Spatial-Immune Multi-omics Refines Prognostication in Early-Stage Estrogen Receptor-Positive Breast Cancer

Spatial-Immune Multi-omics Refines Prognostication in Early-Stage Estrogen Receptor-Positive Breast Cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Spatial-Immune Multi-omics Refines Prognostication in Early-Stage Estrogen Receptor-Positive Breast Cancer Zak Kinsella, Chowdhury Jahangir, Hannah Nyarko, Daria Kalinska-Lysiak, and 14 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7726652/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Despite a strong 5–10 year prognosis, in ER + HER2 − breast cancer relapses are common beyond 10 years. Genomic assays employed to determine recurrence risk at diagnosis are still ambiguous for a significant proportion of patients with Intermediate risk (i.e. Oncotype Dx Recurrence Score (RS) 16–25). We demonstrate that macrophage and T-helper cells facilitate and impede cytotoxic T-cells, are associated with extracellular matrix remodelling and M2-like (SPP1) genes, and cytotoxicity (GZMA, GZMB, PRF1), checkpoints (LAG3, PD-1, PD-L2), and exhaustion (TOX) genes respectively, consistent with restrained T-cell activation. High cytotoxic T-cell density was associated with poorer 15-year iDFS in the High RS (p = 0.017, FDR < 0.1). Critically, in the randomised Intermediate RS, a treatment-biomarker interaction indicated chemotherapy inferiority at higher stromal CD8 density (ΔLR-χ 2 : 7.36, p = 0.007), which was validated orthogonally on whole-resection specimens from the same cohort (ΔLR-χ 2 : 7.48, p = 0.006). Using the candidate biomarker suggested a potential treatment change for up to 50% of the Intermediate RS. Biological sciences/Cancer/Breast cancer Health sciences/Oncology/Cancer/Tumour biomarkers Biological sciences/Cancer/Cancer microenvironment Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION Estrogen Receptor-positive (ER + ) disease is conventionally viewed as an “immunologically cold” subtype of breast cancer, with lower infiltration of tumour-infiltrating lymphocytes (TILs) in comparison to more clinically aggressive subtypes such as Human Epidermal Growth Factor Receptor 2-positive (HER2+) and Triple-Negative Breast Cancer (TNBC) ( 1 , 2 ), as defined by an average infiltration ≤ 10% TILs ( 3 ). Despite this being an accepted threshold for the categorical definition of immune infiltration, i.e. hot or not, there is a mounting body of evidence to suggest that the immune microenvironments of early-stage disease, including the ER + HER2 − subtype, is non-homogenous ( 4 ). This is largely due to underlying genomic instabilities ( 5 ), their phenotypes ( 6 ), and resulting spatial structures (7). Consequently, these data may be useful prognostic indicators or predictive measures of treatment benefit, such as immunotherapies ( 8 – 10 ) or chemotherapy. The immune microenvironment in ER + disease is infiltrated by cells of both myeloid and lymphoid lineages, though the prognostic efficacy of their scoring in large pilot studies has been disappointing. Major effectors of the adaptive immune response (CD8 + T-cells) are not considered to stratify for improved outcome ( 11 ), helper phenotypes such as CD4 + T-helper and regulatory CD4 + FOXP3 + T-reg cells confer marginal prognostic information ( 12 – 14 ) or none at all, though CD20 + B-cells, particularly those enriched in tertiary lymphoid structures, show an association with Immune Checkpoint Inhibitor (ICI) response ( 7 ). Despite lymphoid lineages commanding adaptive immune responses, the tumour microenvironment of ER + disease is dominated by myeloid lineages, mostly macrophage ( 15 , 16 ), but also containing large frequencies of stromal fibroblast subpopulations ( 17 ) that appear to condition the acquiescence of immune responses to genomic events, or chemokine gradients, that would otherwise favour tumour killing. Management of early-stage ER + HER2 − breast cancer remains the subject of much debate, in which 3 ongoing prospective clinical trials (MINDACT, TAILORx, OPTIMA) ( 18 – 20 ) aim to accurately determine the requirement of chemotherapy. While current chemotherapeutic regimens offer benefit to some, others who have an inherently reduced risk of recurrence still receive chemotherapy where the majority would remain cancer-free without it ( 21 , 22 ). The most widely used test to decide whether chemotherapy is required for ER + HER2 − disease is Oncotype Dx. However, it has several shortcomings with room for improvement, such as the uncertainty around the necessity of chemotherapy for managing pre-menopausal patients ( 23 , 24 ). In this study, we sought to provide more granular information on recurrence risk across Oncotype Dx RS categories using region of interest-based spatial transcriptomics and spatial proteomics (Fig. 1 ), in a cohort of Irish patients previously enrolled on the TAILORx clinical trial (Table 1 . CONSORT diagram can be found in Supplementary Fig. S1 ). Table 1 Patient Cohort Characteristics of Irish patients previously enrolled in TAILORx (CTRIAL-IE 12–30, NCT02050750) N = 577 Characteristic Oncotype Dx RS Low (0–15) 224 (38.8%) Intermediate ( 16 – 25 ) 240 (41.6%) High (26–100) 113 (19.6%) Median Age (Range) 53 (25–79) 54 (34–75) 55 (28–74) Menopausal Status Pre/perimenopausal 104 (46.4%) 95 (39.6%) 45 (39.8%) Postmenopausal 120 (53.6%) 145 (60.4%) 70 (60.2%) Tumour Size T1 146 (65.2%) 159 (66.3%) 64 (56.6%) T2 78 (34.8%) 79 (32.9%) 49 (43.4%) Unknown - 2 (0.8%) - Histological Grade Grade I (Low) 56 (25.0%) 31 (12.9%) 3 (2.6%) Grade II (Intermediate) 141 (62.9%) 166 (69.2%) 49 (43.4%) Grade III (High) 22 (9.8%) 43 (17.9%) 61 (54.0%) Unknown 5 (2.2%) - - Histological Subtype Invasive Ductal Carcinoma (IDC) 173 (77.2%) 189 (78.8%) 103 (91.2%) Invasive Lobular Carcinoma (ILC) 31 (13.8%) 40 (16.7%) 6 (5.3%) Other 20 (9.0%) 6 (2.5%) 4 (3.5%) Adjuvant Treatment Hormone Therapy Alone 161 (71.9%) 123 (51.2%) 1 (0.9%) Hormone Therapy + Chemotherapy 63 (28.1%) 117 (48.8%) 112 (99.1%) Luminal Subtype Luminal A (Ki67 < 14%) 143 (64%) 115 (48%) 24 (21%) Luminal B (Ki67 ≥ 14%) 16 (7%) 58 (24%) 59 (52%) No Ki67 data 65 (29%) 68 (28%) 30 (27%) N.B. ‘Other’ histology includes: adenoid cystic, collision tumour IDC-ILC, DCIS, invasive cribform carcinoma, invasive mucinous carcinoma, invasive tubular carcinoma. RESULTS Macrophage and T-helper Cells Facilitate Cytotoxic T-cell Entry to Epithelia Macrophages are the most abundant cell phenotype in ER + disease and exhibit plasticity that can either facilitate or dampen the immune response. We firstly dichotomised the cohort by high vs low macrophage (CD68 + ) cell count (median, 13.8%) and investigated the change in immune percentage and density, across the tumour microenvironment (TME - stroma) and epithelia (tumour) (Fig. 2 A). Macrophage HIGH TMEs were significantly associated with increased T-helper (CD4 + ), cytotoxic T-cell (CD8 + ), and T-regulatory cells (CD4 + FOXP3 + ) (Fig. 2 A). Patients of higher risk also had tumours with greater immune infiltration than those of lower risk (Fig. 2 A – HT + CT). This underscores an immune response that increased in our cohort with macrophage infiltration and clinical risk. Next, we wanted to better understand the proclivity of immune phenotypes toward epithelial migration. We profiled the distance of each cell from an epithelial (PanCK + ) tumour mask across categories dichotomised by median macrophage (270 cells/mm 2 ) and cytotoxic T-cell (27 cells/mm 2 ) density (Fig. 2 B-E.). Immune phenotypes in cytotoxic T-cell HIGH TMEs were not significantly (Cohen’s d ≤ 0.20) associated with distance to epithelia. However, in macrophage HIGH environments, immune phenotypes were significantly spatially oriented in and around the tumour (Cohen’s d > 0.20). These included cytotoxic T-cells (median distance: 5.9µm vs 10.1µm), T-helper cells (median distance: 3.6µm vs 7.5µm), and T-regulatory cells (median distance: 4.0µm vs 8.1µm), suggesting a macrophage-associated rearrangement that could enhance tumour-immune contact and crosstalk. To assess whether immune crosstalk was driven by an architectural or spatial component, we next fit a logistic regression model. We used this model to predict whether immune cells in the TME (densities or cell-cell proximities) facilitated or impeded cytotoxic T-cell entry into epithelia (Fig. 2 F). We found that spatial proximity between phenotypes more strongly predicted facilitation and impediment than bulk densities, but the direction and significance of the odds depended on the interacting pair. Expectedly, a 1 standard deviation (SD) increase of cytotoxic T-cell proximity to tumour cells (~ 11µm decrease in distance) was associated with an 80% increased odds of entry (OR: 1.85, 95%CI: 1.26–2.72, p = 0.002). This was the only significant density or proximity variable across chemoendocrine-treated patients. In endocrine treated patients, however, increased macrophage (1 SD: ~300µm decrease in distance) and T-helper (1 SD: ~280µm decrease in distance) proximity to cytotoxic T-cells significantly increased epithelial accumulation of CD8 + cells by 400 and 280%, respectively (Macrophage OR: 5.18, 95%CI: 2.2–12.0, p = 0.00012. T-helper OR: 3.85. 95%CI: 1.6-9.0, p = 0.002). Macrophage HIGH TMEs also exhibited significantly shorter macrophage-CD8 and T-helper-CD8 distances (Supplementary Fig.S7), suggesting that macrophage-dominated environments can facilitate cytotoxic T-cell entry to epithelia in lower risk tumours; a pattern attenuated in higher risk disease. We next performed a regression analysis of macrophage density and key genes related to antigen presentation (CIITA, HLA-DRA, B2M, CD74, HLA-DRB), extracellular matrix formation (COL1A1, COL3A1, TIMP1, TIMP2, MMP2, MMP9), M1-like macrophages (CXCL9, NOS2, TNF, CD86, CD80, IL1B), and M2-like macrophages (SPP1, CD163, TGFB1, IL10) (Fig. 2 G). An increasing bulk macrophage density (steps + 50 cells/mm 2 ) in the TME was significantly associated with increasing extracellular matrix (COL1A1, COL3A1, TIMP2, MMP2) and M2-like macrophage genes (SPP1). This was true across both treatment strata. However, the expression of antigen presentation genes was only significantly positively associated with density within the endocrine treatment arm (HLA-DRA, B2M, CD74). These data suggest that higher risk diseases may globally shift macrophage toward a more suppressive state in which incumbent cytotoxic T-cells are not primed for tumour killing by antigen display, therefore potentially increasing T-cell exhaustion. We likewise performed a regression analysis for T-helper cell proximity to cytotoxic T-cells (steps − 20µm), across checkpoint (TIGIT, LAG3, CTLA4, PD-1, PD-L1, PD-L2, VSIR), cytotoxicity (GZMA, GZMB, PRF1, IFNG), and exhaustion (TOX, CXCL13, HAVCR2) gene modules (Fig. 2 H). As expected for nominal T-helper cell functionality, cytotoxic genes (GZMA, GZMB, PRF1: FDR < 0.1) were significantly positively associated with increasing proximity to CD8 + cells. Yet exhaustion gene (TOX: FDR < 0.1) and checkpoint genes (LAG3, PD-1, PD-L2: FDR < 0.1) also rose significantly in step. These data are consistent with a model of attempted cytotoxic response by cytotoxic T-cells that is rapidly met by suppressive signalling in the TME. In order to examine whether these cytotoxic T-cells were indeed suppressed or exhausted, we next regressed intra-epithelial cytotoxic T-cell density (steps + 50 cells/mm 2 ) against antigen presentation, cytotoxicity, checkpoints, and co-stimulation (CD40, TNFSF4, ICOSLG, TNFSF9) module genes (Fig. 2 I). This analysis revealed that the epithelial segment markedly upregulated genes in the antigen presentation module (HLA-DRA, HLA-DRB1, CIITA). This was most significant in higher risk diseases (additionally CD74, B2M). Interestingly, cytotoxicity, co-stimulation, and checkpoint module genes were not significantly associated with increasing iTIL cytotoxic T-cell density (Fig. 2 I). These data show a transcriptomic programme consistent with dampened effector cell function. While checkpoint genes were not significantly associated with density, the upregulation of antigen presentation transcripts suggests a potential for immune recognition and killing. If a growing frequency of cytotoxic T-cells were functionally unrestrained, costimulation and cytotoxicity module genes should also increase as anti-tumour immunity increases, yet we did not observe this effect. A highly suppressive milieu may be encountered during the transition of T-cells into epithelia. Spatial Immune Proximity is Independently Associated with Activity & Suppression Adjuvant treatment was administered based on the Oncotype Dx RS category of patients enrolled into our study cohort. Furthermore, patients with an Intermediate RS were randomised to either endocrine therapy only or chemoendocrine therapy. We hypothesised that more granular information on treatment benefit may be found by an analysis of immune phenotypes by genomic risk (Fig. 3 ). Macrophage HIGH TMEs were associated with significant increases in immune phenotypes across the Low and Intermediate RS (Fig. 3 A). In the High RS, only T-regulatory cells (CD4 + FOXP3 + ) significantly accumulated in macrophage HIGH TMEs (Fig. 3 A). We next investigated the distance of immune phenotypes to a tumour mask (Fig. 3 B). Our data again show that macrophage HIGH regions are associated with an increased recruitment of immune cells to the tumour border. This effect is also consistent across Oncotype Dx RS categories (Fig. 3 B). We found T-helper cell (CD4 + ) distances to tumour decrease significantly in the Intermediate and High RS (Cohens d ≥ 0.20). The majority of other immune phenotype distances to tumour were not significantly different (negligible Cohens d < 0.20). This suggests the presence of a macrophage-mediated hub in the peri-tumoural stroma. We therefore next investigated the distances of all immune phenotypes and tumour cells from the nearest macrophage (Fig. 3 C). We found that immune-effector phenotypes had significantly increased proximity to macrophage as genomic risk increases. The High RS had significantly shorter distances of B-cells (median 16.7µm), T-helper cells (median 16.2µm), and cytotoxic T-cells (median 15.9µm) from macrophage. There was no significant difference in tumour cell distance across all RS categories (median ~ 30µm), compatible with the formation of macrophage-mediated peri-tumoural hubs that recruit but suppress effector cells. We next tested whether spatial proximity or density is more significantly associated with the variability of immune effector-versus-suppressor functionality within the TME. We used a multivariable model adjusted for immune densities and distances (Fig. 3 D). This analysis reaffirmed our finding that spatial proximity of immune phenotypes is more strongly associated with immune function than their density in the TME. We recognised T-helper and macrophage proximity to be significantly and independently associated with aberrant immune functioning genes. Macrophage distances to cytotoxic T-cells was significantly associated with decreased cytotoxicity (p = 0.00491, FDR < 0.1) and decreased exhaustion (p = 0.00237, FDR < 0.1), though not for checkpoint programmes (p = 0.113). Similarly, T-helper proximity to cytotoxic T-cells was significantly associated with increased cytotoxicity (p = 0.000276, FDR < 0.1), increased exhaustion (p = 0.00057, FDR < 0.1), and increased checkpoint gene modules (p = 0.00726, FDR < 0.1). No singular immune-proximity pair was associated with simultaneously increased cytotoxic and decreased exhaustion or checkpoints module activity, suggesting no singular phenotype works to modulate the functionality of the milieu alone. However, T-helper, and macrophage, proximity to B-cells was associated with decreased checkpoints (p = 0.00509, FDR < 0.1), and increased cytotoxicity (p = 0.00869, FDR < 0.1), respectively. While B-cell occurrence was much rarer (median 0.07%) than either macrophages (median 13.8%) or T-helper cells (median 7.0%), their presence was favourable to immune function in the TME. Their recruitment, however, was not significantly greater in higher risk diseases, nor in macrophage HIGH environments, despite both of these subsets recruiting significantly more adaptive phenotypes. Cytotoxic T-cell density was associated with increased cytotoxicity (p = 0.0287, FDR < 0.1), confirming the potential for an anti-tumour immune response (Fig. 4 D). Though Low RS (p = 0.0156, FDR < 0.1), Intermediate RS (p = 0.0369, FDR < 0.1), and macrophage density (p = 0.000343, FDR < 0.1) were all significantly associated with decreased cytotoxicity (Fig. 4 D). These data confirm the reduced immunogenicity of lower risk diseases and the suppressive role of macrophages. However, bootstrapped partial R 2 plots indicated that neither Oncotype Dx RS-derived genomic risk nor immune densities strongly account for the explained variance with immune gene activity modules once proximity entered the model (Fig. 4 D, Barchart). This again demonstrates the power of spatial analysis over less granular bulk methodologies toward the functioning of the TME. Immune gene modules were aggregated in previous analyses. We subsequently examined which individual genes were significantly positively or negatively associated with macrophage density (steps + 50 cells/mm 2 . Figure 3 E), and T-helper proximity to cytotoxic T-cells (steps − 20µm. Figure 3 F), across the Oncotype Dx RS, as both were significantly associated with the functioning of the immune milieu. Bulk macrophage density was immunosuppressive (Fig. 3 E), associating with significantly decreased cytotoxicity genes irrespective of RS (GZMA, GZMB: encoding granzymes A and B, respectively). The Intermediate RS and High RS exhibited the largest number of negatively associated genes, which included TIGIT and LAG3 in the Intermediate RS, and TIGIT and CTLA-4 in the High RS, indicative of increased immune exhaustion in these RS categories. Irrespective of RS, T-helper cell distances to cytotoxic T-cells (Fig. 3 F) was significantly positively associated with both exhaustion (TOX) and cytotoxicity genes (GZMA, GZMB, and PRF1: encoding membrane-puncturing protein perforin-1). Interestingly, IFNG (encoding interferon-γ) was significantly positively associated with T-helper-CD8 + proximity only in Low and Intermediate RS. Checkpoint genes PDCD1 (encoding PD-1) and LAG3 were constitutively positively associated with proximity across RS categories, whereas PDCD1LG2 (encoding PD-L2) was only significant in the Low and Intermediate RS. These heatmaps suggest that T-helper cell proximity may trigger cytotoxic T-cell activity. T-helper mediation may then be rapidly met with suppressive signals in the TME, predominantly in the form of PD-L2 and CTLA-4 upregulation. The putative macrophage-mediated hub in the peri-tumoural stroma is likely an immune bottleneck that promotes T-cell exhaustion despite an attempted cytotoxic response. As exhausted milieu are known to impact prognosis in other breast cancer subtypes, we hypothesised that immune phenotypes may be prognostic across Oncotype Dx RS strata. We found in survival analysis of the High RS that continuous iTIL and sTIL cytotoxic T-cells (iTIL CD8%, p = 0.017, FDR < 0.01. sTIL CD8 density, p = 0.009, FDR 0.1) and the Low RS for stromal macrophage percentage (p = 0.028, FDR > 0.1). No proximity variable was significantly and independently associated with iDFS across Oncotype Dx RS (Supplementary Fig.S6). These data imply that, within high risk diseases, abundant cytotoxic T-cells reflect dysregulated or exhausted responses rather than effective anti-tumour immunity. Cytotoxic T-cell Density is a Predictive Marker in the Intermediate RS For patients with an Intermediate RS, the association of cytotoxic T-cell density with outcome was likely confounded by the randomisation of administered treatment. Investigating immune variables across randomised arms (Fig. 4 A) produced data in agreement with the High RS: patients receiving chemotherapy had significantly poorer 15-year iDFS if their stroma (HR: 1.17, 95%CI: 1.07–1.27) or epithelia (HR: 1.1, 95%CI: 1.03–1.18) contained high densities of cytotoxic T-cells (continuous, steps of 50 cells/mm 2 ). This pattern was observed likewise for CD8 + sTIL% (HR: 1.31, 95%CI: 1.07–1.60) and iTIL% (HR: 1.38, 95%CI: 1.10–1.73) (continuous, steps of 2%). Using median cutoff as an illustration for survival curves, high sTIL CD8 + density (> median, 30 cells/mm 2 ) had significantly poorer survival than low density in the chemoendocrine arm (HR: p = 0.0027, 15 year iDFS High: 66.5% vs 91.7% Low), likewise for high iTIL % (> median, 0.3%) (p = 0.0017, 15 year iDFS High: 68.2% vs 91.2% Low). Both high stromal density and epithelial count of cytotoxic T-cells provided significant additional prognostic information to nested models of clinical covariates (Fig. 4 B), underscoring the prognostic utility of immune phenotyping above canonical risk variables in the Intermediate RS. Patients with low CD8 + sTIL density or low iTIL% appeared to benefit from additional chemotherapy, as their survival trended higher than endocrine-only regimes yet was not significantly different (Fig. 4 C, sTIL density p = 0.76, iTIL% p = 0.51). Those with high CD8 + sTIL density and iTIL%, however, had significantly poorer outcome when receiving chemotherapy (sTIL density p = 0.016, 15 year iDFS HTCT: 66.5% vs 87.7% HT-only. iTIL % p = 0.0075, 15 year iDFS HTCT: 68.2% vs 88.6% HT-only) (Fig. 4 C). Interaction plots (Fig. 4 D) demonstrate that additional chemotherapy is significantly inferior to endocrine therapy alone in our cohort as sTIL CD8 + density rises (ΔLR χ²: 7.36, p = 0.007). iTIL% was not a significant discriminator (ΔLR χ²: 3.66, p = 0.056). As patients dropped out of analyses due to insufficient tissue, presence of artefactual cores, or withdrawal from the study, we next examined the proportion of clinical variables across dropout subsets and clinical variables, finding