene expression at the IN identified differentially regulated myeloid cell pathways in ICB-sensitivity versus resistance, suggesting that aberrant myeloid cell function may contribute to ICB-resistance.Additionally, the cell populations at the implant had significantly different ratios of myeloid cells to lymphocytes between ICB-sensitive and ICB-resistant mice.Together with the findings from the TONIC trial, these studies illuminate the potential role of immunosuppressive myeloid cells in ICB-resistance, as well as the efficacy of improving ICB-sensitivity by alleviating this suppression.In fact, a study investigating immunomodulatory nanoparticles that target myeloid cells, found synergistic efficacy when delivering both ICB and nanoparticle-based myeloid cell immunomodulation together in a murine model of TNBC. 100 The IN could serve as a diagnostic for monitoring suppressive mechanisms underlying ICBresistance and to motivate the use of induction strategies to skew ICB-resistance toward sensitivity.The landmark first-in-human clinical trial, QUILT-3.067, is an ongoing study with the goal of evaluating the safety and efficacy of combining multimodal induction therapies, targeting multiple arms of the immune system, with ICB to improve ICB-sensitivity. 101This unique trial combines chemoradiation, IL-15 cytokine administration, tumor-associated antigen vaccine, high-affinity NK cell therapy, and avelumab in patients with refractory, metastatic, or unresectable TNBC tumors.Initial results from the small cohort of patients (n = 9) suggest improved overall response, disease control response, and complete response rates, as compared to ICB monotherapy.This trial points to the mounting evidence that myeloid and NK cells, among other immune cells, play a role in suppressing anti-tumor immune cell responses.Interestingly, in this report, we found that pathways for myeloid and NK cell function were differentially regulated between the ICB-sensitive and resistant cohorts.
Both the phase 2 induction and QUILT-3.067trials highlight the value for a diagnostic that could provide insight into targetable immunosuppressive mechanisms underlying ICB-resistance.While this present study centered on myeloid cell-associated gene expression at the scaffold due to differential pathway regulation, specific features from both myeloid-derived and lymphoid cells have been identified in TNBC tumor microenvironments. 102NK cells were found to be differentially enriched at the IN in ICB-resistant mice after therapy (Figure 2C) and could also be assessed further for diagnostic potential and mechanistic understanding in ICB refractivity. 103 conclusion, we report that the IN implant can be probed to characterize divergent responses to checkpoint blockade in a TNBC murine model.IN-derived gene expression was computationally investigated for monitoring ICB response during and after therapy.Probing the implant identified divergent immune pathways and altered myeloid cell-to-lymphocyte ratios between ICB-sensitivity and resistance, implicating aberrant myeloid cell function in ICB-resistance.This result is noteworthy, given that ICB could be combined with a targeted myeloid cell therapy to improve response rates. 104

Tumor volume measurements and survival monitoring
Tumor size was recorded using electronic calipers (VWR) while mice were anesthetized.Primary tumor volume was calculated (V = 0.5 × L × W 2 , L: length of longest dimension of the tumor, W: length perpendicular to the longest tumor dimension) as previously described. 39Mice were monitored for tumor size and body condition to determine survival.Mice were euthanized if any of the following criteria were met: tumor size of >2 cm in any dimension, ulceration of >50% of the visible tumor, partial paralysis due to tumor invasion of hind limb, labored breathing, ascites, lethargy, or visible weight loss.

Niche implant and tissue isolations
Niche implants were surgically explanted at days 7, 14, and 21 (D7, D14, D21) post-tumor cell inoculation to study gene expression changes associated with ICB response.Mice were anesthetized, and an incision was made above the implant.The implant and any adherent encapsulating tissue were excised, and the incision was sutured closed.