no significant difference in any strata (Supplementary Fig.S5). Secondly, as menopausal status is the current treatment stratifier for patients with an Intermediate RS (RS 16–25), we next wanted to examine the potential treatment change when using cytotoxic T-cell density over menopausal status. Current clinical guidelines suggest that pre-/peri-menopausal patients may benefit from chemoendocrine therapy (RS 16–25) whereas postmenopausal women benefit from endocrine therapy only (RS 0–25) (Fig. 4 E: Current Guidelines). Since our cohort was derived from patients previously enrolled in TAILORx, before these guidelines were updated, a large number of pre- and post-menopausal patients received either therapy (Fig. 4 E: Trial). Using high sTIL CD8 + density (> median) as a predictor of response, we see a potential treatment change of 50.4% (n = 56/111) of postmenopausal women (RS 16–25) from endocrine therapy alone to chemoendocrine therapy. Of pre-/peri-menopausal women (RS 16–25), 49% (n = 37/73) would have a change in treatment from chemoendocrine therapy to endocrine therapy alone. While the distribution is similar to the working trial cohort (Fig. 4 E: Trial, New), the Sankey diagram (Fig. 4 F) demonstrates that a significant proportion of those patients would experience a change in treatment despite a similar split in menopausal status within treatment strata to the trial distribution. Internal Orthogonal Validation on Whole Tissue Resections To mitigate potential TMA sampling bias and in lieu of an accessible validation cohort with similar 1:1 randomisation of the Intermediate RS, we next performed an internal orthogonal validation of cytotoxic T-cell density on whole resection specimens (Fig. 5 A, 5 B). Using an illustrative cut-off of median sTIL CD8 + density (median ~ 125 cells/mm 2 ) showed significantly poorer iDFS of patients receiving additive chemoendocrine therapy (p = 0.0062, 15 year iDFS HTCT: 61.8% vs 89% HT-only, Fig. 5 D). We crucially discovered a treatment-biomarker interaction, indicating poorer outcomes with chemotherapy as cytotoxic T-cell density increases (up to -30% survival difference at 15-years. ΔLR-χ 2 : 7.48, p = 0.00623) (Fig. 5 D). Absolute survival differences across prespecified percentiles (10th -90th, Fig. 5 E) showed a monotonic trend toward chemotherapy inferiority with high (> 50th percentile) sTIL CD8 + density. At the 90th percentile of density, a 3.4% (95CI: 0.5–14.7%, p = 0.006, FDR < 0.05), 8.1% (95%CI: 1.7–28.9%, p = 0.004, FDR < 0.05), and 13.8% (95%CI: 3.0–47.0%, p = 0.004, FDR < 0.05) higher iDFS was observed at 5yrs, 10yrs, and 15yrs, respectively, for patients receiving endocrine therapy only (Fig. 5 E). Over a 15 year horizon, the delta in restricted mean survival time was up to 12 months in favour of endocrine therapy-only (90th percentile, 95%CI: 2.16–37.4, p = 0.008, FDR < 0.05. Figure 5 F). Decision curve analysis demonstrated that sTIL CD8 + density provided higher net benefit than menopausal status across most thresholds. Particularly 0–3% and 8–30% (bootstrap p < 0.004, FDR q = 0.005. p < 0.006, FDR q = 0.062, respectively). To contextualise, a harm threshold of 15% indicates an intention to treat when the probability of recurrence is at least 15%, otherwise spare chemotherapy. Using sTIL CD8 + density at this threshold yielded a net benefit corresponding to 6 more patients per 100 correctly managed over the 15 year period compared with menopausal status (95%CI: 2.65–8.62, p = 0.002, FDR q = 0.0025). Compared with a chemo-for-all policy, sTIL CD8 + density yielded 11 more per 100 correctly managed (95%CI: 7.25–14.19, p = 0.002, q = 0.002). While these results require external validation to increase discriminatory power at lower percentiles, our data suggest that stromal cytotoxic T-cell density is a clinically translatable predictor that may help identify Intermediate RS patients (RS 16–25) more likely to be harmed by, or less likely to benefit from, additive chemotherapy regimens. Crucially, this data is independent of menopausal status. If validated, this could result in up to 50% of all Intermediate RS patients (RS 16–25) experiencing a change in adjuvant treatment from the existing paradigm. DISCUSSION Here, we have demonstrated that immune phenotyping is significantly prognostic in ER + HER2 − disease, but heavily dependent on the underlying genomic risk of the tumour. In whole-cohort assessments agnostic of genomic risk, TIL analysis produced results in agreement with the consensus view of low immunogenicity, poor prognostic dichotomisation, and weak, if no, ability to predict outcome in ER + HER2 − breast cancer overall. However, as the underlying tumour begins to acquire features associated with a decreased reliance on estrogen-related pathways and increased reliance on proliferative mechanisms (i.e. increasing Oncotype Dx RS), the immune milieu changes significantly. We have detailed how immunosuppressive cell niches, associated with macrophage enrichment, become dominant in breast TMEs that are acquiring higher risk features, and importantly demonstrate their presence in a surprising proportion of lower-risk disease. Further, we have outlined how these environments improve the odds of cytotoxic T-cell entry into epithelia, but are simultaneously and significantly associated with immune dysfunction, checkpoint expression, and extracellular matrix remodelling, that may blunt the response of effective tumour killing. The presence of cytotoxic T-cells in higher risk diseases was a significant negative predictor of 15-yr invasive disease-free survival in the High RS. Lastly, using randomised treatment in the Intermediate RS, we show that sTIL CD8 + density is a predictive marker for chemotherapy benefit across both TMA and orthogonal whole-resection specimens, above all clinical covariates, and suggests potential treatment changes for up to 50% of all patients with an Intermediate RS (RS 16–25) regardless of their menopausal status. We firstly outlined how the immune milieu of Oncotype Dx RS categories differ by immunosuppressive cells. Macrophage HIGH TMEs were associated with significantly increased stromal and intraepithelial phenotypes, promoted the recruitment of these cells to the peri-tumoural stroma, and facilitated the odds of epithelial transition of stromal cytotoxic T-cells. Increasing bulk stromal macrophage density was significantly positively associated with MHC-I-like (β2M) and MHC-II-like antigen presentation genes (HLA-DRA, CD74) only in lower risk patients, suggesting a role of T-helper crosstalk. Indeed, these macrophage HIGH environments significantly recruited T-helper cells, forming a macrophage-mediated hub. T-helper distance to cytotoxic T-cells was positively associated with cytotoxicity genes, but also checkpoint genes in PD-1, PD-L2, and LAG3 irrespective of clinical risk (endocrine-only, chemoendocrine arms). These data imply that adaptive recruitment and attempted cytotoxicity is attenuated in suppressive M2-like macrophage HIGH niches. Single-cell omics of luminal breast cancers has recently shown that suppressive myeloid niches can attract T-cells, denoting an exhausted cluster with altered but not abolished cytotoxicity, that corresponds with spatial hubs of macrophage/T-helper/T-reg/Cytotoxic T-cells, and enhanced matrix metalloproteinase expression ( 25 ). We also recognised that bulk macrophage density was significantly associated with extracellular matrix remodelling genes (COL1A1, COL3A1, TIMP2, MMP2), and M2-like gene SPP1. Recent study by Cha et al. has shown in HR + breast cancer an increased propensity of SPP1 + macrophage to interact with T-helper and cytotoxic T-cells, most notable in TIL-high tumours, in which SPP1 + macrophage inhibit T-cell responses to tumour through regulatory signalling and ECM formation ( 26 ). A principal immunosuppressive role of SPP1 + macrophage has also been observed in ovarian cancer ( 27 ), and fibrosis promotion via ECM remodelling by macrophage has been linked in breast cancer cell lines to dysfunctional CD8 + T-cell responses, independent of checkpoint signalling, by both physical exclusion and metabolic reprogramming that impedes antitumour immunity ( 28 ). The impact of an immunosuppressive milieu with reduced T-effector functionality was reflected in our finding that intraepithelial cytotoxic T-cell density is only significantly positively associated with the expression of antigen presentation genes in the epithelia. Paracrine IFN-γ secretion by T-cells in neighbouring immune-hot but suppressed peri-tumoural niches can induce MHC-II-like responses in epithelial cells ( 29 ), but the lack of significant costimulation, cytotoxicity, or checkpoint genes as iTIL CD8 density rises is commensurate with a blunted cytotoxic response. We recognised also that the delta and number of antigen presentation genes associated with iTIL CD8 density is greatest in chemoendocrine-receiving patients, in which estrogen-related signalling is lowest. Recent study suggests that ERα expression is inversely proportional to IFN-γ signalling, and that estradiol downregulates MHC-II and CIITA ( 30 ). Taken together, canonically lower risk, ERα-dependent tumours (i.e. the Low RS) may facilitate immune suppression. As neoantigen load, proliferation, and aggressive behaviour increase (i.e. Oncotype Dx RS shifts to Intermediate and High RS), the TME supplants the role of suppressive regulator in an environment that would otherwise be primed for anti-tumour immunity. This may explain why improved 15-year survival was seen in Intermediate RS patients receiving endocrine therapy alone vs chemoendocrine regimens – high sTIL CD8 + density may be capable of functional cytotoxicity once the estrogen-mediated suppression is lifted by endocrine therapies ( 31 ). More research is required to understand this effect in estrogen-driven breast cancers. With this in mind, we also recognised an association of immune proximities with checkpoint activity (LAG3, PDCD1 – encoding PD-1, and PDCD1LG2 – encoding PD-L2) and exhaustion (TOX). Recent study suggests Intermediate RS tumours contain exhausted T-cells with negative prognostic influence ( 32 ), and that checkpoints such as PD-L1 is positively associated with Oncotype Dx RS ( 33 ). More recently, larger studies have begun to demonstrate PD-L2 may actually be a more granular predictor of early relapse risk in ER + HER2 - breast cancer ( 34 ), an independent variable in multivariable analysis of those patients treated with chemoendocrine therapy. Our data across Oncotype Dx RS categories suggest that these patients contain a macrophage/T-helper/Cytotoxic T-cell spatial hub, containing greater abundances of transcript-level, predominantly PDCD1LG2 (PD-L2), LAG3, and PDCD1 (PD-1) expressing milieu. Patients with high cytotoxic T-cell densities having a paradoxically poorer invasive disease-free survival when receiving chemoendocrine therapy is a rationale for continued exploration of immune checkpoint blockade in early-stage ER + HER2 - breast cancer. This is particularly pressing in Intermediate and High RS disease, where reliance on ERα and estrogen signalling is decreasing, especially considering the successes of recent trials with immune checkpoint blockade in the arena of high risk, later stage ER + disease (Checkmate 7FL: ( 8 ), KEYNOTE 756: ( 10 , 35 ), I-SPY2: ( 36 , 37 )). However, further work is required to investigate with proteomics analyses the potential presence and relationship of checkpoints beyond the transcriptome. We lastly show that cytotoxic T-cell density is a predictive marker in patients with an Intermediate RS. This is, to our knowledge, the first time that a predictive marker of chemotherapy inferiority has been shown for the Intermediate RS. While our data provide strong evidence of improper treatment with additive chemotherapy, it is not yet known whether patients require de-escalation or combination treatment with immunotherapies. On the one hand, chemotherapy administration in early-stage ER + HER2 - disease may ablate existing adaptive responses ( 38 , 39 ), but a pre-treatment TME with increased immunosuppressive signalling will dampen the chemotherapy-induced stimulation of immunogenic cell death ( 40 , 41 ), and induce fibrotic responses through existing wound-healing macrophage ( 42 ) that promotes disease recurrence. External validation of these findings in the larger TAILORx trial is necessary. While we show that up to 50% of all Intermediate RS (RS 16–25) patients, irrespective of their menopausal status, may experience a change in treatment, a larger validation is required before clinical adoption can be seriously discussed. MATERIALS AND METHODS Study Design and TMA The TAILORx Tissue Bank (CTRIAL-IE 12–30, NCT02050750) is an exploratory, translational, non-interventional multicentre biobank sponsored by Cancer Trials Ireland that aims to identify potential biomarkers. Eligibility required prior registration with the TAILORx trial (CTRIAL-IE (ICORG) 06–31, NCT00310180), participation in trial arms and having sufficient tumour material available for immunohistochemical staining. Other than the accrual of patients who were also accrued to the Eastern Cooperative Oncology Group trial TAILORx, there was no connection between the two studies, and the analysis performed on the patient samples from CTRIAL-IE 12–30 did not impinge in any way on the TAILORx trial. RS values were taken as being 0–15 for Low, 16–25 for Intermediate, and 26–100 for High, following from the recommendations of the TAILORx studies ( 43 , 44 ). Cohort characteristics can be found in Table 1 . The primary end-point of this study was invasive disease-free survival (iDFS), as per the STEEP criteria ( 45 ), defined as the first invasive recurrence (distant, ipsilateral, locoregional), second primary invasive cancers, or death from any cause. There had been 108 iDFS events over a median follow-up of 158 months (SD ± 42mo), with 48 events in the Intermediate RS. Of 577 patients entered for TMA construction, 109 had insufficient tumour content for TMA core sampling. Each core was sampled with a diameter of 1000µm and in triplicate per patient, leaving n = 468 patients and n = 1404 TMA total cores for spatial omics analyses. Due to too few cells or no cores, artefactual staining, sectioning, or imaging, and limitations of the GeoMx DSP instrument, a further 58 patients were dropped leaving 410 patients with whole-transcriptome spatial transcriptomics data. One core was selected per patient and segmented in the cytokeratin channel for epithelia vs microenvironment. Similarly, due to no tumour or too few cells, and artefacts present within cores, 26 patients were dropped leaving 442 patients for spatial proteomics analysis. Three cores were used per patient for proteomics analyses. Orthogonal validation was performed on n = 453 whole-resection specimens taken from the source tumour blocks of the same patient material used to construct TMA cores. A more detailed consort diagram can be found in Supplementary Fig. S1 . Spatial Proteomics mIF staining was carried out using sequential OPAL™ tyramide signal amplification (TSA) multiplexing method on a Bond-RXm Automated Research Stainer (Leica Biosystems, Newcastle, UK) on tissues within one month of their sectioning to minimise epitope degradation. Primary antibodies were applied sequentially with heat-induced epitope retrieval times, staining conditions, and pH values optimised per antibody and OPAL™ fluorophore combination. The staining order was applied (Supplementary Table 1.), and counterstained with spectral DAPI. Briefly, slides were blocked for 10min with 150µL of Akoya blocking buffer (#ARD1001EA, Akoya Biosciences, Menlo Park, CA) before incubating with 150µL antibody at room temperature (RT) for 30 mins. Slides were washed before applying 150µL rabbit or mouse linker (DAKO) for 20min at room temperature, where used. 150µL TSA-DIG was applied for 10 mins, when used. Linkers/ TSA-DIG steps were followed by a wash and 150µL of DAKO horseradish peroxidase for 20 mins, wash, and 150µL of the requisite OPAL™ fluorophore for 30 mins, or 20 mins for OPAL™ 780. Slides were then thoroughly washed and subjected to epitope retrieval for the subsequent antibody, for 20 mins at 95ºC, using either Leica Bond ER solution 1 (pH6) or solution 2 (pH9). IHC staining (Supplementary Table S1 .) was carried using DAKO EnVision FLEX kit (#K802321-1, Agilent Technologies, Stockport, UK) with a DAKO PT-Link and Link-48 system. Antigen retrieval was performed at 97ºC for 20 minutes at pH9 in EnVision FLEX high pH. Staining was performed in 48-slide batches with 5 minutes of FLEX peroxidase block, 20 minutes with CD8 (#IR62361-2, Agilent Technologies, Stockport, UK), and 5 minutes of FLEX DAB + Sub-Chromo for 5 minutes. Slides were washed with buffer for two cycles before FLEX haematoxylin was applied for 3 minutes, washed, and slides dehydrated at 60ºC for 1hr. 1mL drop of Sigma DPX was used for mounting and cover slipped. Tonsil tissue was used within each staining batch as a positive control to ensure intra-batch consistency. Spatial Transcriptomics A Nanostring GeoMx Morphology Marker Kit was stained for CD45, PanCK, and SYTO13 to FFPE-derived TMA tissues as outlined by the manufacturer, using a Leica Bond RX-m (Leica Biosystems, Newcastle UK). Antigen retrieval was performed in a Leica Bond RxM at 100ºC for 20mins in Tris-EDTA, with staining performed as per the manufacturer’s instruction (Nanostring GeoMx DSP Automated Slide Preparation User Manual). RNA targets were exposed by digestion in 0.1µg/ml proteinase k for 15 minutes. RNA in-situ hybridisation was performed at room temperature for 16hrs. The Whole Transcriptome Atlas (WTA) was applied, with tumour vs microenvironment regions semantically segmented and RNA probes aspirated on the Nanostring Digital Spatial Profiler instrument ( 46 ). One 660µm region was taken per patient for segmentation, using PanCK to semantically segment two Areas of Interest (AOIs): the tumour microenvironment from epithelia. Serial UV illumination of each compartment was used to sequentially collect mRNA probe barcodes from each segmented region. Collected probes were stored at -80ºC immediately after collection to preserve probes for sequencing. Library Preparation was performed at the Genomics Core Facility, Queen’s University of Belfast, Northern Ireland, according to MAN-10153-03 (Version Feb-2023). Resulting Sample Pools underwent AMPure Cleanup with KAPA Pure Beads (07983298001) followed by Quality Control checks with Qubit 1X dsDNA High Sensitivity Assay (Q33231) and Agilent Tapestation D1000 Assay (5067–5582). Two Sequencing Pools were made based on total ROI area of the Sample Pools. These underwent further QC checks with Qubit 1X dsDNA High Sensitivity Assay (Q33231) and Fragment Analyser HS Fragment Analyser Assay (DNF-474) to determine Molarity. A Sequencing Depth Factor of 100 was chosen (for Whole Transcriptome Atlas/WTA). Each Sequencing Pool was run on the Illumina NovaSeq 6000 Sequencer (S4 200 Cycle v1.5 (20028313) and S2 100 Cycle v1.5 (20028316) carts utilised for the two different Sequencing Pools). The following Read Structure was utilised (Read 1–27. Index Read 1–8, Index Read 2–8, Read 2–27). Sequenced reads were obtained in FASTQ format and processed through the GeoMx NGS pipeline to generate raw gene counts. Raw counts were imported into R Studio for quality control via Nanostring’s GeoMxWorkflow pipeline. Limit of quantitation (LOQ) was defined as the geometric mean of the negative control probes multiplied by the geometric standard deviation. Targets consistently below the LOQ were excluded. For normalisation, upper quartile (Q3) normalisation was applied: briefly, the count in one segment was divided by the 3rd quartile value for that segment, and subsequently multiplied by the geometric mean of the 3rd quartile of all segments. Visualisation of gene expression data was performed on log2-transformed, Q3-normalised data (R packages: ggplot2 and pheatmap ). Differential expression analysis between groups was performed with LimmaVoom pipeline, and Linear Mixed Models (LMM). False discovery rate (FDR) was controlled using the Benjamini-Hochberg method, and genes with a FDR ≤ 0.05 were identified as differentially expressed genes. Image Capture Fluorescent staining was captured using the Akoya PhenoImager™ HT Automated Quantitative Pathology Imaging System (Akoya Biosciences, Menlo Park, CA) at 20X magnification. Fluorescent channels were optimised using a spectral library of breast tissues stained single-plex for each of the fluorophores and subjected to the outlined protocol (antigen retrieval times and pH conditions equal to their order in the staining stack). The spectral library was constructed to minimise both spectral bleed-through and tissue autofluorescence, and to maximise the captured signal:noise ratio of each respective marker during imaging. The final resolution of captured image scans was 0.4992µm/px. Chromogenic IHC on WSIs were captured using a Leica Aperio AT2 (Leica Biosystems, Newcastle, UK) at 20x magnification, with a final resolution of 0.5025 µm/px. Digital Pathology and Phenotype Scoring All digital image analysis (DIA) was undertaken in QuPath ( 47 ). A bespoke script was utilised to stitch individual TMA core TIFF files to a singular pyramidal .ome.tif image (code available via GitHub: https://gist.github.com/coltegelston/7c5ef58b32dd3a1b6a3f34926f01d6b2 ). Nuclear detection in the DAPI channel was performed using the StarDist ( 48 ) QuPath plugin, or watershed cell detection in the haematoxylin channel (Supplementary Table S4-S5). Object detection classifiers were trained for individual markers using a Random Forest ( 49 ) classifier within QuPath with default features. A Random Forest pixel-based classifier was trained to distinguish macrophage due to their irregular shape, which made object detection-based methods perform sub-optimally, and used to generate detections with a filtering step by a minimum size of 15µm 2 and maximum 150µm 2 . These thresholds were chosen as they performed best in separating individual macrophage instances while reducing overcalling from spectral bleed-through, seen with the PanCK (Opal 690) channel. For separation of the TME compartments into epithelia versus microenvironment, a random forest pixel classifier was also trained with default features in the cytokeratin channel (mIF) or haematoxylin (IHC) and used to generate annotation masks of epithelia and TME. Subsequently, combination classifiers were constructed from single-marker object classifiers to discern all cells, including double-positive scoring such as CD4 + FOXP3 + T-regs or interacting cells, and resolved hierarchically so that detections were binned into either epithelia or microenvironment annotation mask. This method afforded further granularity in analysing whether TILs were intratumoural (within epithelial compartment) or stromal (within microenvironment). Detected and phenotyped objects were exported and compared core-core, per patient, demonstrating strong intrapatient correlation (Supplementary Fig.S2.). Artefactual cores (e.g. shearing, staining, blur, core translocation) and those with < 100 cells were omitted from analysis, whereas artefact was manually annotated out of the analysis area when these occurred in whole-slide images. In order to generate a singular score per patient, the mean percentage of each cell phenotype was taken across available cores and used as a singular TIL score. Overall TIL scores were calculated using all detected objects within both tumour epithelia and microenvironment masks. Microenvironmental (sTIL) and epithelial (iTIL) scores were calculated similarly, using the detected cells within each segment respectively. Density was calculated using count/mm 2 of the annotation mask. All immune scores across mIF and IHC can be found in Supplementary Table S2. Spatial analyses were performed using “Detection centroid distances 2D” script commands within QuPath. This command computes the 2D Euclidean distance (µm) between the centroid of a cell of a source class to the nearest centroid of a target class. These data were exported and analysed further in R Studio, summarised as the median distance per core. Examples of QuPath classifier generation for spatial proteomics and orthogonal validation can be seen in Supplementary Fig.S3-S4. All Ki67 values were calculated as outlined previously ( 50 ). Statistical Analyses Normality of data distribution was assessed using the Kolmogorov-Smirnov test with Lilliefors correction (R package nortest ). Differences in values between two categories was assessed using pairwise Mann-Whitney or Student’s t-test (R package stats ). Multiple sample overall difference was assessed using the Kruskal-Wallis or ANOVA tests (R package stats, FSA ), and either Dunn’s test with a correction for multiple testing by the Benjamini-Hochberg method (flagged at FDR < 0.05), or Tukey’s test was used to examine pairwise differences. Considering the large sample size for immune distances (Total N phenotypes = 4,700,000) and how small deviations in the distribution at these sample sizes would affect calculations of pairwise statistical difference, Cohen’s d was computed (R package effectsize ) to examine the magnitude of the effect size of immune distances and classified as negligible (≤ 0.2), or non-negligible (> 0.2). Drop-out and missingness analysis were assessed at the patient level within the Intermediate RS, comparing baseline covariates between included/excluded cases for each omics dataset. For categorical variables, Pearson’s χ 2 test was used for frequencies > 5, otherwise Fisher’s exact test was used (two-sided) (R package stats ). For continuous variables, Wilcoxon rank-sum (two-sided) was used. Within each modality, comparisons were adjusted for multiple comparisons using the Benjamini-Hochberg method (FDR < 0.05). To investigate which, if any, phenotype may facilitate or impede cytotoxic T-cell epithelial entry as stromal cytotoxic T-cell density increases, we fit patient-level binomial logistic regression models with a logit link (R package stats ) separately by treatment arm (endocrine-only, chemoendocrine). The response was supplied with n in and n out , corresponding to counts of CD8 cells inside epithelium and in stroma, respectively. Each model included stromal CD8 density (sCD8), a candidate predictor (densities and proximities) and their interaction, plus a uniform adjustment set of stromal densities (all except sCD8). The interaction tests whether, for the same + 1SD increase in stromal CD8 density, a higher level of the predictor multiplies the odds that a CD8 cell is iTIL rather than sTIL. An interaction OR > 1 indicates facilitation, OR < 1 indicates impediment. When the predictor was itself a density, it was omitted to prevent overadjustment. Oncotype Dx RS was included as a factor. Densities were log1p-transformed and z-scored. Distances were z-scored and sign flipped so that larger values corresponded with closer proximity. Both sCD8 and the predictors were Winsorized at 1% tails to limit outlier leverage. Inference used robust Wald statistics with covariate estimates (R package sandwich , coeftest ). The interaction term was summarised as an odds ratio (OR) with robust 95%CI. Multiple testing was controlled within each treatment arm using the Benjamini-Hochberg method (flagged at a discovery FDR < 0.1). Tautological proximity pairs (e.g. CD20-CD20) and variables with n ≤ 100 patients per arm were excluded. To relate immune densities or spatial proximity to functional or dysfunctional immune activity, we modelled gene expression modules against stromal immune densities, spatial proximities, and genomic risk. Predefined gene modules were derived by aggregating the mean expression of curated gene sets, for: antigen presentation (HLA-DRA, HLA-DRB1, CIITA, CD74, β2M), cytotoxicity (GZMA, GZMB, PRF1, IFNG), exhaustion (PDCD1, TOX, CXCL13, HAVCR2, TIGIT, LAG3), checkpoint inhibition (PD-L1, PD-L2, CTLA4, LAG3, TIGIT, VSIR), M1-like macrophage (CXCL9, NOS2, TNF, CD80, CD86, IL1B), M2-like macrophage (SPP1, CD163, TGFB1, IL10), and extracellular matrix (COL1A1, COL3A1, TIMP1, TIMP2, MMP2, MMP9). Module scores were z-standardised within the stromal compartment. Linear regression models (R package stats ) were fitted separately for each module, to include immune phenotype count and density, pairwise spatial proximity, and Oncotype Dx RS as predictors. Densities were log-transformed and z-scored. Distances were z-scored and negated so that larger values represented greater proximity. For each model, the regression coefficient with 95% CI were estimated (R package broom ) and p-values corrected for multiple testing using the Benjamini-Hochberg (BH) method (flagged at discovery FDR < 0.1). In order to quantify unique variance explained by each predictor, semi-partial R 2 values were calculated using bootstrap resampling (R package boot , B = 5000 resamples). The Delta-R 2 (variance loss) across resamples quantified the unique contribution of each predictor. To investigate whether macrophage density, cytotoxic T-cell density, or T-helper proximity to cytotoxic T-cells was related to immune gene activity, we modelled individual gene z-scores from each curated gene module above, as a function of density (cells/mm 2 ) or distance (µm). Densities were log-transformed, and z-score normalised within the stroma (macrophage) or epithelia (cytotoxic T-cell). Ordinary Least Squares (OLS) linear models (R package stats ) were fitted, with cluster robust standard errors (R package sandwich) , stratified by either treatment category (endocrine-only, chemoendocrine) or Oncotype Dx RS (Low, Intermediate, High). Models were adjusted for densities and distances, with group-specific slopes estimated (R package emmeans) to give effect sizes per + 1SD change in the variable. Slopes were back-transformed to enable interpretation, where changes in gene expression z-score could be viewed as intuitive steps in the predictor (Density: +50 cells/mm 2 . Distance: -20µm proximity). Multiple testing was controlled using the Benjamini-Hochberg method (flagged at discovery FDR < 0.1). Survival Analyses Survival analyses were performed via the Kaplan-Meier method (R package survival ), and Likelihood ratio-χ 2 tests and Harrell’s C-index were obtained from nested uni- and multivariable Cox models. All statistical tests were two sided, with α 0.05, p 0.001 (**), and 0.001 ≥ p (***). For all models the proportional hazard assumption was verified by cox.zph tests (R package survival. Supplementary Table S6-S9). To assess chemotherapy inferiority/superiority, we fit multivariable Cox models containing an interaction between adjuvant treatment (HT vs HT + CT) and the continuous immune variable (counts/densities), adjusting for clinical covariates (menopausal status, age < 65 vs ≥ 65, tumour size, histological grade, luminal subtype). Non-linearity was explored with restricted cubic splines (3 knots at Harrell’s defaults of 10th, 50th, 90th percentile) using rms, with 95% CIs from model standard errors ( 51 ). Absolute survival differences (ARDs) were computed by standardization (R package survival ). From the fitted Cox model with a continuous treatment*biomarker interaction, we generated model-based survival curves with covariates fixed at typical values (numerics at the cohort median; factors at the modal level) and baseline stromal CD8 density set to prespecified levels (percentiles, median). ARD was calculated at 5, 10, and 15 years where positive values favour endocrine therapy above chemoendocrine. Uncertainty was quantified with patient-level bootstrap resampling (1,000 resamples) with percentile 95% CIs. Stromal CD8 density was modelled on the z-scale for estimation and translated to cells/mm² for interpretation and display. Restricted mean survival time (RMST) was used to summarise treatment effects over the follow-up period (R package survival ). For a timepoint of 15 years, RMST equals the area under the survival curve up to 15-years. Delta-RMST was defined as: $$\:{\Delta\:}RMST\left(15yrs\right)=RMS{T}_{HT}-{RMST}_{HT}+CT\:$$ RMSTs were obtained by integrating the model-based survival functions to 15 years (from survfit) with bootstrap 95% CIs (1,000 resamples). Decision-curve analysis (DCA) was performed with a benefit-based variant ( 52 ). Clinical utility of treatment selection in the randomised arms of the Intermediate RS was evaluated with DCA at the 15-year timepoint (R package dcurves ). For each patient we estimated counterfactual 15-year absolute risks, under endocrine therapy alone and chemoendocrine therapy, from Cox proportional hazard models. From these, we defined chemotherapy harm at 15-years as the absolute risk increase: $$\:{\Delta\:}Risk=P\left(event\:\right|\:HT+CT)-P\left(event\:\right|\:HT)$$ So that positive values indicated chemotherapy inferiority. Absolute risks were obtained from each Cox model via the baseline cumulative hazard and individual linear predictors. We considered harm thresholds t ranging from 0 to 30% (in 1% steps). For each threshold, we evaluated a chemotherapy sparing policy: withhold chemotherapy if \(\:{\Delta\:}Risk\) ≥ t. For a given t , net benefit was computed per-patient and reported per 100 patients: $$\:Net\:benefit=\left({Risk}_{HT}-{Risk}_{Policy}\right)-\:\frac{t}{1-t}\:\times\:\:P\left(chemo\:under\:policy\left(t\right)\right),$$ Where \(\:{Risk}_{HT}\) is the mean 15-year risk if all patients received endocrine therapy only, and \(\:{Risk}_{Policy}\) is the mean 15-year risk under the treatment allocation rule (e.g. menopausal status toward endocrine-only or chemoendocrine, etc.). Net benefit was plotted for the biomarker policy (treatment*sTIL CD8 density), menopausal status policy, and endocrine therapy for all, chemoendocrine for all. We obtained 95% bootstrap percentile intervals by resampling over 1,000 replicates, recomputing net benefit over each replicate. Plots show median net benefit and 95% intervals. Declarations Acknowledgements/Funding We would like to thank Dr Heiko Dussman for assistance with the operation of the Bruker NanoString GeoMx platform; Dr Philip Schouten, Dr Elena Provezano, and Dr Aris Sionakidis for their help in reviewing and copy editing manuscripts. ZK was supported by the SFI Strategic Partnership “Precision Oncology Ireland” grant# 18/SPP/3522. DOC was supported by 18/SPP/3522 and the Irish Cancer Society Collaborative Cancer Research Centre “Breast-Predict” grant# CCRC13GAL. HN was supported by the SFI Centre for Research Training in Genomics Data Science and the Royal College of Surgeons in Ireland grant# 18/CRT/6214-RCSI-DGF-2022 and DKL was supported by the SFI Centre for Research Training in Genomics Data Science Grant# 18/CRT/6214. AC was supported by Grant# 18/SPP/3522. WMG was supported by SFI Grant# 18/SPP/3522, Irish Cancer Society Grant# CCRC13GAL as well as Science Foundation Ireland (SFI) under the Investigator Programme OPTi-PREDICT Grant# 15/IA/3104. JHMP was supported by Research Ireland grants 18/RI/5792 and 21/RI/9787. Conflicts of Interest: None to report. Data Access: where appropriate, source data files will be provided with the article. De-identified data collected in the Irish TAILORx study will be made available to researchers if access and use of the data has been approved by the respective trial management group. References Denkert C, Loibl S, Noske A, Roller M, Müller BM, Komor M et al (2010) Tumor-associated lymphocytes as an independent predictor of response to neoadjuvant chemotherapy in breast cancer. J Clin Oncol 28(1):105–113 Loi S, Michiels S, Salgado R, Sirtaine N, Jose V, Fumagalli D et al (2014) Tumor infiltrating lymphocytes are prognostic in triple negative breast cancer and predictive for trastuzumab benefit in early breast cancer: results from the FinHER trial. Ann Oncol 25(8):1544–1550 Salgado R, Denkert C, Demaria S, Sirtaine N, Klauschen F, Pruneri G et al (2015) The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol 26(2):259–271 Wagner J, Rapsomaniki MA, Chevrier S, Anzeneder T, Langwieder C, Dykgers A et al (2019) A Single-Cell Atlas of the Tumor and Immune Ecosystem of Human Breast Cancer. Cell 177(5):1330–45e18 Curtis C, Shah SP, Chin S-F, Turashvili G, Rueda OM, Dunning MJ et al (2012) The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486(7403):346–352 Ali HR, Jackson HW, Zanotelli VRT, Danenberg E, Fischer JR, Bardwell H et al (2020) Imaging mass cytometry and multiplatform genomics define the phenogenomic landscape of breast cancer. Nat Cancer 1(2):163–175 Danenberg E, Bardwell H, Zanotelli VRT, Provenzano E, Chin S-F, Rueda OM et al (2022) Breast tumor microenvironment structures are associated with genomic features and clinical outcome. Nat Genet 54(5):660–669 Loi S, Salgado R, Curigliano G, Romero Díaz RI, Delaloge S, Rojas García CI et al (2025) Neoadjuvant nivolumab and chemotherapy in early estrogen receptor-positive breast cancer: a randomized phase 3 trial. Nat Med 31(2):433–441 Ríos-Hoyo A, Cobain E, Huppert LA, Beitsch PD, Buchholz TA, Esserman L et al (2024) Neoadjuvant Chemotherapy and Immunotherapy for Estrogen Receptor–Positive Human Epidermal Growth Factor 2–Negative Breast Cancer. J Clin Oncol 42(22):2632–2636 Cardoso F, O’Shaughnessy J, Liu Z, McArthur H, Schmid P, Cortes J et al (2025) Pembrolizumab and chemotherapy in high-risk, early-stage, ER+/HER2 – breast cancer: a randomized phase 3 trial. Nat Med 31(2):442–448 Ali HR, Provenzano E, Dawson SJ, Blows FM, Liu B, Shah M et al (2014) Association between CD8 + T-cell infiltration and breast cancer survival in 12 439 patients. Ann Oncol 25(8):1536–1543 Liu S, Foulkes WD, Leung S, Gao D, Lau S, Kos Z et al (2014) Prognostic significance of FOXP3 + tumor-infiltrating lymphocytes in breast cancer depends on estrogen receptor and human epidermal growth factor receptor-2 expression status and concurrent cytotoxic T-cell infiltration. Breast Cancer Res 16(5):432 Huang Y, Ma C, Zhang Q, Ye J, Wang F, Zhang Y et al (2015) CD4 + and CD8 + T cells have opposing roles in breast cancer progression and outcome. Oncotarget 6(19):17462–17478 Shang B, Liu Y, Jiang S-j, Liu Y (2015) Prognostic value of tumor-infiltrating FoxP3 + regulatory T cells in cancers: a systematic review and meta-analysis. Sci Rep 5(1):15179 Onkar S, Cui J, Zou J, Cardello C, Cillo AR, Uddin MR et al (2023) Immune landscape in invasive ductal and lobular breast cancer reveals a divergent macrophage-driven microenvironment. Nat Cancer 4(4):516–534 O’Meara T, Marczyk M, Qing T, Yaghoobi V, Blenman K, Cole K et al (2020) Immunological Differences Between Immune-Rich Estrogen Receptor–Positive and Immune-Rich Triple-Negative Breast Cancers. JCO Precision Oncol. (4):767–779 Nederlof I, Hajizadeh S, Sobhani F, Raza SEA, AbdulJabbar K, Harkes R et al (2022) Spatial interplay of lymphocytes and fibroblasts in estrogen receptor-positive HER2-negative breast cancer. npj Breast Cancer 8(1):56 Cardoso F, van't Veer LJ, Bogaerts J, Slaets L, Viale G, Delaloge S et al (2016) 70-Gene Signature as an Aid to Treatment Decisions in Early-Stage Breast Cancer. N Engl J Med 375(8):717–729 Sparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF et al (2018) Adjuvant Chemotherapy Guided by a 21-Gene Expression Assay in Breast Cancer. N Engl J Med 379(2):111–121 Stein RC, Dunn JA, Bartlett JM, Campbell AF, Marshall A, Hall P et al (2016) OPTIMA prelim: a randomised feasibility study of personalised care in the treatment of women with early breast cancer. Health Technol Assess 20(10):xxiii–xxix O'Brien MER, Borthwick A, Rigg A, Leary A, Assersohn L, Last K et al (2006) Mortality within 30 days of chemotherapy: a clinical governance benchmarking issue for oncology patients. Br J Cancer 95(12):1632–1636 van Dorp W, Haupt R, Anderson RA, Mulder RL, van den Heuvel-Eibrink MM, van Dulmen-den Broeder E et al (2018) Reproductive Function and Outcomes in Female Survivors of Childhood, Adolescent, and Young Adult Cancer: A Review. J Clin Oncol 36(21):2169–2180 Regan MM, Fleming GF, Walley B, Francis PA, Pagani O (2019) Adjuvant Systemic Treatment of Premenopausal Women With Hormone Receptor-Positive Early Breast Cancer: Lights and Shadows. J Clin Oncol 37(11):862–866 Vaz-Luis I, Francis PA, Di Meglio A, Stearns V (2021) Challenges in Adjuvant Therapy for Premenopausal Women Diagnosed With Luminal Breast Cancers. Am Soc Clin Oncol Educ Book 41:1–15 Tietscher S, Wagner J, Anzeneder T, Langwieder C, Rees M, Sobottka B et al (2023) A comprehensive single-cell map of T cell exhaustion-associated immune environments in human breast cancer. Nat Commun 14(1):98 Cha SM, Park J-W, Lee YJ, Lee HJ, Lee H, Lee IW et al (2024) SPP1 + macrophages in HR + breast cancer are associated with tumor-infiltrating lymphocytes. npj Breast Cancer 10(1):83 Hou L, Jiang M, Li Y, Cheng J, Liu F, Han X et al (2025) Targeting SPP1 + macrophages via the SPP1-CD44 axis reveals a key mechanism of immune suppression and tumor progression in ovarian cancer. Int Immunopharmacol 160:114906 Tharp KM, Kersten K, Maller O, Timblin GA, Stashko C, Canale FP et al (2024) Tumor-associated macrophages restrict CD8 + T cell function through collagen deposition and metabolic reprogramming of the breast cancer microenvironment. Nat Cancer 5(7):1045–1062 Axelrod ML, Cook RS, Johnson DB, Balko JM (2019) Biological Consequences of MHC-II Expression by Tumor Cells in Cancer. Clin Cancer Res 25(8):2392–2402 Mostafa AA, Codner D, Hirasawa K, Komatsu Y, Young MN, Steimle V et al (2014) Activation of ERα Signaling Differentially Modulates IFN-γ Induced HLA-Class II Expression in Breast Cancer Cells. PLoS ONE 9(1):e87377 Huang H, Zhou J, Chen H, Li J, Zhang C, Jiang X et al (2021) The immunomodulatory effects of endocrine therapy in breast cancer. J Experimental Clin Cancer Res 40(1):19 Egelston CA, Guo W, Tan J, Avalos C, Simons DL, Lim MH et al (2022) Tumor-infiltrating exhausted CD8 + T cells dictate reduced survival in premenopausal estrogen receptor–positive breast cancer. JCI Insight. ;7(3) Rozenblit M, Blenman K, Harigopal M, Reisenbichler E, Singh K, Qing T et al (2022) PD-L1 protein expression in relation to recurrence score values in early-stage ER + breast cancer. Breast Cancer Res Treat Chervoneva I, Peck AR, Sun Y, Yi M, Udhane SS, Langenheim JF et al (2023) High PD-L2 Predicts Early Recurrence of ER-Positive Breast Cancer. JCO Precision Oncol. (7):e2100498 Cardoso F, McArthur HL, Schmid P, Cortés J, Harbeck N, Telli ML, et al. LBA21 KEYNOTE-756:Phase III study of neoadjuvant pembrolizumab (pembro) or placebo (pbo) + chemotherapy(chemo), followed by adjuvant pembro or pbo + endocrine therapy (ET) for early-stage high-risk ER+/HER2– breast cancer. Annals of Oncology. 