For RNA analyses, the implants were then flash-frozen in isopentane on dry ice and stored at −80

Differentially expressed genes and differentially regulated pathways
Normalized RNA-seq counts were screened to identify DEGs by response to ICB.T-tests were first performed for each gene between ICB-sensitive and ICB-resistant mouse implants.These screened genes were then probed with a two-step process using EN regularization with random resampling to determine gene signatures associated with variable resistance phenotypes.First, data were randomly sampled without replacement to generate 1000 subsets.The resampled subsets spanned 90% of the original sample size to reduce potential outliers in feature selection.EN regularization was then applied to each of the 1000 resampled subsets to select features most associated with the outcome variables.To facilitate mechanistic insight following feature selection, the EN hyperparameter, α, was set with equal weights between the least absolute shrinkage and selection and ridge regression penalties (α = 0.5) to promote sparsity and group selection. 106e final signature was selected to include genes which were chosen in >85% of all EN iterations.For all EN models, k-fold cross-validation (k = 5) was used to generate the model with the lowest misclassification error.Multivariate gene signatures were visualized with PCA for dimensionality reduction.ROC curves were generated with sensitivity and specificity for identifying ICB response for the gene panels identified by EN.Also, a two-metric scoring system was derived as a clinical means of assigning a simple numeric and visual representation of the serial gene panel in identifying ICB response for a given IN sample.One metric is an unsupervised clustering of samples through SVD, while the second method is a supervised bagged-tree Random Forest (RF) machine learning algorithm.All computations were performed using R except for the EN-based analysis, which used the Glmnet package in MATLAB. 107GSEA was performed on DESeq2-normalized counts with the human orthologs to assess pathways associated with divergent ICB responses from 4726 identified gene sets (Hallmark, Reactome, PID, and Gene Ontology). 108We set a normalized enrichment score (NES) cutoff of |NES| > 1 to find the most differentially regulated pathways and categorized the pathways by immune response, excluding irrelevant inflammatory pathways (e.g., miRNAs Involvement in the Immune Response in Sepsis).GSVA was also performed to identify sparse pathway lists classifying ACAR distinctly from healthy graft recipients. 109r pathway analysis, mouse gene symbols were converted to human gene orthologs using the biomaRt package.

Cross-validation via sample permutations and PLS-DA
A variance validation was performed for each signature via sample permutation testing.In brief, the signature expression was first grouped according to ICB response.PCA transformed the primary data into a set of independent principal components and their eigenvalues.Next, 1000 permutations of the original sample groups were performed to generate a distribution of PCA eigenvalues under the null hypothesis, which assumes no specific directional relationship between the geneset and sample groups.Finally, the mean of the 95th percentile of these permuted eigenvalues for each principal component was compared with the initially calculated eigenvalues to test the null hypothesis.If the actual eigenvalue for a principal component exceeds its randomly sampled counterpart, it suggests that the specific component explains a degree of variance in the dataset significantly higher than chance.
A PLS-DA with permutation testing was conducted at each time point to evaluate the discriminative power and robustness of our EN-identified signature in distinguishing ICB responses.Signature expression was partitioned into training (seven samples) and test sets (three samples), with the training set used to construct a PLS-DA model and perform cross-validation.Each of the 1000 permutations of data partitioning generated cross-validation and test error rates, creating a null distribution against which error rates from randomly-generated gene panels were compared.To generate these random gene panels, 100,000 iterations were performed, each involving the selection of a random gene panel of up to 15 genes from the DEG gene set.These panels underwent similar data partitioning, PLS-DA model testing, and error rate calculation to the niche-derived gene signatures to highlight that the discriminative power of these signatures is not the result of random chance.

Flow cytometry
The primary tumor, spleen, and IN were disassociated and digested as previously detailed. 39,44Tissues were filtered through a 70-μm cell strainer (Corning).Single cell suspensions were then prepared by erythrocyte lysis in ACK buffer (Fisher) and washed in PBS (2 mM EDTA, 0.5% bovine serum albumin) by centrifuging at 500 × g for 5 min.