2023;34:S1260-S1 Nanda R, Liu MC, Yau C, Shatsky R, Pusztai L, Wallace A et al (2020) Effect of Pembrolizumab Plus Neoadjuvant Chemotherapy on Pathologic Complete Response in Women With Early-Stage Breast Cancer: An Analysis of the Ongoing Phase 2 Adaptively Randomized I-SPY2 Trial. JAMA Oncol 6(5):676–684 Wolf DM, Yau C, Campbell M, Glas A, Barcaru A, Mittempergher L et al (2025) Immune Subtyping Identifies Patients With Hormone Receptor–Positive Early-Stage Breast Cancer Who Respond to Neoadjuvant Immunotherapy (IO): Results From Five IO Arms of the I-SPY2 Trial. JCO Precision Oncol. (9):e2400776 Dixon-Douglas J, Virassamy B, Clarke K, Hun M, Luen SJ, Savas P et al (2024) Sustained lymphocyte decreases after treatment for early breast cancer. npj Breast Cancer 10(1):94 Park YH, Lal S, Lee JE, Choi Y-L, Wen J, Ram S et al (2020) Chemotherapy induces dynamic immune responses in breast cancers that impact treatment outcome. Nat Commun 11(1):6175 Michaud M, Martins I, Sukkurwala AQ, Adjemian S, Ma Y, Pellegatti P et al (2011) Autophagy-Dependent Anticancer Immune Responses Induced by Chemotherapeutic Agents in Mice. Science 334(6062):1573–1577 Galluzzi L, Humeau J, Buqué A, Zitvogel L, Kroemer G (2020) Immunostimulation with chemotherapy in the era of immune checkpoint inhibitors. Nat Reviews Clin Oncol 17(12):725–741 Monteran L, Ershaid N, Doron H, Zait Y, Scharff Ye, Ben-Yosef S et al (2022) Chemotherapy-induced complement signaling modulates immunosuppression and metastatic relapse in breast cancer. Nat Commun 13(1):5797 Sparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF et al (2015) Prospective Validation of a 21-Gene Expression Assay in Breast Cancer. N Engl J Med 373(21):2005–2014 Sparano JA (2018) Prognostic gene expression assays in breast cancer: are two better than one? npj Breast Cancer 4(1):11 Tolaney SM, Garrett-Mayer E, White J, Blinder VS, Foster JC, Amiri-Kordestani L et al (2021) Updated Standardized Definitions for Efficacy End Points (STEEP) in Adjuvant Breast Cancer Clinical Trials: STEEP Version 2.0. J Clin Oncol 39(24):2720–2731 Merritt CR, Ong GT, Church SE, Barker K, Danaher P, Geiss G et al (2020) Multiplex digital spatial profiling of proteins and RNA in fixed tissue. Nat Biotechnol 38(5):586–599 Bankhead P, Loughrey MB, Fernández JA, Dombrowski Y, McArt DG, Dunne PD et al (2017) QuPath: Open source software for digital pathology image analysis. Sci Rep 7(1):16878 Schmidt U, Weigert M, Broaddus C, Myers G (eds) Cell Detection with Star-Convex Polygons. Medical Image Computing and Computer Assisted Intervention – MICCAI 2018; 2018 2018//; Cham: Springer International Publishing Breiman L (2001) Random Forests. Mach Learn 45(1):5–32 Lynch SM, Russell NM, Barron S, Wang C-JA, Loughman T, Dynoodt P, et al. Prognostic value of the 6-gene OncoMasTR test in hormone receptor–positive HER2-negative early-stage breast cancer: Comparative analysis with standard clinicopathological factors. European Journal of Cancer. 2021;152:78–89 Gauthier J, Wu QV, Gooley TA (2020) Cubic splines to model relationships between continuous variables and outcomes: a guide for clinicians. Bone Marrow Transplant 55(4):675–680 Vickers AJ, van Calster B, Steyerberg EW (2019) A simple, step-by-step guide to interpreting decision curve analysis. Diagn Prognostic Res 3(1):18 Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryMaterials.docx Supplementary Materials Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7726652","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":526485715,"identity":"a7bd7222-b92d-4e87-9659-a876e2bdd009","order_by":0,"name":"Zak Kinsella","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYFACNmaGBCDFx87A+ABI8xCvBUgxGxCvBUqxSRDlLPn2Y8kGD3fUybExMz+rrqipk2FgP3x0Az4tjD1phxMSzxw2ZmNmM7t55thhHgaetLQb+LQwM6Q3H0hsAyJmBrObjQ0HeBgkeMzwamHjfw7SUlffxsz+rbCxoY6wFh4JkMPamBPYmHnMGBsbmAlrkZB4lmyQ2HbYsI2Zp1iyAegXNkJ+ke9PM5b82VYnz8/evvFjQ02dPT/74WN4tWDxHWnKR8EoGAWjYBRgAwAPiD4sLEDqqwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-6176-8895","institution":"Royal College of Surgeons in Ireland","correspondingAuthor":true,"prefix":"","firstName":"Zak","middleName":"","lastName":"Kinsella","suffix":""},{"id":526485716,"identity":"11347444-aec7-427c-b8d3-ef5037b60819","order_by":1,"name":"Chowdhury Jahangir","email":"","orcid":"","institution":"School of Biomolecular and Biomedical Science Master of Science","correspondingAuthor":false,"prefix":"","firstName":"Chowdhury","middleName":"","lastName":"Jahangir","suffix":""},{"id":526485717,"identity":"58cfcca0-991e-4903-b723-0b73c822017f","order_by":2,"name":"Hannah Nyarko","email":"","orcid":"","institution":"Royal College of Surgeons in Ireland","correspondingAuthor":false,"prefix":"","firstName":"Hannah","middleName":"","lastName":"Nyarko","suffix":""},{"id":526485718,"identity":"562144cb-e5b8-4bb7-b3e4-16e2cb2400dc","order_by":3,"name":"Daria Kalinska-Lysiak","email":"","orcid":"","institution":"Royal College of Surgeons in Ireland","correspondingAuthor":false,"prefix":"","firstName":"Daria","middleName":"","lastName":"Kalinska-Lysiak","suffix":""},{"id":526485719,"identity":"a78e23ce-010c-47e6-8d2e-9a5cd794fb35","order_by":4,"name":"Claudia Gonzalez","email":"","orcid":"","institution":"University College Dublin","correspondingAuthor":false,"prefix":"","firstName":"Claudia","middleName":"","lastName":"Gonzalez","suffix":""},{"id":526485720,"identity":"3e7b79fe-49b1-4f79-bcee-9e8d3fe98cd3","order_by":5,"name":"Verena Murphy","email":"","orcid":"","institution":"St Vincent's University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Verena","middleName":"","lastName":"Murphy","suffix":""},{"id":526485721,"identity":"08fb4c6c-b472-4e40-a283-f9e4c92ddc47","order_by":6,"name":"Tony O'Grady","email":"","orcid":"","institution":"Beaumont Hospital","correspondingAuthor":false,"prefix":"","firstName":"Tony","middleName":"","lastName":"O'Grady","suffix":""},{"id":526485722,"identity":"d57ec772-369e-42cf-a1d6-f60b621ae6c4","order_by":7,"name":"Joanna Fay","email":"","orcid":"","institution":"Beaumont Hospital","correspondingAuthor":false,"prefix":"","firstName":"Joanna","middleName":"","lastName":"Fay","suffix":""},{"id":526485723,"identity":"b2277b5a-9139-4c92-9311-90def13dfb7a","order_by":8,"name":"Katherine Sheehan","email":"","orcid":"","institution":"Beaumont Hospital","correspondingAuthor":false,"prefix":"","firstName":"Katherine","middleName":"","lastName":"Sheehan","suffix":""},{"id":526485724,"identity":"cde3a443-ebb3-46be-8016-1a40960f3582","order_by":9,"name":"Arman Rahman","email":"","orcid":"","institution":"University College Dublin","correspondingAuthor":false,"prefix":"","firstName":"Arman","middleName":"","lastName":"Rahman","suffix":""},{"id":526485725,"identity":"14031514-e9cd-4efc-8a58-641a5182133c","order_by":10,"name":"John Crown","email":"","orcid":"","institution":"St Vincent's University Hospital","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Crown","suffix":""},{"id":526485726,"identity":"f84d8c57-10df-4d02-9731-230db78760b4","order_by":11,"name":"Catherine Kelly","email":"","orcid":"","institution":"UCD School of Biomolecular and Biomedical Science","correspondingAuthor":false,"prefix":"","firstName":"Catherine","middleName":"","lastName":"Kelly","suffix":""},{"id":526485727,"identity":"5983a79a-184a-490b-bb13-49620a2d0248","order_by":12,"name":"Simon McDade","email":"","orcid":"https://orcid.org/0000-0002-3024-4773","institution":"Queen's University Belfast","correspondingAuthor":false,"prefix":"","firstName":"Simon","middleName":"","lastName":"McDade","suffix":""},{"id":526485728,"identity":"1b4cbe9b-4352-4fea-bc15-9cd99989a204","order_by":13,"name":"Ina Woods","email":"","orcid":"","institution":"Royal College of Surgeons in Ireland","correspondingAuthor":false,"prefix":"","firstName":"Ina","middleName":"","lastName":"Woods","suffix":""},{"id":526485729,"identity":"49f1ef84-219a-4987-a8b0-e5e7269c5c76","order_by":14,"name":"Niamh Connolly","email":"","orcid":"","institution":"Royal College of Surgeons in Ireland","correspondingAuthor":false,"prefix":"","firstName":"Niamh","middleName":"","lastName":"Connolly","suffix":""},{"id":526485730,"identity":"b88939ff-97bd-432e-b7a7-55996b7ecc2e","order_by":15,"name":"Jochen Prehn","email":"","orcid":"https://orcid.org/0000-0003-3479-7794","institution":"RCSI University of Medicine and Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Jochen","middleName":"","lastName":"Prehn","suffix":""},{"id":526485731,"identity":"fc0ea30f-2ef4-4cc8-8176-f029eed5f293","order_by":16,"name":"William Gallagher","email":"","orcid":"","institution":"University College Dublin","correspondingAuthor":false,"prefix":"","firstName":"William","middleName":"","lastName":"Gallagher","suffix":""},{"id":526485732,"identity":"66d339d8-748d-4703-842d-fcd6f6a26352","order_by":17,"name":"Darran O’Connor","email":"","orcid":"https://orcid.org/0000-0002-8835-1547","institution":"RCSI University of Medicine and Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Darran","middleName":"","lastName":"O’Connor","suffix":""}],"badges":[],"createdAt":"2025-09-27 07:20:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7726652/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7726652/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93341836,"identity":"d25accd8-527b-464a-a959-8fcbd1972b18","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4852879,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/8a51becb57e4f2056071fb63.png"},{"id":93343919,"identity":"abf64069-1622-4145-98df-33fcc5a1ce6f","added_by":"auto","created_at":"2025-10-12 14:56:58","extension":"png","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3402324,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/08efe4840f14ff3f6f07cfab.png"},{"id":93341844,"identity":"6797f569-2f08-4bc0-9f5e-0f54ec0135bf","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4117939,"visible":true,"origin":"","legend":"","description":"","filename":"SpOmicsERHER2ZKNatureComms.docx","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/aca4f88bb6aa001f76fd4274.docx"},{"id":93343921,"identity":"87f36f9f-e515-4cb3-b34e-be1f8bd58bd5","added_by":"auto","created_at":"2025-10-12 14:56:58","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2565887,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/e7eeeca3bf7e6ad49a2cb7de.png"},{"id":93343143,"identity":"e21a1aeb-4016-40eb-939f-6714091ac0a7","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1795813,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/a32a2fe023ac34913682399d.png"},{"id":93341846,"identity":"6777697e-34f7-4bad-aa55-8e88623ebd42","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9970450,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/48376ee7c313600df863f793.png"},{"id":93343150,"identity":"d35063a0-1f50-4886-928f-3825e5f27d42","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16736,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2575907.json","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/4b4fa1cb808394ebcd1bd974.json"},{"id":93343922,"identity":"6a611d71-114c-4a0d-b9e8-23f74fab4ce0","added_by":"auto","created_at":"2025-10-12 14:56:58","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5369507,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/7768bc6de67e385d96af3a00.docx"},{"id":93341855,"identity":"5a358fbb-1e31-40e2-a67d-8d2707eab4c8","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":183654,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25759070enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/f4c2f17dc7cf6e25c2011eb4.xml"},{"id":93341853,"identity":"0e26e022-706a-4f25-bdc3-3a45d3153a52","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4852879,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/9c4d8a7a9237dfe8c91a7cda.png"},{"id":93343146,"identity":"782e7f3b-e0ab-4ae6-813b-2da2589b35b6","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3402324,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/7295398f08de0e6780862ce1.png"},{"id":93341849,"identity":"9e0c8af2-1066-4329-811d-d28c7fec5f37","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2565887,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/cf381ac683da3ae5e61a60eb.png"},{"id":93341859,"identity":"69619dc5-8890-48ab-a7e8-ad4107784f32","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1795813,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/d1cb183b95bc37b3b2afd102.png"},{"id":93343151,"identity":"8259c942-aeeb-426c-8663-111b384ab30e","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9970450,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/3d7107e1efc49c65bbaeeec7.png"},{"id":93341848,"identity":"3fcddf6d-99bf-48be-b5e7-a299e1ca4601","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":880084,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/ec47e155084bb10a7718fd31.png"},{"id":93341858,"identity":"d90e7298-5c73-43ba-aca6-bb55e0ae1f36","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":682154,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/b93e741c105aa452a7309005.png"},{"id":93343152,"identity":"49c0787b-054b-4552-9712-6b99d92b9ef4","added_by":"auto","created_at":"2025-10-12 14:48:59","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":570767,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/6b1f3dfee1ee94a6ec7d568a.png"},{"id":93341863,"identity":"db6ab481-dd42-42d8-8769-c4154834fcc1","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":406035,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/d8487ebbe3423325f3653d50.png"},{"id":93343149,"identity":"fb89b1fd-6ab1-4849-aed3-5ef9adbaf501","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"jpeg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1323771,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/e9f7836e589d501c902c84f4.jpeg"},{"id":93341860,"identity":"1ebcc9b0-a032-42a3-bd44-ace3e09d6b03","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":897497,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig1New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/751e04c98875d1d10fff3df8.png"},{"id":93341865,"identity":"72573433-7423-46fc-a845-f0162aa1fa63","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":515416,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig2New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/c066cec91894f8e52745c402.png"},{"id":93341872,"identity":"f26d8075-6f2f-4164-8247-1a6136c00609","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":400769,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig3New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/11252f9abbf284edae3679ab.png"},{"id":93343147,"identity":"271f4624-1fc1-494f-89f8-ecbfc354f291","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":271003,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig4New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/a100357aa5953f47b37ffaf9.png"},{"id":93341861,"identity":"f02417ca-1a3a-47bc-83ff-7272f15bd980","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1456431,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig5New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/f25f9f2ab5209cd2929c753f.png"},{"id":93341870,"identity":"d0d411e4-1e55-4fdd-9ad4-f4c1fea69904","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":165341,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/09b76945bff52d47e035641e.png"},{"id":93341873,"identity":"7568d1d3-a540-4966-ba6d-bb9ef5f97d85","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110420,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/0dd2f0d0abfcbdfe41b3e2da.png"},{"id":93341847,"identity":"1358a160-dafe-4256-b008-1eb7f3caa444","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":88591,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/5d5d49183d9263b1f74cec0c.png"},{"id":93343153,"identity":"7cc1e9b9-1be3-4b76-b3bf-31d0e6207ec3","added_by":"auto","created_at":"2025-10-12 14:48:59","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":61408,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/f792becb66911e1bb3de93ce.png"},{"id":93341851,"identity":"7b8786c3-3a02-4b30-a3ae-a569237fa25a","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":346038,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/3ada9ae1cf4beb9756466ec7.png"},{"id":93341869,"identity":"02f17bb6-70fb-4f4d-b9c7-fd6e3365961a","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"xml","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":178890,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25759070structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/48b87f06a320c35d3d5fac10.xml"},{"id":93341871,"identity":"43a7aac7-30d1-4919-bb10-00f29bb503e9","added_by":"auto","created_at":"2025-10-12 14:40:59","extension":"html","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":192960,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/aab5f34b89ac0b0d11c179ab.html"},{"id":93341838,"identity":"98337609-ae0b-4839-ad94-56a69c8505dc","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":4852879,"visible":true,"origin":"","legend":"\u003cp\u003eAn overview of the workflow using whole resection specimen FFPE blocks for orthogonal validation, and TMA-blocks for spatial-omics samples. Further information on patient sample size after dropout can be seen in CONSORT diagrams, Supplementary Fig.S1.\u003c/p\u003e","description":"","filename":"Fig1New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/a0af72259be23f4d70fa6e77.png"},{"id":93343141,"identity":"3383dd88-fe15-475b-8507-01fe06afedf0","added_by":"auto","created_at":"2025-10-12 14:48:58","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":3402324,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial-Immune Context and Gene Readouts. A. \u003c/strong\u003eThe distribution of immune cell phenotype proportion and density, across the TME (stroma, left) and epithelia (tumour, right) stratified by cohort median macrophage percentage (13%). Colour opacity shows statistically significant pairwise comparisons.\u0026nbsp; \u003cstrong\u003eB. \u003c/strong\u003eRidgeline plots showing phenotype distances to a tumour mask (µm), dichotomised by cohort median macrophage (red) or cytotoxic T-cell (yellow) density. Colour opacity shows a Cohen’s d value ≥0.2 (non-negligible). Negative values show cell distances from the boundary within a tumour mask. \u003cstrong\u003eC.-E. \u003c/strong\u003eIllustrative examples of interacting cell pairs from spatial proteomics analyses. Scale bar shows 20 µm. \u003cstrong\u003eC. \u003c/strong\u003eMacrophage-Tumour, and Macrophage-Cytotoxic T-cell; \u003cstrong\u003eD. \u003c/strong\u003eT-helper cell-Cytotoxic T-cell; \u003cstrong\u003eE. \u003c/strong\u003eCytotoxic T-cell-Tumour cell. \u003cstrong\u003eF. \u003c/strong\u003eForest plots of proximity and density odds ratios (OR) for sTIL to iTIL CD8 conversion, stratified by received treatment. Points show OR per +1 SD in the predictor. Error bars show 95% Wald CI computed from robust standard errors. Models adjust for immune cell densities and the Oncotype Dx RS. \u003cstrong\u003eG.-I. \u003c/strong\u003eHeatmaps of regression slopes (Δ-gene z-score) for immune module genes by cell phenotype density or proximity covariates, adjusted for all other immune densities and proximities. Black outlines show tiles passing a discovery FDR\u0026lt;0.1. \u003cstrong\u003eG. \u003c/strong\u003eIncreasing stromal macrophage density by 50cells/mm\u003csup\u003e2\u003c/sup\u003e. \u003cstrong\u003eH. \u003c/strong\u003eDecreasing T-helper to Cytotoxic T-cell distance by 20 µm. \u003cstrong\u003eI. \u003c/strong\u003eIncreasing epithelial Cytotoxic T-cell density by 50cells/mm\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e","description":"","filename":"Fig2New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/a37ce8a1e6dc4b13f164d6b4.png"},{"id":93341845,"identity":"93ad3130-c23a-4939-91cb-c64851c73e4a","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2565887,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eMacrophage-mediated Spatial Hubs Associate with Immune Activity. A. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eThe distribution of immune phenotype proportion and density, across the TME (stroma, left) and epithelia (tumour, right) stratified by cohort median macrophage percentage (13%). Darker opacity shows statistically significant pairwise comparisons. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eB. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eRidgeline plots of immune phenotype distances to a tumour mask (µm), by Oncotype Dx RS categories, faceted by low (left) and high macrophage density (right). Colour opacity shows a Cohen’s d value ≥0.2 (non-negligible) versus Low RS. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eC. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eRidgeline plots of immune phenotype median distances to a macrophage, by Oncotype Dx RS categories. Colour opacity shows a Cohen’s d value ≥0.2 (non-negligible) versus Low RS. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eD. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eForest plots of regression coefficients for predicting gene module scores within TME (left), with bar charts of partial R\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e variance (right, 95%CI from 5,000 bootstrapped resamples). Module gene: Cytotoxic = GZMA, GZMB, PRF1, IFNG. Checkpoints = TIGIT, LAG3, CTLA4, CD274, PDCD1, PDCD1LG2, VSIR. Exhaustion = TOX, CXCL13, HAVCR2. Models adjust for immune densities. White points indicate p\u0026lt;0.05, gold points and bold labels denote FDR\u0026lt;0.1. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eE.-F. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eHeatmaps of regression slopes (Δ-gene z-score) for module genes against density and proximity covariates, adjusted for all other immune densities and proximities. Black outlines denote FDR\u0026lt;0.1 within RS*predictor strata. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eE. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eIncreasing stromal macrophage density by 50cells/mm\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eF. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eDecreasing T-helper to Cytotoxic T-cell distance by 20\u003c/em\u003e µm\u003cem\u003e. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eG. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eMultivariable Cox models: hazard ratios (HR) with 95% CI for continuous immune densities and proportion (per step in the label), adjusted for clinical covariates (age, menopausal status, histological grade, tumour size, luminal status). White points show p\u0026lt;0.05, gold points and bold annotation text denote FDR\u0026lt;0.1. Proximity variables were also tested but did not meet significance thresholds.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Fig3New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/28861587f51183c408032d30.png"},{"id":93344063,"identity":"db595286-8d46-4dd8-b415-7df6fe871f51","added_by":"auto","created_at":"2025-10-12 15:04:58","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1795813,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eImmune Phenotyping in the Intermediate RS. A. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eMultivariable Cox models: hazard ratios (HR) with 95% CI for immune densities and proportions (per step in the label), adjusted for clinical covariates (age, menopausal status, histological grade, tumour size, luminal status) across 1:1 randomised treatment arms. White points denote p\u0026lt;0.05, gold points denote FDR\u0026lt;0.1. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eB. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eKaplan-Meier curves for two exemplar biomarkers, dichotomised at the cohort median (left). Right panels show paired bar plots of model performance (ΔLR χ² and ΔC-index). Light green denotes the baseline clinical variable, dark green denotes bivariable models. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eC. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eKaplan-Meier curves for the same biomarkers, stratified by received adjuvant treatment (HT = endocrine therapy, HT+CT = chemoendocrine). Annotations report unadjusted Cox model HR with 95% CI and estimated 15-year invasive disease-free survival (iDFS) for each strata. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eD. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003ePredicted difference in 15-year iDFS (HT+CT – HT) from the spline-interaction Cox model as a function of the continuous biomarker (bottom), with a histogram and density curve of the biomarker distribution (bins 50 cells/mm\u003c/em\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e) (top). Shaded regions show 95% CI. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eE. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eStacked bar plots showing distributions of menopausal status and received treatment under Trial criteria, current guidelines, and the proposed new biomarker. Labels indicate patient counts. \u003c/em\u003e\u003cem\u003e\u003cstrong\u003eF. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eSankey diagram illustrating potential treatment reclassification from Trial criteria to proposed biomarker, with flows coloured by treatment and biomarker strata.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"Fig4New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/87c88895959fb443cc62cf76.png"},{"id":93341854,"identity":"1ed9c627-b76e-4d30-bece-182b3c141404","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":9970450,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOrthogonal Validation of the Proposed Biomarker. A. \u003c/strong\u003eExample CD8-IHC WSI. Scale bar shows 2mm. Yellow inset shows a region of interest (ROI). \u003cstrong\u003eB. \u003c/strong\u003eROI panels: raw image (top left), positive cell detection mask (top right), epithelia segmentation mask (bottom left), and CD8 density by nearest-neighbour (NN) of 50 µm (bottom right). Scale bars show 100 µm. \u003cstrong\u003eC. \u003c/strong\u003eKaplan-Meier curves of stromal CD8 density stratified by received adjuvant therapy (HT = endocrine therapy, HT+CT = chemoendocrine therapy) and dichotomised at the cohort median. \u003cstrong\u003eD. \u003c/strong\u003ePredicted difference in 15-year iDFS (HT+CT – HT) from the spline-interaction Cox model as a function of continuous stromal CD8 density (bottom), adjusted for all clinical covariates (age: ≥65, \u0026lt;65, menopausal status, histological grade, tumour size, luminal status), with a histogram and density curve of the biomarker distribution (top). Shaded areas correspond to 95% CI. \u003cstrong\u003eE. \u003c/strong\u003eAbsolute survival difference (HT – HTCT) for biomarker deciles across 5, 10, and 15 years (point colour encodes the time horizon). Error bars show 95% CI (bootstrap, 1,000 resamples). Colour opacity shows significance. \u003cstrong\u003eF. \u003c/strong\u003eDifference in restricted mean survival time at 15-years (Δ-RMST, HT – HT+CT) at the same biomarker percentiles. Error bars show 95% CI (bootstrap, 1,000 resamples). Colour opacity shows significance. \u003cstrong\u003eG. \u003c/strong\u003eDecision curve analysis comparing treatment policies across minimum chemotherapy harm thresholds. Curves show standardised benefit (higher is better). Teal lines show the proposed biomarker policy, purple lines show menopausal status policy. Red and green lines show treat-all (red) versus treat-none (green) policies. Shaded areas show 95% CI (bootstrap, 1,000 iterations). For interpretation, at a chosen threshold (e.g. 5%) the vertical distance between curves is the gain in net benefit, approximating the additional correct treatment decisions per 100 patients compared with the alternative policies.\u003c/p\u003e","description":"","filename":"Fig5New.png","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/d6cc59ac7b4f8819d3c5c1e5.png"},{"id":93344496,"identity":"faf1170e-d5eb-47ce-94ce-e6945fd74212","added_by":"auto","created_at":"2025-10-12 15:13:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":23937675,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/8684d45e-3db3-435a-8a1e-42f14268651d.pdf"},{"id":93341841,"identity":"74367f78-8293-40f1-91d1-3ac4cd7b0dab","added_by":"auto","created_at":"2025-10-12 14:40:58","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5369507,"visible":true,"origin":"","legend":"Supplementary Materials","description":"","filename":"SupplementaryMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-7726652/v1/dddc8a58cfdb53d304879d21.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Spatial-Immune Multi-omics Refines Prognostication in Early-Stage Estrogen Receptor-Positive Breast Cancer","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eEstrogen Receptor-positive (ER\u003csup\u003e+\u003c/sup\u003e) disease is conventionally viewed as an \u0026ldquo;immunologically cold\u0026rdquo; subtype of breast cancer, with lower infiltration of tumour-infiltrating lymphocytes (TILs) in comparison to more clinically aggressive subtypes such as Human Epidermal Growth Factor Receptor 2-positive (HER2+) and Triple-Negative Breast Cancer (TNBC) (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e), as defined by an average infiltration\u0026thinsp;\u0026le;\u0026thinsp;10% TILs (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Despite this being an accepted threshold for the categorical definition of immune infiltration, i.e. hot or not, there is a mounting body of evidence to suggest that the immune microenvironments of early-stage disease, including the ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e subtype, is non-homogenous (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). This is largely due to underlying genomic instabilities (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e), their phenotypes (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e), and resulting spatial structures (7). Consequently, these data may be useful prognostic indicators or predictive measures of treatment benefit, such as immunotherapies (\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e) or chemotherapy.\u003c/p\u003e\u003cp\u003eThe immune microenvironment in ER\u003csup\u003e+\u003c/sup\u003e disease is infiltrated by cells of both myeloid and lymphoid lineages, though the prognostic efficacy of their scoring in large pilot studies has been disappointing. Major effectors of the adaptive immune response (CD8\u003csup\u003e+\u003c/sup\u003e T-cells) are not considered to stratify for improved outcome (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e), helper phenotypes such as CD4\u003csup\u003e+\u003c/sup\u003e T-helper and regulatory CD4\u003csup\u003e+\u003c/sup\u003eFOXP3\u003csup\u003e+\u003c/sup\u003e T-reg cells confer marginal prognostic information (\u003cspan additionalcitationids=\"CR13\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) or none at all, though CD20\u003csup\u003e+\u003c/sup\u003e B-cells, particularly those enriched in tertiary lymphoid structures, show an association with Immune Checkpoint Inhibitor (ICI) response (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Despite lymphoid lineages commanding adaptive immune responses, the tumour microenvironment of ER\u003csup\u003e+\u003c/sup\u003e disease is dominated by myeloid lineages, mostly macrophage (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), but also containing large frequencies of stromal fibroblast subpopulations (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e) that appear to condition the acquiescence of immune responses to genomic events, or chemokine gradients, that would otherwise favour tumour killing.\u003c/p\u003e\u003cp\u003eManagement of early-stage ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e breast cancer remains the subject of much debate, in which 3 ongoing prospective clinical trials (MINDACT, TAILORx, OPTIMA) (\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e) aim to accurately determine the requirement of chemotherapy. While current chemotherapeutic regimens offer benefit to some, others who have an inherently reduced risk of recurrence still receive chemotherapy where the majority would remain cancer-free without it (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). The most widely used test to decide whether chemotherapy is required for ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e disease is Oncotype Dx. However, it has several shortcomings with room for improvement, such as the uncertainty around the necessity of chemotherapy for managing pre-menopausal patients (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). In this study, we sought to provide more granular information on recurrence risk across Oncotype Dx RS categories using region of interest-based spatial transcriptomics and spatial proteomics (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), in a cohort of Irish patients previously enrolled on the TAILORx clinical trial (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. CONSORT diagram can be found in Supplementary Fig.\u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePatient Cohort Characteristics of Irish patients previously enrolled in TAILORx (CTRIAL-IE 12\u0026ndash;30, NCT02050750)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eN\u0026thinsp;=\u0026thinsp;577\u003c/p\u003e\u003cp\u003eCharacteristic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eOncotype Dx RS\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003eLow (0\u0026ndash;15)\u003c/b\u003e\u003c/p\u003e\u003cp\u003e224 (38.8%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003eIntermediate\u003c/b\u003e (\u003cspan additionalcitationids=\"CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e)\u003c/p\u003e\u003cp\u003e240 (41.6%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003eHigh (26\u0026ndash;100)\u003c/b\u003e\u003c/p\u003e\u003cp\u003e113 (19.6%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eMedian Age\u003c/em\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003e(Range)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e53 (25\u0026ndash;79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e54 (34\u0026ndash;75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e55 (28\u0026ndash;74)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003e\u003cem\u003eMenopausal Status\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePre/perimenopausal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e104 (46.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e95 (39.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e45 (39.8%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePostmenopausal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e120 (53.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e145 (60.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e70 (60.2%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003e\u003cem\u003eTumour Size\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e146 (65.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e159 (66.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e64 (56.6%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e78 (34.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e79 (32.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e49 (43.4%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnknown\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2 (0.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003e\u003cem\u003eHistological Grade\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGrade I (Low)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e56 (25.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e31 (12.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3 (2.6%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGrade II (Intermediate)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e141 (62.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e166 (69.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e49 (43.4%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGrade III (High)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e22 (9.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e43 (17.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e61 (54.0%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUnknown\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5 (2.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003e\u003cem\u003eHistological Subtype\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInvasive Ductal Carcinoma (IDC)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e173 (77.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e189 (78.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e103 (91.2%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInvasive Lobular Carcinoma (ILC)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e31 (13.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e40 (16.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e6 (5.3%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOther\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e20 (9.0%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6 (2.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4 (3.5%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003e\u003cem\u003eAdjuvant Treatment\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHormone Therapy\u003c/p\u003e\u003cp\u003eAlone\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e161 (71.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e123 (51.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1 (0.9%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHormone Therapy\u0026thinsp;+\u0026thinsp;Chemotherapy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e63 (28.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e117 (48.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e112 (99.1%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eLuminal Subtype\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLuminal A (Ki67\u0026thinsp;\u0026lt;\u0026thinsp;14%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e143 (64%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e115 (48%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e24 (21%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLuminal B (Ki67\u0026thinsp;\u0026ge;\u0026thinsp;14%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16 (7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e58 (24%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e59 (52%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo Ki67 data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e65 (29%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e68 (28%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e30 (27%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eN.B.\u003c/b\u003e \u003cem\u003e\u0026lsquo;Other\u0026rsquo; histology includes: adenoid cystic, collision tumour IDC-ILC, DCIS, invasive cribform carcinoma, invasive mucinous carcinoma, invasive tubular carcinoma.\u003c/em\u003e\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eMacrophage and T-helper Cells Facilitate Cytotoxic T-cell Entry to Epithelia\u003c/h2\u003e\u003cp\u003eMacrophages are the most abundant cell phenotype in ER\u003csup\u003e+\u003c/sup\u003e disease and exhibit plasticity that can either facilitate or dampen the immune response. We firstly dichotomised the cohort by high vs low macrophage (CD68\u003csup\u003e+\u003c/sup\u003e) cell count (median, 13.8%) and investigated the change in immune percentage and density, across the tumour microenvironment (TME - stroma) and epithelia (tumour) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Macrophage\u003csup\u003eHIGH\u003c/sup\u003e TMEs were significantly associated with increased T-helper (CD4\u003csup\u003e+\u003c/sup\u003e), cytotoxic T-cell (CD8\u003csup\u003e+\u003c/sup\u003e), and T-regulatory cells (CD4\u003csup\u003e+\u003c/sup\u003eFOXP3\u003csup\u003e+\u003c/sup\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Patients of higher risk also had tumours with greater immune infiltration than those of lower risk (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA \u0026ndash; HT\u0026thinsp;+\u0026thinsp;CT). This underscores an immune response that increased in our cohort with macrophage infiltration and clinical risk.\u003c/p\u003e\u003cp\u003eNext, we wanted to better understand the proclivity of immune phenotypes toward epithelial migration. We profiled the distance of each cell from an epithelial (PanCK\u003csup\u003e+\u003c/sup\u003e) tumour mask across categories dichotomised by median macrophage (270 cells/mm\u003csup\u003e2\u003c/sup\u003e) and cytotoxic T-cell (27 cells/mm\u003csup\u003e2\u003c/sup\u003e) density (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB-E.). Immune phenotypes in cytotoxic T-cell\u003csup\u003eHIGH\u003c/sup\u003e TMEs were not significantly (Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e\u0026thinsp;\u0026le;\u0026thinsp;0.20) associated with distance to epithelia. However, in macrophage\u003csup\u003eHIGH\u003c/sup\u003e environments, immune phenotypes were significantly spatially oriented in and around the tumour (Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.20). These included cytotoxic T-cells (median distance: 5.9\u0026micro;m vs 10.1\u0026micro;m), T-helper cells (median distance: 3.6\u0026micro;m vs 7.5\u0026micro;m), and T-regulatory cells (median distance: 4.0\u0026micro;m vs 8.1\u0026micro;m), suggesting a macrophage-associated rearrangement that could enhance tumour-immune contact and crosstalk.\u003c/p\u003e\u003cp\u003eTo assess whether immune crosstalk was driven by an architectural or spatial component, we next fit a logistic regression model. We used this model to predict whether immune cells in the TME (densities or cell-cell proximities) facilitated or impeded cytotoxic T-cell entry into epithelia (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF). We found that spatial proximity between phenotypes more strongly predicted facilitation and impediment than bulk densities, but the direction and significance of the odds depended on the interacting pair. Expectedly, a 1 standard deviation (SD) increase of cytotoxic T-cell proximity to tumour cells (~\u0026thinsp;11\u0026micro;m decrease in distance) was associated with an 80% increased odds of entry (OR: 1.85, 95%CI: 1.26\u0026ndash;2.72, p\u0026thinsp;=\u0026thinsp;0.002). This was the only significant density or proximity variable across chemoendocrine-treated patients. In endocrine treated patients, however, increased macrophage (1 SD: ~300\u0026micro;m decrease in distance) and T-helper (1 SD: ~280\u0026micro;m decrease in distance) proximity to cytotoxic T-cells significantly increased epithelial accumulation of CD8\u003csup\u003e+\u003c/sup\u003e cells by 400 and 280%, respectively (Macrophage OR: 5.18, 95%CI: 2.2\u0026ndash;12.0, p\u0026thinsp;=\u0026thinsp;0.00012. T-helper OR: 3.85. 95%CI: 1.6-9.0, p\u0026thinsp;=\u0026thinsp;0.002). Macrophage\u003csup\u003eHIGH\u003c/sup\u003e TMEs also exhibited significantly shorter macrophage-CD8 and T-helper-CD8 distances (Supplementary Fig.S7), suggesting that macrophage-dominated environments can facilitate cytotoxic T-cell entry to epithelia in lower risk tumours; a pattern attenuated in higher risk disease.\u003c/p\u003e\u003cp\u003eWe next performed a regression analysis of macrophage density and key genes related to antigen presentation (CIITA, HLA-DRA, B2M, CD74, HLA-DRB), extracellular matrix formation (COL1A1, COL3A1, TIMP1, TIMP2, MMP2, MMP9), M1-like macrophages (CXCL9, NOS2, TNF, CD86, CD80, IL1B), and M2-like macrophages (SPP1, CD163, TGFB1, IL10) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG). An increasing bulk macrophage density (steps\u0026thinsp;+\u0026thinsp;50 cells/mm\u003csup\u003e2\u003c/sup\u003e) in the TME was significantly associated with increasing extracellular matrix (COL1A1, COL3A1, TIMP2, MMP2) and M2-like macrophage genes (SPP1). This was true across both treatment strata. However, the expression of antigen presentation genes was only significantly positively associated with density within the endocrine treatment arm (HLA-DRA, B2M, CD74). These data suggest that higher risk diseases may globally shift macrophage toward a more suppressive state in which incumbent cytotoxic T-cells are not primed for tumour killing by antigen display, therefore potentially increasing T-cell exhaustion.\u003c/p\u003e\u003cp\u003eWe likewise performed a regression analysis for T-helper cell proximity to cytotoxic T-cells (steps \u0026minus;\u0026thinsp;20\u0026micro;m), across checkpoint (TIGIT, LAG3, CTLA4, PD-1, PD-L1, PD-L2, VSIR), cytotoxicity (GZMA, GZMB, PRF1, IFNG), and exhaustion (TOX, CXCL13, HAVCR2) gene modules (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH). As expected for nominal T-helper cell functionality, cytotoxic genes (GZMA, GZMB, PRF1: FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1) were significantly positively associated with increasing proximity to CD8\u003csup\u003e+\u003c/sup\u003e cells. Yet exhaustion gene (TOX: FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1) and checkpoint genes (LAG3, PD-1, PD-L2: FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1) also rose significantly in step. These data are consistent with a model of attempted cytotoxic response by cytotoxic T-cells that is rapidly met by suppressive signalling in the TME.