Study approval
All procedures were performed in accordance with guidelines and protocols approved by the University of Michigan Institutional Animal Care and Use Committee.
sensitive or resistant indicated by primary tumor growth and extended survival.The IN was sampled during and after ICB and sequenced to identify gene expression signatures that correlate with sensitivity or resistance.We also analyzed gene expression at the IN prior to ICB treatment to derive markers predicting therapeutic response.Longitudinally interrogating an IN, to monitor changes associated with ICB response, presents a new opportunity to personalize care and investigate mechanisms underlying treatment resistance.

F I G U R E 1
Response to antibody to programmed cell death protein 1 (anti-PD-1) immune checkpoint blockade treatment.(A) Schematic representation of tumor cell inoculation and immune checkpoint blockade (ICB) treatment and response.(B) Tumor growth following inoculation with n = 10 per group, *p < 0.05, and values = mean ± standard error mean (SEM).(C) Mouse survival after tumor cell inoculation with n = 5 per group, *p < 0.05 by Log-rank (Mantel-Cox) test.(D-F) Immunohistochemistry of PD-L1 expression in the primary tumor of mice (D) 7 days after inoculation and before ICB treatment, (E) 21 days after inoculation in ICB-sensitive mice, and (F) 21 days after inoculation in ICB-resistant mice.Scale bar = 500 μm.Images representative of n = 10-15 independent samples per group.(G) Quantification of immunohistochemistry staining for PD-L1 in primary tumor samples at day 21, corresponding to the representative images above.n = 10-15 independent samples per group; **p < 0.01 and ns indicates not significant.
The IN were retrieved prior to the initiation of ICB (D7), during ICB administration (D14), and following completion of therapy (D21).The mice were similarly stratified following ICB treatment with IN implants as ICB-sensitive or ICB-resistant (Figure S2A, B).Key leukocyte populations were analyzed in response to ICB at the IN, PT, and the spleen after ICB therapy (D21) (Figure 2B-E, Figures S2- Cellular pathways related to chemotaxis, differentiation, and degranulation were downregulated at the niche in the ICB-resistant cohort.For IN lymphocytes, including adaptive immune cells (T and B cells) and NK cells that contribute to tumor cell killing (Figure 3C), pathways were associated with activation, differentiation, proliferation, and function.Proinflammatory pathways, including activation of T cells, B cells, and NK cells were enriched in the ICB-sensitive cohort.Cytokine/chemokine pathways are common mechanisms of communication among the innate and adaptive immune cells, and several factors upregulated in the ICB-sensitive cohort contribute to anti-tumor, pro-inflammatory responses (Figure S6B).The most differentially regulated pathways of cytokine/chemokine regulation are those associated with interferon (IFN) signaling in cancer, including IFNγ and type 1 IFN, interleukin-12 (IL-12), IL-10, IL-8, IL-2, and IL-1.Many of the depleted cytokine/chemokine pathways in ICB-resistance are driven by myeloid cells and play a role in dampening T cell responses.We then coupled this enrichment analysis with a gene set variation analysis (GSVA) to specify the most differentially regulated pathways in ICB refractivity (Figure S7).The pathways with the lowest false discovery rate (< 0.1) F I G U R E 2 Cell phenotypes and ratios at the immunological niche after immune checkpoint blockade (ICB) treatment.(A) Schematic representation of immunological niche implants and explants during tumor cell inoculation and ICB treatment, where ICB refractivity was determined as previously on day 19 (D19).(B-E) Flow cytometry analysis of tissues isolated on D21 post-tumor cell inoculation with fluorophores labeling (B) myeloid cells including monocytes, neutrophils (Neut), and macrophages (Mac), and (C) lymphocytes including CD4 and CD8 T cells, B cells, and NK cells.Cell proportions quantified as percentage of the CD45+ population.(D) Ratio of neutrophils to CD8 T cells (left) or B cells (right).(E) Ratio of neutrophils (left) or macrophages (right) to NK cells.Two-tailed unpaired t-tests assuming unequal variance were performed for single comparisons between two conditions.n = 5-6 per group, *p < 0.05, and values = mean ± standard error mean (SEM).Cell markers: neutrophils (CD45+ CD11b+ Ly-6G+ Ly-6C−), monocytes (CD45+ CD11b+ Ly-6C+ Ly-6G−), macrophages (CD45+ CD11b+ F4/80+), CD4 T cells (CD45+ CD3+ CD4+), CD8 T cells (CD45+ CD3+ CD8+), B cells (CD45+ CD19+), and NK cells (CD45+ CD49b+).