\u003c/p\u003e\u003cp\u003eIn order to examine whether these cytotoxic T-cells were indeed suppressed or exhausted, we next regressed intra-epithelial cytotoxic T-cell density (steps\u0026thinsp;+\u0026thinsp;50 cells/mm\u003csup\u003e2\u003c/sup\u003e) against antigen presentation, cytotoxicity, checkpoints, and co-stimulation (CD40, TNFSF4, ICOSLG, TNFSF9) module genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eI). This analysis revealed that the epithelial segment markedly upregulated genes in the antigen presentation module (HLA-DRA, HLA-DRB1, CIITA). This was most significant in higher risk diseases (additionally CD74, B2M). Interestingly, cytotoxicity, co-stimulation, and checkpoint module genes were not significantly associated with increasing iTIL cytotoxic T-cell density (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eI). These data show a transcriptomic programme consistent with dampened effector cell function. While checkpoint genes were not significantly associated with density, the upregulation of antigen presentation transcripts suggests a potential for immune recognition and killing. If a growing frequency of cytotoxic T-cells were functionally unrestrained, costimulation and cytotoxicity module genes should also increase as anti-tumour immunity increases, yet we did not observe this effect. A highly suppressive milieu may be encountered during the transition of T-cells into epithelia.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eSpatial Immune Proximity is Independently Associated with Activity \u0026 Suppression\u003c/h3\u003e\n\u003cp\u003eAdjuvant treatment was administered based on the Oncotype Dx RS category of patients enrolled into our study cohort. Furthermore, patients with an Intermediate RS were randomised to either endocrine therapy only or chemoendocrine therapy. We hypothesised that more granular information on treatment benefit may be found by an analysis of immune phenotypes by genomic risk (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Macrophage\u003csup\u003eHIGH\u003c/sup\u003e TMEs were associated with significant increases in immune phenotypes across the Low and Intermediate RS (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). In the High RS, only T-regulatory cells (CD4\u003csup\u003e+\u003c/sup\u003eFOXP3\u003csup\u003e+\u003c/sup\u003e) significantly accumulated in macrophage\u003csup\u003eHIGH\u003c/sup\u003e TMEs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA).\u003c/p\u003e\u003cp\u003eWe next investigated the distance of immune phenotypes to a tumour mask (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Our data again show that macrophage\u003csup\u003eHIGH\u003c/sup\u003e regions are associated with an increased recruitment of immune cells to the tumour border. This effect is also consistent across Oncotype Dx RS categories (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). We found T-helper cell (CD4\u003csup\u003e+\u003c/sup\u003e) distances to tumour decrease significantly in the Intermediate and High RS (Cohens \u003cem\u003ed\u003c/em\u003e\u0026thinsp;\u0026ge;\u0026thinsp;0.20). The majority of other immune phenotype distances to tumour were not significantly different (negligible Cohens \u003cem\u003ed\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.20). This suggests the presence of a macrophage-mediated hub in the peri-tumoural stroma. We therefore next investigated the distances of all immune phenotypes and tumour cells from the nearest macrophage (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC). We found that immune-effector phenotypes had significantly increased proximity to macrophage as genomic risk increases. The High RS had significantly shorter distances of B-cells (median 16.7\u0026micro;m), T-helper cells (median 16.2\u0026micro;m), and cytotoxic T-cells (median 15.9\u0026micro;m) from macrophage. There was no significant difference in tumour cell distance across all RS categories (median\u0026thinsp;~\u0026thinsp;30\u0026micro;m), compatible with the formation of macrophage-mediated peri-tumoural hubs that recruit but suppress effector cells.\u003c/p\u003e\u003cp\u003eWe next tested whether spatial proximity or density is more significantly associated with the variability of immune effector-versus-suppressor functionality within the TME. We used a multivariable model adjusted for immune densities and distances (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). This analysis reaffirmed our finding that spatial proximity of immune phenotypes is more strongly associated with immune function than their density in the TME. We recognised T-helper and macrophage proximity to be significantly and independently associated with aberrant immune functioning genes. Macrophage distances to cytotoxic T-cells was significantly associated with decreased cytotoxicity (p\u0026thinsp;=\u0026thinsp;0.00491, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1) and decreased exhaustion (p\u0026thinsp;=\u0026thinsp;0.00237, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), though not for checkpoint programmes (p\u0026thinsp;=\u0026thinsp;0.113). Similarly, T-helper proximity to cytotoxic T-cells was significantly associated with increased cytotoxicity (p\u0026thinsp;=\u0026thinsp;0.000276, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), increased exhaustion (p\u0026thinsp;=\u0026thinsp;0.00057, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), and increased checkpoint gene modules (p\u0026thinsp;=\u0026thinsp;0.00726, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1). No singular immune-proximity pair was associated with simultaneously increased cytotoxic and decreased exhaustion or checkpoints module activity, suggesting no singular phenotype works to modulate the functionality of the milieu alone. However, T-helper, and macrophage, proximity to B-cells was associated with decreased checkpoints (p\u0026thinsp;=\u0026thinsp;0.00509, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), and increased cytotoxicity (p\u0026thinsp;=\u0026thinsp;0.00869, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), respectively. While B-cell occurrence was much rarer (median 0.07%) than either macrophages (median 13.8%) or T-helper cells (median 7.0%), their presence was favourable to immune function in the TME. Their recruitment, however, was not significantly greater in higher risk diseases, nor in macrophage\u003csup\u003eHIGH\u003c/sup\u003e environments, despite both of these subsets recruiting significantly more adaptive phenotypes.\u003c/p\u003e\u003cp\u003eCytotoxic T-cell density was associated with increased cytotoxicity (p\u0026thinsp;=\u0026thinsp;0.0287, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), confirming the potential for an anti-tumour immune response (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). Though Low RS (p\u0026thinsp;=\u0026thinsp;0.0156, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), Intermediate RS (p\u0026thinsp;=\u0026thinsp;0.0369, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1), and macrophage density (p\u0026thinsp;=\u0026thinsp;0.000343, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1) were all significantly associated with decreased cytotoxicity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). These data confirm the reduced immunogenicity of lower risk diseases and the suppressive role of macrophages. However, bootstrapped partial R\u003csup\u003e2\u003c/sup\u003e plots indicated that neither Oncotype Dx RS-derived genomic risk nor immune densities strongly account for the explained variance with immune gene activity modules once proximity entered the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD, Barchart). This again demonstrates the power of spatial analysis over less granular bulk methodologies toward the functioning of the TME.\u003c/p\u003e\u003cp\u003eImmune gene modules were aggregated in previous analyses. We subsequently examined which individual genes were significantly positively or negatively associated with macrophage density (steps\u0026thinsp;+\u0026thinsp;50 cells/mm\u003csup\u003e2\u003c/sup\u003e. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE), and T-helper proximity to cytotoxic T-cells (steps \u0026minus;\u0026thinsp;20\u0026micro;m. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF), across the Oncotype Dx RS, as both were significantly associated with the functioning of the immune milieu. Bulk macrophage density was immunosuppressive (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE), associating with significantly decreased cytotoxicity genes irrespective of RS (GZMA, GZMB: encoding granzymes A and B, respectively). The Intermediate RS and High RS exhibited the largest number of negatively associated genes, which included TIGIT and LAG3 in the Intermediate RS, and TIGIT and CTLA-4 in the High RS, indicative of increased immune exhaustion in these RS categories. Irrespective of RS, T-helper cell distances to cytotoxic T-cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF) was significantly positively associated with both exhaustion (TOX) and cytotoxicity genes (GZMA, GZMB, and PRF1: encoding membrane-puncturing protein perforin-1). Interestingly, IFNG (encoding interferon-γ) was significantly positively associated with T-helper-CD8\u003csup\u003e+\u003c/sup\u003e proximity only in Low and Intermediate RS. Checkpoint genes PDCD1 (encoding PD-1) and LAG3 were constitutively positively associated with proximity across RS categories, whereas PDCD1LG2 (encoding PD-L2) was only significant in the Low and Intermediate RS. These heatmaps suggest that T-helper cell proximity may trigger cytotoxic T-cell activity. T-helper mediation may then be rapidly met with suppressive signals in the TME, predominantly in the form of PD-L2 and CTLA-4 upregulation. The putative macrophage-mediated hub in the peri-tumoural stroma is likely an immune bottleneck that promotes T-cell exhaustion despite an attempted cytotoxic response.\u003c/p\u003e\u003cp\u003eAs exhausted milieu are known to impact prognosis in other breast cancer subtypes, we hypothesised that immune phenotypes may be prognostic across Oncotype Dx RS strata. We found in survival analysis of the High RS that continuous iTIL and sTIL cytotoxic T-cells (iTIL CD8%, p\u0026thinsp;=\u0026thinsp;0.017, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.01. sTIL CD8 density, p\u0026thinsp;=\u0026thinsp;0.009, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.01) are significantly and independently associated with poorer 15-year iDFS (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG). Trends were also recognised in the Intermediate RS for sTIL cytotoxic T-cell density (p\u0026thinsp;=\u0026thinsp;0.027, FDR\u0026thinsp;\u0026gt;\u0026thinsp;0.1) and the Low RS for stromal macrophage percentage (p\u0026thinsp;=\u0026thinsp;0.028, FDR\u0026thinsp;\u0026gt;\u0026thinsp;0.1). No proximity variable was significantly and independently associated with iDFS across Oncotype Dx RS (Supplementary Fig.S6). These data imply that, within high risk diseases, abundant cytotoxic T-cells reflect dysregulated or exhausted responses rather than effective anti-tumour immunity.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eCytotoxic T-cell Density is a Predictive Marker in the Intermediate RS\u003c/h3\u003e\n\u003cp\u003eFor patients with an Intermediate RS, the association of cytotoxic T-cell density with outcome was likely confounded by the randomisation of administered treatment. Investigating immune variables across randomised arms (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA) produced data in agreement with the High RS: patients receiving chemotherapy had significantly poorer 15-year iDFS if their stroma (HR: 1.17, 95%CI: 1.07\u0026ndash;1.27) or epithelia (HR: 1.1, 95%CI: 1.03\u0026ndash;1.18) contained high densities of cytotoxic T-cells (continuous, steps of 50 cells/mm\u003csup\u003e2\u003c/sup\u003e). This pattern was observed likewise for CD8\u003csup\u003e+\u003c/sup\u003e sTIL% (HR: 1.31, 95%CI: 1.07\u0026ndash;1.60) and iTIL% (HR: 1.38, 95%CI: 1.10\u0026ndash;1.73) (continuous, steps of 2%).\u003c/p\u003e\u003cp\u003eUsing median cutoff as an illustration for survival curves, high sTIL CD8\u003csup\u003e+\u003c/sup\u003e density (\u0026gt;\u0026thinsp;median, 30 cells/mm\u003csup\u003e2\u003c/sup\u003e) had significantly poorer survival than low density in the chemoendocrine arm (HR: p\u0026thinsp;=\u0026thinsp;0.0027, 15\u0026nbsp;year iDFS High: 66.5% vs 91.7% Low), likewise for high iTIL % (\u0026gt;\u0026thinsp;median, 0.3%) (p\u0026thinsp;=\u0026thinsp;0.0017, 15\u0026nbsp;year iDFS High: 68.2% vs 91.2% Low). Both high stromal density and epithelial count of cytotoxic T-cells provided significant additional prognostic information to nested models of clinical covariates (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB), underscoring the prognostic utility of immune phenotyping above canonical risk variables in the Intermediate RS.\u003c/p\u003e\u003cp\u003ePatients with low CD8\u003csup\u003e+\u003c/sup\u003e sTIL density or low iTIL% appeared to benefit from additional chemotherapy, as their survival trended higher than endocrine-only regimes yet was not significantly different (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC, sTIL density p\u0026thinsp;=\u0026thinsp;0.76, iTIL% p\u0026thinsp;=\u0026thinsp;0.51). Those with high CD8\u003csup\u003e+\u003c/sup\u003e sTIL density and iTIL%, however, had significantly poorer outcome when receiving chemotherapy (sTIL density p\u0026thinsp;=\u0026thinsp;0.016, 15\u0026nbsp;year iDFS HTCT: 66.5% vs 87.7% HT-only. iTIL % p\u0026thinsp;=\u0026thinsp;0.0075, 15\u0026nbsp;year iDFS HTCT: 68.2% vs 88.6% HT-only) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC). Interaction plots (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD) demonstrate that additional chemotherapy is significantly inferior to endocrine therapy alone in our cohort as sTIL CD8\u003csup\u003e+\u003c/sup\u003e density rises (ΔLR χ\u0026sup2;: 7.36, p\u0026thinsp;=\u0026thinsp;0.007). iTIL% was not a significant discriminator (ΔLR χ\u0026sup2;: 3.66, p\u0026thinsp;=\u0026thinsp;0.056).\u003c/p\u003e\u003cp\u003eAs patients dropped out of analyses due to insufficient tissue, presence of artefactual cores, or withdrawal from the study, we next examined the proportion of clinical variables across dropout subsets and clinical variables, finding no significant difference in any strata (Supplementary Fig.S5). Secondly, as menopausal status is the current treatment stratifier for patients with an Intermediate RS (RS 16\u0026ndash;25), we next wanted to examine the potential treatment change when using cytotoxic T-cell density over menopausal status. Current clinical guidelines suggest that pre-/peri-menopausal patients may benefit from chemoendocrine therapy (RS 16\u0026ndash;25) whereas postmenopausal women benefit from endocrine therapy only (RS 0\u0026ndash;25) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE: Current Guidelines). Since our cohort was derived from patients previously enrolled in TAILORx, before these guidelines were updated, a large number of pre- and post-menopausal patients received either therapy (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE: Trial). Using high sTIL CD8\u003csup\u003e+\u003c/sup\u003e density (\u0026gt;\u0026thinsp;median) as a predictor of response, we see a potential treatment change of 50.4% (n\u0026thinsp;=\u0026thinsp;56/111) of postmenopausal women (RS 16\u0026ndash;25) from endocrine therapy alone to chemoendocrine therapy. Of pre-/peri-menopausal women (RS 16\u0026ndash;25), 49% (n\u0026thinsp;=\u0026thinsp;37/73) would have a change in treatment from chemoendocrine therapy to endocrine therapy alone. While the distribution is similar to the working trial cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE: Trial, New), the Sankey diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eF) demonstrates that a significant proportion of those patients would experience a change in treatment despite a similar split in menopausal status within treatment strata to the trial distribution.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eInternal Orthogonal Validation on Whole Tissue Resections\u003c/h3\u003e\n\u003cp\u003eTo mitigate potential TMA sampling bias and in lieu of an accessible validation cohort with similar 1:1 randomisation of the Intermediate RS, we next performed an internal orthogonal validation of cytotoxic T-cell density on whole resection specimens (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB). Using an illustrative cut-off of median sTIL CD8\u003csup\u003e+\u003c/sup\u003e density (median\u0026thinsp;~\u0026thinsp;125 cells/mm\u003csup\u003e2\u003c/sup\u003e) showed significantly poorer iDFS of patients receiving additive chemoendocrine therapy (p\u0026thinsp;=\u0026thinsp;0.0062, 15\u0026nbsp;year iDFS HTCT: 61.8% vs 89% HT-only, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). We crucially discovered a treatment-biomarker interaction, indicating poorer outcomes with chemotherapy as cytotoxic T-cell density increases (up to -30% survival difference at 15-years. ΔLR-χ\u003csup\u003e2\u003c/sup\u003e: 7.48, p\u0026thinsp;=\u0026thinsp;0.00623) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). Absolute survival differences across prespecified percentiles (10th -90th, Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE) showed a monotonic trend toward chemotherapy inferiority with high (\u0026gt;\u0026thinsp;50th percentile) sTIL CD8\u003csup\u003e+\u003c/sup\u003e density. At the 90th percentile of density, a 3.4% (95CI: 0.5\u0026ndash;14.7%, p\u0026thinsp;=\u0026thinsp;0.006, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05), 8.1% (95%CI: 1.7\u0026ndash;28.9%, p\u0026thinsp;=\u0026thinsp;0.004, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05), and 13.8% (95%CI: 3.0\u0026ndash;47.0%, p\u0026thinsp;=\u0026thinsp;0.004, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05) higher iDFS was observed at 5yrs, 10yrs, and 15yrs, respectively, for patients receiving endocrine therapy only (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE).\u003c/p\u003e\u003cp\u003eOver a 15 year horizon, the delta in restricted mean survival time was up to 12 months in favour of endocrine therapy-only (90th percentile, 95%CI: 2.16\u0026ndash;37.4, p\u0026thinsp;=\u0026thinsp;0.008, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eF). Decision curve analysis demonstrated that sTIL CD8\u003csup\u003e+\u003c/sup\u003e density provided higher net benefit than menopausal status across most thresholds. Particularly 0\u0026ndash;3% and 8\u0026ndash;30% (bootstrap p\u0026thinsp;\u0026lt;\u0026thinsp;0.004, FDR q\u0026thinsp;=\u0026thinsp;0.005. p\u0026thinsp;\u0026lt;\u0026thinsp;0.006, FDR q\u0026thinsp;=\u0026thinsp;0.062, respectively). To contextualise, a harm threshold of 15% indicates an intention to treat when the probability of recurrence is at least 15%, otherwise spare chemotherapy. Using sTIL CD8\u003csup\u003e+\u003c/sup\u003e density at this threshold yielded a net benefit corresponding to 6 more patients per 100 correctly managed over the 15 year period compared with menopausal status (95%CI: 2.65\u0026ndash;8.62, p\u0026thinsp;=\u0026thinsp;0.002, FDR q\u0026thinsp;=\u0026thinsp;0.0025). Compared with a chemo-for-all policy, sTIL CD8\u003csup\u003e+\u003c/sup\u003e density yielded 11 more per 100 correctly managed (95%CI: 7.25\u0026ndash;14.19, p\u0026thinsp;=\u0026thinsp;0.002, q\u0026thinsp;=\u0026thinsp;0.002). While these results require external validation to increase discriminatory power at lower percentiles, our data suggest that stromal cytotoxic T-cell density is a clinically translatable predictor that may help identify Intermediate RS patients (RS 16\u0026ndash;25) more likely to be harmed by, or less likely to benefit from, additive chemotherapy regimens. Crucially, this data is independent of menopausal status. If validated, this could result in up to 50% of all Intermediate RS patients (RS 16\u0026ndash;25) experiencing a change in adjuvant treatment from the existing paradigm.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eHere, we have demonstrated that immune phenotyping is significantly prognostic in ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e disease, but heavily dependent on the underlying genomic risk of the tumour. In whole-cohort assessments agnostic of genomic risk, TIL analysis produced results in agreement with the consensus view of low immunogenicity, poor prognostic dichotomisation, and weak, if no, ability to predict outcome in ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e breast cancer overall. However, as the underlying tumour begins to acquire features associated with a decreased reliance on estrogen-related pathways and increased reliance on proliferative mechanisms (i.e. increasing Oncotype Dx RS), the immune milieu changes significantly. We have detailed how immunosuppressive cell niches, associated with macrophage enrichment, become dominant in breast TMEs that are acquiring higher risk features, and importantly demonstrate their presence in a surprising proportion of lower-risk disease. Further, we have outlined how these environments improve the odds of cytotoxic T-cell entry into epithelia, but are simultaneously and significantly associated with immune dysfunction, checkpoint expression, and extracellular matrix remodelling, that may blunt the response of effective tumour killing. The presence of cytotoxic T-cells in higher risk diseases was a significant negative predictor of 15-yr invasive disease-free survival in the High RS. Lastly, using randomised treatment in the Intermediate RS, we show that sTIL CD8\u003csup\u003e+\u003c/sup\u003e density is a predictive marker for chemotherapy benefit across both TMA and orthogonal whole-resection specimens, above all clinical covariates, and suggests potential treatment changes for up to 50% of all patients with an Intermediate RS (RS 16\u0026ndash;25) regardless of their menopausal status.\u003c/p\u003e\u003cp\u003eWe firstly outlined how the immune milieu of Oncotype Dx RS categories differ by immunosuppressive cells. Macrophage\u003csup\u003eHIGH\u003c/sup\u003e TMEs were associated with significantly increased stromal and intraepithelial phenotypes, promoted the recruitment of these cells to the peri-tumoural stroma, and facilitated the odds of epithelial transition of stromal cytotoxic T-cells. Increasing bulk stromal macrophage density was significantly positively associated with MHC-I-like (β2M) and MHC-II-like antigen presentation genes (HLA-DRA, CD74) only in lower risk patients, suggesting a role of T-helper crosstalk. Indeed, these macrophage\u003csup\u003eHIGH\u003c/sup\u003e environments significantly recruited T-helper cells, forming a macrophage-mediated hub. T-helper distance to cytotoxic T-cells was positively associated with cytotoxicity genes, but also checkpoint genes in PD-1, PD-L2, and LAG3 irrespective of clinical risk (endocrine-only, chemoendocrine arms). These data imply that adaptive recruitment and attempted cytotoxicity is attenuated in suppressive M2-like macrophage\u003csup\u003eHIGH\u003c/sup\u003e niches. Single-cell omics of luminal breast cancers has recently shown that suppressive myeloid niches can attract T-cells, denoting an exhausted cluster with altered but not abolished cytotoxicity, that corresponds with spatial hubs of macrophage/T-helper/T-reg/Cytotoxic T-cells, and enhanced matrix metalloproteinase expression (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). We also recognised that bulk macrophage density was significantly associated with extracellular matrix remodelling genes (COL1A1, COL3A1, TIMP2, MMP2), and M2-like gene SPP1. Recent study by Cha \u003cem\u003eet al.\u003c/em\u003e has shown in HR\u003csup\u003e+\u003c/sup\u003e breast cancer an increased propensity of SPP1\u003csup\u003e+\u003c/sup\u003e macrophage to interact with T-helper and cytotoxic T-cells, most notable in TIL-high tumours, in which SPP1\u003csup\u003e+\u003c/sup\u003e macrophage inhibit T-cell responses to tumour through regulatory signalling and ECM formation (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). A principal immunosuppressive role of SPP1\u003csup\u003e+\u003c/sup\u003e macrophage has also been observed in ovarian cancer (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e), and fibrosis promotion via ECM remodelling by macrophage has been linked in breast cancer cell lines to dysfunctional CD8\u0026thinsp;+\u0026thinsp;T-cell responses, independent of checkpoint signalling, by both physical exclusion and metabolic reprogramming that impedes antitumour immunity (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe impact of an immunosuppressive milieu with reduced T-effector functionality was reflected in our finding that intraepithelial cytotoxic T-cell density is only significantly positively associated with the expression of antigen presentation genes in the epithelia. Paracrine IFN-γ secretion by T-cells in neighbouring immune-hot but suppressed peri-tumoural niches can induce MHC-II-like responses in epithelial cells (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e), but the lack of significant costimulation, cytotoxicity, or checkpoint genes as iTIL CD8 density rises is commensurate with a blunted cytotoxic response. We recognised also that the delta and number of antigen presentation genes associated with iTIL CD8 density is greatest in chemoendocrine-receiving patients, in which estrogen-related signalling is lowest. Recent study suggests that ERα expression is inversely proportional to IFN-γ signalling, and that estradiol downregulates MHC-II and CIITA (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). Taken together, canonically lower risk, ERα-dependent tumours (i.e. the Low RS) may facilitate immune suppression. As neoantigen load, proliferation, and aggressive behaviour increase (i.e. Oncotype Dx RS shifts to Intermediate and High RS), the TME supplants the role of suppressive regulator in an environment that would otherwise be primed for anti-tumour immunity. This may explain why improved 15-year survival was seen in Intermediate RS patients receiving endocrine therapy alone vs chemoendocrine regimens \u0026ndash; high sTIL CD8\u003csup\u003e+\u003c/sup\u003e density may be capable of functional cytotoxicity once the estrogen-mediated suppression is lifted by endocrine therapies (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e). More research is required to understand this effect in estrogen-driven breast cancers.\u003c/p\u003e\u003cp\u003eWith this in mind, we also recognised an association of immune proximities with checkpoint activity (LAG3, PDCD1 \u0026ndash; encoding PD-1, and PDCD1LG2 \u0026ndash; encoding PD-L2) and exhaustion (TOX). Recent study suggests Intermediate RS tumours contain exhausted T-cells with negative prognostic influence (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e), and that checkpoints such as PD-L1 is positively associated with Oncotype Dx RS (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e). More recently, larger studies have begun to demonstrate PD-L2 may actually be a more granular predictor of early relapse risk in ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e-\u003c/sup\u003e breast cancer (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e), an independent variable in multivariable analysis of those patients treated with chemoendocrine therapy. Our data across Oncotype Dx RS categories suggest that these patients contain a macrophage/T-helper/Cytotoxic T-cell spatial hub, containing greater abundances of transcript-level, predominantly PDCD1LG2 (PD-L2), LAG3, and PDCD1 (PD-1) expressing milieu. Patients with high cytotoxic T-cell densities having a paradoxically poorer invasive disease-free survival when receiving chemoendocrine therapy is a rationale for continued exploration of immune checkpoint blockade in early-stage ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e-\u003c/sup\u003e breast cancer. This is particularly pressing in Intermediate and High RS disease, where reliance on ERα and estrogen signalling is decreasing, especially considering the successes of recent trials with immune checkpoint blockade in the arena of high risk, later stage ER\u003csup\u003e+\u003c/sup\u003e disease (Checkmate 7FL: (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e), KEYNOTE 756: (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e), I-SPY2: (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e)). However, further work is required to investigate with proteomics analyses the potential presence and relationship of checkpoints beyond the transcriptome.\u003c/p\u003e\u003cp\u003eWe lastly show that cytotoxic T-cell density is a predictive marker in patients with an Intermediate RS. This is, to our knowledge, the first time that a predictive marker of chemotherapy inferiority has been shown for the Intermediate RS. While our data provide strong evidence of improper treatment with additive chemotherapy, it is not yet known whether patients require de-escalation or combination treatment with immunotherapies. On the one hand, chemotherapy administration in early-stage ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e-\u003c/sup\u003e disease may ablate existing adaptive responses (\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e), but a pre-treatment TME with increased immunosuppressive signalling will dampen the chemotherapy-induced stimulation of immunogenic cell death (\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e), and induce fibrotic responses through existing wound-healing macrophage (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e) that promotes disease recurrence. External validation of these findings in the larger TAILORx trial is necessary. While we show that up to 50% of all Intermediate RS (RS 16\u0026ndash;25) patients, irrespective of their menopausal status, may experience a change in treatment, a larger validation is required before clinical adoption can be seriously discussed.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003eStudy Design and TMA\u003c/h2\u003e\u003cp\u003eThe TAILORx Tissue Bank (CTRIAL-IE 12\u0026ndash;30, NCT02050750) is an exploratory, translational, non-interventional multicentre biobank sponsored by Cancer Trials Ireland that aims to identify potential biomarkers. Eligibility required prior registration with the TAILORx trial (CTRIAL-IE (ICORG) 06\u0026ndash;31, NCT00310180), participation in trial arms and having sufficient tumour material available for immunohistochemical staining. Other than the accrual of patients who were also accrued to the Eastern Cooperative Oncology Group trial TAILORx, there was no connection between the two studies, and the analysis performed on the patient samples from CTRIAL-IE 12\u0026ndash;30 did not impinge in any way on the TAILORx trial. RS values were taken as being 0\u0026ndash;15 for Low, 16\u0026ndash;25 for Intermediate, and 26\u0026ndash;100 for High, following from the recommendations of the TAILORx studies (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e). Cohort characteristics can be found in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The primary end-point of this study was invasive disease-free survival (iDFS), as per the STEEP criteria (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e), defined as the first invasive recurrence (distant, ipsilateral, locoregional), second primary invasive cancers, or death from any cause. There had been 108 iDFS events over a median follow-up of 158 months (SD\u0026thinsp;\u0026plusmn;\u0026thinsp;42mo), with 48 events in the Intermediate RS.\u003c/p\u003e\u003cp\u003eOf 577 patients entered for TMA construction, 109 had insufficient tumour content for TMA core sampling. Each core was sampled with a diameter of 1000\u0026micro;m and in triplicate per patient, leaving n\u0026thinsp;=\u0026thinsp;468 patients and n\u0026thinsp;=\u0026thinsp;1404 TMA total cores for spatial omics analyses. Due to too few cells or no cores, artefactual staining, sectioning, or imaging, and limitations of the GeoMx DSP instrument, a further 58 patients were dropped leaving 410 patients with whole-transcriptome spatial transcriptomics data. One core was selected per patient and segmented in the cytokeratin channel for epithelia vs microenvironment. Similarly, due to no tumour or too few cells, and artefacts present within cores, 26 patients were dropped leaving 442 patients for spatial proteomics analysis. Three cores were used per patient for proteomics analyses. Orthogonal validation was performed on n\u0026thinsp;=\u0026thinsp;453 whole-resection specimens taken from the source tumour blocks of the same patient material used to construct TMA cores. A more detailed consort diagram can be found in Supplementary Fig.\u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eSpatial Proteomics\u003c/h3\u003e\n\u003cp\u003emIF staining was carried out using sequential OPAL\u0026trade; tyramide signal amplification (TSA) multiplexing method on a Bond-RXm Automated Research Stainer (Leica Biosystems, Newcastle, UK) on tissues within one month of their sectioning to minimise epitope degradation. Primary antibodies were applied sequentially with heat-induced epitope retrieval times, staining conditions, and pH values optimised per antibody and OPAL\u0026trade; fluorophore combination. The staining order was applied (Supplementary Table\u0026nbsp;1.), and counterstained with spectral DAPI. Briefly, slides were blocked for 10min with 150\u0026micro;L of Akoya blocking buffer (#ARD1001EA, Akoya Biosciences, Menlo Park, CA) before incubating with 150\u0026micro;L antibody at room temperature (RT) for 30 mins. Slides were washed before applying 150\u0026micro;L rabbit or mouse linker (DAKO) for 20min at room temperature, where used. 150\u0026micro;L TSA-DIG was applied for 10 mins, when used. Linkers/ TSA-DIG steps were followed by a wash and 150\u0026micro;L of DAKO horseradish peroxidase for 20 mins, wash, and 150\u0026micro;L of the requisite OPAL\u0026trade; fluorophore for 30 mins, or 20 mins for OPAL\u0026trade; 780. Slides were then thoroughly washed and subjected to epitope retrieval for the subsequent antibody, for 20 mins at 95\u0026ordm;C, using either Leica Bond ER solution 1 (pH6) or solution 2 (pH9).\u003c/p\u003e\u003cp\u003eIHC staining (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.) was carried using DAKO EnVision FLEX kit (#K802321-1, Agilent Technologies, Stockport, UK) with a DAKO PT-Link and Link-48 system. Antigen retrieval was performed at 97\u0026ordm;C for 20 minutes at pH9 in EnVision FLEX high pH. Staining was performed in 48-slide batches with 5 minutes of FLEX peroxidase block, 20 minutes with CD8 (#IR62361-2, Agilent Technologies, Stockport, UK), and 5 minutes of FLEX DAB\u0026thinsp;+\u0026thinsp;Sub-Chromo for 5 minutes. Slides were washed with buffer for two cycles before FLEX haematoxylin was applied for 3 minutes, washed, and slides dehydrated at 60\u0026ordm;C for 1hr. 1mL drop of Sigma DPX was used for mounting and cover slipped. Tonsil tissue was used within each staining batch as a positive control to ensure intra-batch consistency.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eSpatial Transcriptomics\u003c/h2\u003e\u003cp\u003eA Nanostring GeoMx Morphology Marker Kit was stained for CD45, PanCK, and SYTO13 to FFPE-derived TMA tissues as outlined by the manufacturer, using a Leica Bond RX-m (Leica Biosystems, Newcastle UK). Antigen retrieval was performed in a Leica Bond RxM at 100\u0026ordm;C for 20mins in Tris-EDTA, with staining performed as per the manufacturer\u0026rsquo;s instruction (Nanostring GeoMx DSP Automated Slide Preparation User Manual). RNA targets were exposed by digestion in 0.1\u0026micro;g/ml proteinase k for 15 minutes. RNA \u003cem\u003ein-situ\u003c/em\u003e hybridisation was performed at room temperature for 16hrs. The Whole Transcriptome Atlas (WTA) was applied, with tumour vs microenvironment regions semantically segmented and RNA probes aspirated on the Nanostring Digital Spatial Profiler instrument (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e). One 660\u0026micro;m region was taken per patient for segmentation, using PanCK to semantically segment two Areas of Interest (AOIs): the tumour microenvironment from epithelia. Serial UV illumination of each compartment was used to sequentially collect mRNA probe barcodes from each segmented region. Collected probes were stored at -80\u0026ordm;C immediately after collection to preserve probes for sequencing. Library Preparation was performed at the Genomics Core Facility, Queen\u0026rsquo;s University of Belfast, Northern Ireland, according to MAN-10153-03 (Version Feb-2023). Resulting Sample Pools underwent AMPure Cleanup with KAPA Pure Beads (07983298001) followed by Quality Control checks with Qubit 1X dsDNA High Sensitivity Assay (Q33231) and Agilent Tapestation D1000 Assay (5067\u0026ndash;5582). Two Sequencing Pools were made based on total ROI area of the Sample Pools. These underwent further QC checks with Qubit 1X dsDNA High Sensitivity Assay (Q33231) and Fragment Analyser HS Fragment Analyser Assay (DNF-474) to determine Molarity. A Sequencing Depth Factor of 100 was chosen (for Whole Transcriptome Atlas/WTA). Each Sequencing Pool was run on the Illumina NovaSeq 6000 Sequencer (S4 200 Cycle v1.5 (20028313) and S2 100 Cycle v1.5 (20028316) carts utilised for the two different Sequencing Pools). The following Read Structure was utilised (Read 1\u0026ndash;27. Index Read 1\u0026ndash;8, Index Read 2\u0026ndash;8, Read 2\u0026ndash;27). Sequenced reads were obtained in FASTQ format and processed through the GeoMx NGS pipeline to generate raw gene counts. Raw counts were imported into R Studio for quality control via Nanostring\u0026rsquo;s GeoMxWorkflow pipeline. Limit of quantitation (LOQ) was defined as the geometric mean of the negative control probes multiplied by the geometric standard deviation. Targets consistently below the LOQ were excluded. For normalisation, upper quartile (Q3) normalisation was applied: briefly, the count in one segment was divided by the 3rd quartile value for that segment, and subsequently multiplied by the geometric mean of the 3rd quartile of all segments. Visualisation of gene expression data was performed on log2-transformed, Q3-normalised data (R packages: \u003cem\u003eggplot2\u003c/em\u003e and \u003cem\u003epheatmap\u003c/em\u003e). Differential expression analysis between groups was performed with LimmaVoom pipeline, and Linear Mixed Models (LMM). False discovery rate (FDR) was controlled using the Benjamini-Hochberg method, and genes with a FDR\u0026thinsp;\u0026le;\u0026thinsp;0.05 were identified as differentially expressed genes.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eImage Capture\u003c/h2\u003e\u003cp\u003eFluorescent staining was captured using the Akoya PhenoImager\u0026trade; HT Automated Quantitative Pathology Imaging System (Akoya Biosciences, Menlo Park, CA) at 20X magnification. Fluorescent channels were optimised using a spectral library of breast tissues stained single-plex for each of the fluorophores and subjected to the outlined protocol (antigen retrieval times and pH conditions equal to their order in the staining stack). The spectral library was constructed to minimise both spectral bleed-through and tissue autofluorescence, and to maximise the captured signal:noise ratio of each respective marker during imaging. The final resolution of captured image scans was 0.4992\u0026micro;m/px. Chromogenic IHC on WSIs were captured using a Leica Aperio AT2 (Leica Biosystems, Newcastle, UK) at 20x magnification, with a final resolution of 0.5025 \u0026micro;m/px.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eDigital Pathology and Phenotype Scoring\u003c/h2\u003e\u003cp\u003eAll digital image analysis (DIA) was undertaken in QuPath (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e). A bespoke script was utilised to stitch individual TMA core TIFF files to a singular pyramidal .ome.tif image (code available via GitHub: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gist.github.com/coltegelston/7c5ef58b32dd3a1b6a3f34926f01d6b2\u003c/span\u003e\u003cspan address=\"https://gist.github.com/coltegelston/7c5ef58b32dd3a1b6a3f34926f01d6b2\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Nuclear detection in the DAPI channel was performed using the StarDist (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e) QuPath plugin, or watershed cell detection in the haematoxylin channel (Supplementary Table S4-S5). Object detection classifiers were trained for individual markers using a Random Forest (\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e) classifier within QuPath with default features. A Random Forest pixel-based classifier was trained to distinguish macrophage due to their irregular shape, which made object detection-based methods perform sub-optimally, and used to generate detections with a filtering step by a minimum size of 15\u0026micro;m\u003csup\u003e2\u003c/sup\u003e and maximum 150\u0026micro;m\u003csup\u003e2\u003c/sup\u003e. These thresholds were chosen as they performed best in separating individual macrophage instances while reducing overcalling from spectral bleed-through, seen with the PanCK (Opal 690) channel. For separation of the TME compartments into epithelia versus microenvironment, a random forest pixel classifier was also trained with default features in the cytokeratin channel (mIF) or haematoxylin (IHC) and used to generate annotation masks of epithelia and TME. Subsequently, combination classifiers were constructed from single-marker object classifiers to discern all cells, including double-positive scoring such as CD4\u003csup\u003e+\u003c/sup\u003eFOXP3\u003csup\u003e+\u003c/sup\u003e T-regs or interacting cells, and resolved hierarchically so that detections were binned into either epithelia or microenvironment annotation mask. This method afforded further granularity in analysing whether TILs were intratumoural (within epithelial compartment) or stromal (within microenvironment). Detected and phenotyped objects were exported and compared core-core, per patient, demonstrating strong intrapatient correlation (Supplementary Fig.S2.). Artefactual cores (e.g. shearing, staining, blur, core translocation) and those with \u0026lt;\u0026thinsp;100 cells were omitted from analysis, whereas artefact was manually annotated out of the analysis area when these occurred in whole-slide images. In order to generate a singular score per patient, the mean percentage of each cell phenotype was taken across available cores and used as a singular TIL score. Overall TIL scores were calculated using all detected objects within both tumour epithelia and microenvironment masks. Microenvironmental (sTIL) and epithelial (iTIL) scores were calculated similarly, using the detected cells within each segment respectively. Density was calculated using count/mm\u003csup\u003e2\u003c/sup\u003e of the annotation mask. All immune scores across mIF and IHC can be found in Supplementary Table S2. Spatial analyses were performed using \u0026ldquo;Detection centroid distances 2D\u0026rdquo; script commands within QuPath. This command computes the 2D Euclidean distance (\u0026micro;m) between the centroid of a cell of a source class to the nearest centroid of a target class. These data were exported and analysed