of these 243 IN-derived DEGs after ICB (Figure S8A-D) can distinguish sensitive from resistant mice and motivates the development of a minimal gene signature of ICB response.Feature selection using elastic net (EN) regularization with random resampling selected the most divergent DEGs from this 243-gene panel to establish a fewest-necessary biomarker signature of ICB response.This EN analysis of DEGs reduced features to a panel of 21 genes that distinguish mice based on ICB F I G U R E 3 Differential pathways and genes at the immunological niche after immune checkpoint blockade (ICB) treatment.(A, B) Highly enriched or depleted immunologic niche (IN) pathways for ICB response (normalized enrichment score, |NES| > 1), including (A) innate immune cell pathways and (B) adaptive immune cell pathways.(C) Heat map of EN-identified signature of 21 genes that sparsely distinguish ICB response, with red being resistant and blue sensitive.n = 10 mice per group where each column represents the IN biopsy from a distinct mouse.(D) Principal component analysis (PCA) clustering of 21-gene signature, principal component 1 (PC1) versus PC2.n = 10 mice per group where points indicate individual IN biopsies from distinct mice and ellipses = 70% confidence intervals (arbitrary units).

F
I G U R E 4 Serial-normalized gene expression at the immunologic niche (IN) correlates to immune checkpoint blockade (ICB)-response.(A) Schematic representation of serial IN analyses.(B) Heat map with hierarchical clustering of 22-gene panel for both serial analyses D14-D7 and D21-D7.(C, D) Singular value decomposition metric versus Random Forest metric scoring system for (C) D14-D7 and (D) D21-D7.n = 10 per group and ellipses = 70% confidence intervals (arbitrary units).(E) Myeloid scores for 22 gene panel indicated by the purple to green coloring.Note that the dark purple color is associated with a positive score, and that the larger circles are associated with a greater differential response between sensitive and resistant.
These gene expression signatures indicate that, while lymphocyte and myeloid cell pathways at the IN are differentially regulated between ICB-sensitive and resistant mice, the most prognostic IN-derived signature monitoring ICB sensitivity is predominantly correlated with myeloid cells.Also, previous work has shown that the IN implant recruits tumor cells and immune cells, 39-41 yet these highly enriched and depleted pathways at the IN following ICB treatment identify the immune microenvironment, not just the cancer cells themselves, as contributors to ICB refractivity.

F I G U R E 5
Analysis of immunologic niche (IN)-derived analytes before administering therapy identifies predictive signature for immune checkpoint blockade (ICB)-response.(A, B) Gene set enrichment analysis (GSEA) analysis of gene expression before therapy for (A) lymphocyte (adaptive immune cell) pathways, and (B) myeloid cell (innate immune cell) pathways.Cutoff for normalized enrichment score (NES) > 1 used for GSEA analysis.(C) Heat map of IN-identified predictive signature of 15 genes.(D) Clustering of mice by principal component analysis before administering therapy based on 15-gene predictive signature.n = 10 scaffolds from independent mice per group.(E) Sensitivity and specificity of correctly identifying scaffolds from ICB-sensitive or ICB-resistant mice in 100-fold cross validation for the 15-gene before ICB signature with or without sample partitioning for cross validation compared to the Oncogene DX panel represented by ˆand panels associated with triple negative breast cancer (TNBC) responsiveness to ICB from references 83 and 84 represented by # and + , respectively.(F) Cross-validation error and test data classification error of the 15-gene panel partial least squares discriminant analysis (PLS-DA) model compared to a PLS-DA model derived from the 99th percentile of 100,000 random 15-gene panel permutations.
tures for monitoring ICB response at the IN and their association with myeloid cells.Niche-derived biomarkers were analyzed by normalizing gene expression during or after therapy to gene expression before initiating ICB.This serial, individualized analysis of the IN captured dynamic immune responses to ICB, and IN-derived gene expression was monitored to define ICB response both during and after ICB.Such a technology has the potential to augment radiographic imaging for monitoring ICB response, though clinical transition will require validation of optimal niche biopsy timing before, during, and after immunotherapy.An implant-derived gene signature was also identified for predicting ICB response prior to therapy administration.This signature of ICB sensitivity is distinct from the mid-and post-therapy signatures, signifying that treatment alters responses.Considering the toxic adverse effects and high cost associated with ineffectually administering ICB to ICB-resistant patients, a significant clinical need is identifying which patients would benefit from ICB prior to administration.Identifying patients in which combination immunotherapy would overcome ICB refractivity would also provide great clinical benefit,52,94 and understanding how the gene expression and pathway regulation at the IN before and during monotherapy resistance differs from combination therapy remains uncharacterized.The standard of care for predicting ICB response has been limited to PD-L1 expression, leukocyte infiltration, and tumor mutational burden.12While clinical trials have found improved response rates among breast cancer patients with high TMB and TNBC patients with PD-L1+ tumors, these biomarkers have failed to identify ICB-sensitive patients by predicting ICB response.Selecting specific TNBC patients with PD-L1+ tumor-infiltrating lymphocytes has enriched for patient subpopulations with the best ICB response, yet even the majority of this subpopulation is ICB-resistant.95The FDA has recently approved ICB in early TNBC, administered in the neoadjuvant and adjuvant setting.Neoadjuvant ICB is delivered prior to resection with the goal of reducing PT volume, and then adjuvant ICB is administered following resection to mitigate recurrence.11Not all patients require adjuvant ICB, and yet, no method is available to predict which patients would benefit from adjuvant ICB, following PT resection.Analysis of implant-derived gene expression prior to ICB identified multivariate analytes predictive of ICB response before initiating therapy.While immune cell types and dynamics were largely unchanged between ICB-sensitive and ICB-resistant mice after treatment, phenotypic changes at the level of gene expression were abundant both after and before therapy.Illustrative of this point, for mice sensitive to therapy pathways including myeloid cell homeostasis, myeloid cell differentiation, and neutrophil-mediated immunity were enriched within the IN before ICB, whereas macrophage tolerance and macrophage M1 versus M2 pathways were depleted at the IN before ICB compared to ICB-resistant mice.In this work, we subsequently focused on the IN-derived myeloid cells as acquiring cells from metastatic sites require invasive procedures and longitudinal biopsy of vital organs is not clinically translatable.Analyses of the engineered niche revealed the contribution of myeloid cell dysfunction in ICB-resistance before and during the progression of therapy.Myeloid cell variations exhibit distinct molecular landscapes related to therapy resistance.Solid tumors induce systemic immune alterations in which neutrophils and monocytes accumulate in immune niches and develop immunosuppressive activity,52,96 and the high proportion of myeloid cells present in the IN may be more susceptible to these systemic alterations.While this work explores primary tumor associated immune responses at the IN, metastatic cells may also differentially modulate the environment at the IN.
Finally, an IN-derived gene signature was identified that predicts ICB response prior to therapy initiation.The IN provides a unique tool to investigate the role of differentially regulated immune cell pathways responsible for heterologous ICB-resistance.Insight into dysregulated immune pathways underlying ICB-resistance could motivate selecting chemo-, immuno-, or radiation therapies as an induction strategy to condition patients predicted to be ICB-resistant, prior to ICB administration, toward ICB-sensitivity.The ability to longitudinally biopsy an accessible site for ICB response-associated biomarkers in real time, as well as to investigate mechanisms underlying therapy resistance, could dramatically impact how clinicians utilize ICB for cancer management.
• C until surgery.Three scaffolds were implanted in the dorsal subcutaneous space of 8-10-week-old female BALB/c mice (Jackson Laboratories, 000651).All animal care and procedures were performed in accordance with standards from the Guide for the Care and Use of Laboratory Animals and were carried out in compliance with protocols approved by the Institutional Animal Care and Use Committee (IACUC Protocol 00009715) at the University of Michigan (UM).Mice were housed in a pathogen-free environment under a 12-h light-dark cycle.Sample sizes were calculated based on previous experience by the investigators.The investigators were not blinded to allocation during experiments or analyses.No outlier values were excluded.Mice were anesthetized with 2% v/v isoflurane prior to implant and received subcutaneous carprofen (5 mg/kg) immediately before surgery and 24 h after surgery.

4 . 2
Tumor cell culture and orthotopic cell inoculations Orthotopic tumor cell inoculation was performed 2 weeks after IN implantation.4T1-luc2-tdTomato murine TNBC cells (PerkinElmer) were cultured in RPMI 1640 (Thermo Fisher Scientific) containing 10% fetal bovine serum (FBS) for 5 days.Tumor cells were enzymatically lifted from the tissue culture flask with trypsin for 10 min at 37 • C and resuspended in culture medium.Cells were centrifuged at 500 × g for 5 min and resuspended in phosphate buffered saline (PBS) at 40E6 cells/mL.The tumor cell line was previously confirmed to be pathogen-free and authenticated by short tandem repeat DNA analysis and compared to the ATCC STR profile database (DDC Medical).Orthotopic inoculations were performed by injecting 2E6 4T1 tumor cells to the fourth right mammary fat pad of 10-12-week-old mice.
Single cell suspensions were prepared from explanted INs and magnetically sorted against CD11b (Miltenyi Biotec).The enriched CD11b+ myeloid cell fraction and the CD11b− non-myeloid cell fractions were resuspended in Trizol and stored at −80 • C for RNA isolation.After RNA-seq of D21 myeloid (my) and non-myeloid (nm) cells from the niche, each gene was ascribed an ICB resistance differential (RD) score and a score associated with myeloid-cell score from the ICB-resistant (Res) and ICB-sensitive (Sen) cohorts as follows: RD = |myRes − nmRes| (1) Myeloid Score = (|myRes − mySen| − |nmRes − nmSen|) * RD (2) Two-tailed unpaired t-tests assuming unequal variance were performed for comparisons between conditions, namely ICB-sensitive and ICB-resistance.Median survival and Kaplan-Meier survival curves were analyzed with log-rank Mantel-Cox test for significance.Following RNA-seq normalization, t-tests were performed for single gene comparisons (p < 0.05).DEGs were then parsed via EN regularization, as outlined.GraphPad Prism 9, Microsoft Excel, and R were used for these analyses, with p < 0.05 considered significant.Error bars on plotted data are calculated as standard error mean (SEM).Data for flow cytometry, immunohistochemistry, tumor monitoring, and sequencing can be found at https://doi.org/10.7302/wthd-z193,with sequencing data and analyses code also found at https://doi.org/10.24433/CO.2502469.v1.All other data are available in the main text or the Supplementary Materials.
• C until the study endpoints to confirm ICB refractivity for each mouse.For flow cytometry analyses, mice were euthanized at study endpoint (D21) after determining ICB refractivity for each mouse and the primary tumor, spleen, and implant isolated and stored in cold PBS for immediate dissociation for flow cytometry.