further in R Studio, summarised as the median distance per core. Examples of QuPath classifier generation for spatial proteomics and orthogonal validation can be seen in Supplementary Fig.S3-S4. All Ki67 values were calculated as outlined previously (\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analyses\u003c/h2\u003e\u003cp\u003eNormality of data distribution was assessed using the Kolmogorov-Smirnov test with Lilliefors correction (R package \u003cem\u003enortest\u003c/em\u003e). Differences in values between two categories was assessed using pairwise Mann-Whitney or Student\u0026rsquo;s t-test (R package \u003cem\u003estats\u003c/em\u003e). Multiple sample overall difference was assessed using the Kruskal-Wallis or ANOVA tests (R package \u003cem\u003estats, FSA\u003c/em\u003e), and either Dunn\u0026rsquo;s test with a correction for multiple testing by the Benjamini-Hochberg method (flagged at FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05), or Tukey\u0026rsquo;s test was used to examine pairwise differences. Considering the large sample size for immune distances (Total N\u003csub\u003ephenotypes\u003c/sub\u003e = 4,700,000) and how small deviations in the distribution at these sample sizes would affect calculations of pairwise statistical difference, Cohen\u0026rsquo;s \u003cem\u003ed\u003c/em\u003e was computed (R package \u003cem\u003eeffectsize\u003c/em\u003e) to examine the magnitude of the effect size of immune distances and classified as negligible (\u0026le;\u0026thinsp;0.2), or non-negligible (\u0026gt;\u0026thinsp;0.2). Drop-out and missingness analysis were assessed at the patient level within the Intermediate RS, comparing baseline covariates between included/excluded cases for each omics dataset. For categorical variables, Pearson\u0026rsquo;s χ\u003csup\u003e2\u003c/sup\u003e test was used for frequencies\u0026thinsp;\u0026gt;\u0026thinsp;5, otherwise Fisher\u0026rsquo;s exact test was used (two-sided) (R package \u003cem\u003estats\u003c/em\u003e). For continuous variables, Wilcoxon rank-sum (two-sided) was used. Within each modality, comparisons were adjusted for multiple comparisons using the Benjamini-Hochberg method (FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003cp\u003eTo investigate which, if any, phenotype may facilitate or impede cytotoxic T-cell epithelial entry as stromal cytotoxic T-cell density increases, we fit patient-level binomial logistic regression models with a logit link (R package \u003cem\u003estats\u003c/em\u003e) separately by treatment arm (endocrine-only, chemoendocrine). The response was supplied with \u003cem\u003en\u003c/em\u003e\u003csub\u003ein\u003c/sub\u003e and \u003cem\u003en\u003c/em\u003e\u003csub\u003eout\u003c/sub\u003e, corresponding to counts of CD8 cells inside epithelium and in stroma, respectively. Each model included stromal CD8 density (sCD8), a candidate predictor (densities and proximities) and their interaction, plus a uniform adjustment set of stromal densities (all except sCD8). The interaction tests whether, for the same\u0026thinsp;+\u0026thinsp;1SD increase in stromal CD8 density, a higher level of the predictor multiplies the odds that a CD8 cell is iTIL rather than sTIL. An interaction OR\u0026thinsp;\u0026gt;\u0026thinsp;1 indicates facilitation, OR\u0026thinsp;\u0026lt;\u0026thinsp;1 indicates impediment. When the predictor was itself a density, it was omitted to prevent overadjustment. Oncotype Dx RS was included as a factor. Densities were log1p-transformed and z-scored. Distances were z-scored and sign flipped so that larger values corresponded with closer proximity. Both sCD8 and the predictors were Winsorized at 1% tails to limit outlier leverage. Inference used robust Wald statistics with covariate estimates (R package \u003cem\u003esandwich\u003c/em\u003e, \u003cem\u003ecoeftest\u003c/em\u003e). The interaction term was summarised as an odds ratio (OR) with robust 95%CI. Multiple testing was controlled within each treatment arm using the Benjamini-Hochberg method (flagged at a discovery FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1). Tautological proximity pairs (e.g. CD20-CD20) and variables with n\u0026thinsp;\u0026le;\u0026thinsp;100 patients per arm were excluded.\u003c/p\u003e\u003cp\u003eTo relate immune densities or spatial proximity to functional or dysfunctional immune activity, we modelled gene expression modules against stromal immune densities, spatial proximities, and genomic risk. Predefined gene modules were derived by aggregating the mean expression of curated gene sets, for: antigen presentation (HLA-DRA, HLA-DRB1, CIITA, CD74, β2M), cytotoxicity (GZMA, GZMB, PRF1, IFNG), exhaustion (PDCD1, TOX, CXCL13, HAVCR2, TIGIT, LAG3), checkpoint inhibition (PD-L1, PD-L2, CTLA4, LAG3, TIGIT, VSIR), M1-like macrophage (CXCL9, NOS2, TNF, CD80, CD86, IL1B), M2-like macrophage (SPP1, CD163, TGFB1, IL10), and extracellular matrix (COL1A1, COL3A1, TIMP1, TIMP2, MMP2, MMP9). Module scores were z-standardised within the stromal compartment. Linear regression models (R package \u003cem\u003estats\u003c/em\u003e) were fitted separately for each module, to include immune phenotype count and density, pairwise spatial proximity, and Oncotype Dx RS as predictors. Densities were log-transformed and z-scored. Distances were z-scored and negated so that larger values represented greater proximity. For each model, the regression coefficient with 95% CI were estimated (R package \u003cem\u003ebroom\u003c/em\u003e) and p-values corrected for multiple testing using the Benjamini-Hochberg (BH) method (flagged at discovery FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1). In order to quantify unique variance explained by each predictor, semi-partial R\u003csup\u003e2\u003c/sup\u003e values were calculated using bootstrap resampling (R package \u003cem\u003eboot\u003c/em\u003e, B\u0026thinsp;=\u0026thinsp;5000 resamples). The Delta-R\u003csup\u003e2\u003c/sup\u003e (variance loss) across resamples quantified the unique contribution of each predictor.\u003c/p\u003e\u003cp\u003eTo investigate whether macrophage density, cytotoxic T-cell density, or T-helper proximity to cytotoxic T-cells was related to immune gene activity, we modelled individual gene z-scores from each curated gene module above, as a function of density (cells/mm\u003csup\u003e2\u003c/sup\u003e) or distance (\u0026micro;m). Densities were log-transformed, and z-score normalised within the stroma (macrophage) or epithelia (cytotoxic T-cell). Ordinary Least Squares (OLS) linear models (R package \u003cem\u003estats\u003c/em\u003e) were fitted, with cluster robust standard errors (R package \u003cem\u003esandwich)\u003c/em\u003e, stratified by either treatment category (endocrine-only, chemoendocrine) or Oncotype Dx RS (Low, Intermediate, High). Models were adjusted for densities and distances, with group-specific slopes estimated (R package \u003cem\u003eemmeans)\u003c/em\u003e to give effect sizes per +\u0026thinsp;1SD change in the variable. Slopes were back-transformed to enable interpretation, where changes in gene expression z-score could be viewed as intuitive steps in the predictor (Density: +50 cells/mm\u003csup\u003e2\u003c/sup\u003e. Distance: -20\u0026micro;m proximity). Multiple testing was controlled using the Benjamini-Hochberg method (flagged at discovery FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eSurvival Analyses\u003c/h2\u003e\u003cp\u003eSurvival analyses were performed via the Kaplan-Meier method (R package \u003cem\u003esurvival\u003c/em\u003e), and Likelihood ratio-χ\u003csup\u003e2\u003c/sup\u003e tests and Harrell\u0026rsquo;s C-index were obtained from nested uni- and multivariable Cox models. All statistical tests were two sided, with α\u0026thinsp;\u0026lt;\u0026thinsp;0.05 considered statistically significant. Asterisk values of significance are shown as follows: ns p\u0026thinsp;\u0026gt;\u0026thinsp;0.05, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 (*), 0.01\u0026thinsp;\u0026ge;\u0026thinsp;p\u0026thinsp;\u0026gt;\u0026thinsp;0.001 (**), and 0.001\u0026thinsp;\u0026ge;\u0026thinsp;p (***). For all models the proportional hazard assumption was verified by cox.zph tests (R package \u003cem\u003esurvival.\u003c/em\u003e Supplementary Table S6-S9). To assess chemotherapy inferiority/superiority, we fit multivariable Cox models containing an interaction between adjuvant treatment (HT vs HT\u0026thinsp;+\u0026thinsp;CT) and the continuous immune variable (counts/densities), adjusting for clinical covariates (menopausal status, age\u0026thinsp;\u0026lt;\u0026thinsp;65 vs\u0026thinsp;\u0026ge;\u0026thinsp;65, tumour size, histological grade, luminal subtype). Non-linearity was explored with restricted cubic splines (3 knots at Harrell\u0026rsquo;s defaults of 10th, 50th, 90th percentile) using rms, with 95% CIs from model standard errors (\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eAbsolute survival differences (ARDs) were computed by standardization (R package \u003cem\u003esurvival\u003c/em\u003e). From the fitted Cox model with a continuous treatment*biomarker interaction, we generated model-based survival curves with covariates fixed at typical values (numerics at the cohort median; factors at the modal level) and baseline stromal CD8 density set to prespecified levels (percentiles, median). ARD was calculated at 5, 10, and 15 years where positive values favour endocrine therapy above chemoendocrine. Uncertainty was quantified with patient-level bootstrap resampling (1,000 resamples) with percentile 95% CIs. Stromal CD8 density was modelled on the z-scale for estimation and translated to cells/mm\u0026sup2; for interpretation and display.\u003c/p\u003e\u003cp\u003eRestricted mean survival time (RMST) was used to summarise treatment effects over the follow-up period (R package \u003cem\u003esurvival\u003c/em\u003e). For a timepoint of 15 years, RMST equals the area under the survival curve up to 15-years. Delta-RMST was defined as:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{\\Delta\\:}RMST\\left(15yrs\\right)=RMS{T}_{HT}-{RMST}_{HT}+CT\\:$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eRMSTs were obtained by integrating the model-based survival functions to 15 years (from \u003cem\u003esurvfit)\u003c/em\u003e with bootstrap 95% CIs (1,000 resamples).\u003c/p\u003e\u003cp\u003eDecision-curve analysis (DCA) was performed with a benefit-based variant (\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e). Clinical utility of treatment selection in the randomised arms of the Intermediate RS was evaluated with DCA at the 15-year timepoint (R package \u003cem\u003edcurves\u003c/em\u003e). For each patient we estimated counterfactual 15-year absolute risks, under endocrine therapy alone and chemoendocrine therapy, from Cox proportional hazard models. From these, we defined chemotherapy harm at 15-years as the absolute risk increase:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{\\Delta\\:}Risk=P\\left(event\\:\\right|\\:HT+CT)-P\\left(event\\:\\right|\\:HT)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eSo that positive values indicated chemotherapy inferiority. Absolute risks were obtained from each Cox model via the baseline cumulative hazard and individual linear predictors. We considered harm thresholds \u003cem\u003et\u003c/em\u003e ranging from 0 to 30% (in 1% steps). For each threshold, we evaluated a chemotherapy sparing policy: withhold chemotherapy if \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\Delta\\:}Risk\\)\u003c/span\u003e\u003c/span\u003e \u0026ge; \u003cem\u003et.\u003c/em\u003e For a given \u003cem\u003et\u003c/em\u003e, net benefit was computed per-patient and reported per 100 patients:\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:Net\\:benefit=\\left({Risk}_{HT}-{Risk}_{Policy}\\right)-\\:\\frac{t}{1-t}\\:\\times\\:\\:P\\left(chemo\\:under\\:policy\\left(t\\right)\\right),$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{Risk}_{HT}\\)\u003c/span\u003e\u003c/span\u003e is the mean 15-year risk if all patients received endocrine therapy only, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{Risk}_{Policy}\\)\u003c/span\u003e\u003c/span\u003e is the mean 15-year risk under the treatment allocation rule (e.g. menopausal status toward endocrine-only or chemoendocrine, etc.). Net benefit was plotted for the biomarker policy (treatment*sTIL CD8 density), menopausal status policy, and endocrine therapy for all, chemoendocrine for all. We obtained 95% bootstrap percentile intervals by resampling over 1,000 replicates, recomputing net benefit over each replicate. Plots show median net benefit and 95% intervals.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements/Funding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank Dr Heiko Dussman for assistance with the operation of the Bruker NanoString GeoMx platform; Dr Philip Schouten, Dr Elena Provezano, and Dr Aris Sionakidis for their help in reviewing and copy editing manuscripts.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eZK was supported by the SFI Strategic Partnership \u0026ldquo;Precision Oncology Ireland\u0026rdquo; grant# 18/SPP/3522. DOC was supported by 18/SPP/3522 and the Irish Cancer Society Collaborative Cancer Research Centre \u0026ldquo;Breast-Predict\u0026rdquo; grant# CCRC13GAL. HN was supported by the SFI Centre for Research Training in Genomics Data Science and the Royal College of Surgeons in Ireland grant# 18/CRT/6214-RCSI-DGF-2022 and DKL was supported by the SFI Centre for Research Training in Genomics Data Science Grant# 18/CRT/6214. AC was supported by Grant# 18/SPP/3522. WMG was supported by SFI Grant# 18/SPP/3522, Irish Cancer Society Grant# CCRC13GAL as well as Science Foundation Ireland (SFI) under the Investigator Programme OPTi-PREDICT Grant# 15/IA/3104. JHMP was supported by Research Ireland grants 18/RI/5792 and 21/RI/9787.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest: None to report.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData Access: where appropriate, source data files will be provided with the article. De-identified data collected in the Irish TAILORx study will be made available to researchers if access and use of the data has been approved by the respective trial management group.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eDenkert C, Loibl S, Noske A, Roller M, M\u0026uuml;ller BM, Komor M et al (2010) Tumor-associated lymphocytes as an independent predictor of response to neoadjuvant chemotherapy in breast cancer. J Clin Oncol 28(1):105\u0026ndash;113\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLoi S, Michiels S, Salgado R, Sirtaine N, Jose V, Fumagalli D et al (2014) Tumor infiltrating lymphocytes are prognostic in triple negative breast cancer and predictive for trastuzumab benefit in early breast cancer: results from the FinHER trial. Ann Oncol 25(8):1544\u0026ndash;1550\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalgado R, Denkert C, Demaria S, Sirtaine N, Klauschen F, Pruneri G et al (2015) The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol 26(2):259\u0026ndash;271\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWagner J, Rapsomaniki MA, Chevrier S, Anzeneder T, Langwieder C, Dykgers A et al (2019) A Single-Cell Atlas of the Tumor and Immune Ecosystem of Human Breast Cancer. Cell 177(5):1330\u0026ndash;45e18\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCurtis C, Shah SP, Chin S-F, Turashvili G, Rueda OM, Dunning MJ et al (2012) The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486(7403):346\u0026ndash;352\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAli HR, Jackson HW, Zanotelli VRT, Danenberg E, Fischer JR, Bardwell H et al (2020) Imaging mass cytometry and multiplatform genomics define the phenogenomic landscape of breast cancer. Nat Cancer 1(2):163\u0026ndash;175\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDanenberg E, Bardwell H, Zanotelli VRT, Provenzano E, Chin S-F, Rueda OM et al (2022) Breast tumor microenvironment structures are associated with genomic features and clinical outcome. Nat Genet 54(5):660\u0026ndash;669\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLoi S, Salgado R, Curigliano G, Romero D\u0026iacute;az RI, Delaloge S, Rojas Garc\u0026iacute;a CI et al (2025) Neoadjuvant nivolumab and chemotherapy in early estrogen receptor-positive breast cancer: a randomized phase 3 trial. Nat Med 31(2):433\u0026ndash;441\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR\u0026iacute;os-Hoyo A, Cobain E, Huppert LA, Beitsch PD, Buchholz TA, Esserman L et al (2024) Neoadjuvant Chemotherapy and Immunotherapy for Estrogen Receptor\u0026ndash;Positive Human Epidermal Growth Factor 2\u0026ndash;Negative Breast Cancer. J Clin Oncol 42(22):2632\u0026ndash;2636\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCardoso F, O\u0026rsquo;Shaughnessy J, Liu Z, McArthur H, Schmid P, Cortes J et al (2025) Pembrolizumab and chemotherapy in high-risk, early-stage, ER+/HER2\u0026thinsp;\u0026ndash;\u0026thinsp;breast cancer: a randomized phase 3 trial. Nat Med 31(2):442\u0026ndash;448\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAli HR, Provenzano E, Dawson SJ, Blows FM, Liu B, Shah M et al (2014) Association between CD8\u0026thinsp;+\u0026thinsp;T-cell infiltration and breast cancer survival in 12 439 patients. Ann Oncol 25(8):1536\u0026ndash;1543\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu S, Foulkes WD, Leung S, Gao D, Lau S, Kos Z et al (2014) Prognostic significance of FOXP3\u0026thinsp;+\u0026thinsp;tumor-infiltrating lymphocytes in breast cancer depends on estrogen receptor and human epidermal growth factor receptor-2 expression status and concurrent cytotoxic T-cell infiltration. Breast Cancer Res 16(5):432\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang Y, Ma C, Zhang Q, Ye J, Wang F, Zhang Y et al (2015) CD4\u0026thinsp;+\u0026thinsp;and CD8\u0026thinsp;+\u0026thinsp;T cells have opposing roles in breast cancer progression and outcome. Oncotarget 6(19):17462\u0026ndash;17478\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShang B, Liu Y, Jiang S-j, Liu Y (2015) Prognostic value of tumor-infiltrating FoxP3\u0026thinsp;+\u0026thinsp;regulatory T cells in cancers: a systematic review and meta-analysis. Sci Rep 5(1):15179\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOnkar S, Cui J, Zou J, Cardello C, Cillo AR, Uddin MR et al (2023) Immune landscape in invasive ductal and lobular breast cancer reveals a divergent macrophage-driven microenvironment. Nat Cancer 4(4):516\u0026ndash;534\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eO\u0026rsquo;Meara T, Marczyk M, Qing T, Yaghoobi V, Blenman K, Cole K et al (2020) Immunological Differences Between Immune-Rich Estrogen Receptor\u0026ndash;Positive and Immune-Rich Triple-Negative Breast Cancers. JCO Precision Oncol. (4):767\u0026ndash;779\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNederlof I, Hajizadeh S, Sobhani F, Raza SEA, AbdulJabbar K, Harkes R et al (2022) Spatial interplay of lymphocytes and fibroblasts in estrogen receptor-positive HER2-negative breast cancer. npj Breast Cancer 8(1):56\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCardoso F, van't Veer LJ, Bogaerts J, Slaets L, Viale G, Delaloge S et al (2016) 70-Gene Signature as an Aid to Treatment Decisions in Early-Stage Breast Cancer. N Engl J Med 375(8):717\u0026ndash;729\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF et al (2018) Adjuvant Chemotherapy Guided by a 21-Gene Expression Assay in Breast Cancer. N Engl J Med 379(2):111\u0026ndash;121\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStein RC, Dunn JA, Bartlett JM, Campbell AF, Marshall A, Hall P et al (2016) OPTIMA prelim: a randomised feasibility study of personalised care in the treatment of women with early breast cancer. Health Technol Assess 20(10):xxiii\u0026ndash;xxix\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eO'Brien MER, Borthwick A, Rigg A, Leary A, Assersohn L, Last K et al (2006) Mortality within 30 days of chemotherapy: a clinical governance benchmarking issue for oncology patients. Br J Cancer 95(12):1632\u0026ndash;1636\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Dorp W, Haupt R, Anderson RA, Mulder RL, van den Heuvel-Eibrink MM, van Dulmen-den Broeder E et al (2018) Reproductive Function and Outcomes in Female Survivors of Childhood, Adolescent, and Young Adult Cancer: A Review. J Clin Oncol 36(21):2169\u0026ndash;2180\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRegan MM, Fleming GF, Walley B, Francis PA, Pagani O (2019) Adjuvant Systemic Treatment of Premenopausal Women With Hormone Receptor-Positive Early Breast Cancer: Lights and Shadows. J Clin Oncol 37(11):862\u0026ndash;866\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVaz-Luis I, Francis PA, Di Meglio A, Stearns V (2021) Challenges in Adjuvant Therapy for Premenopausal Women Diagnosed With Luminal Breast Cancers. Am Soc Clin Oncol Educ Book 41:1\u0026ndash;15\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTietscher S, Wagner J, Anzeneder T, Langwieder C, Rees M, Sobottka B et al (2023) A comprehensive single-cell map of T cell exhaustion-associated immune environments in human breast cancer. Nat Commun 14(1):98\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCha SM, Park J-W, Lee YJ, Lee HJ, Lee H, Lee IW et al (2024) SPP1\u0026thinsp;+\u0026thinsp;macrophages in HR\u0026thinsp;+\u0026thinsp;breast cancer are associated with tumor-infiltrating lymphocytes. npj Breast Cancer 10(1):83\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHou L, Jiang M, Li Y, Cheng J, Liu F, Han X et al (2025) Targeting SPP1\u0026thinsp;+\u0026thinsp;macrophages via the SPP1-CD44 axis reveals a key mechanism of immune suppression and tumor progression in ovarian cancer. Int Immunopharmacol 160:114906\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTharp KM, Kersten K, Maller O, Timblin GA, Stashko C, Canale FP et al (2024) Tumor-associated macrophages restrict CD8\u0026thinsp;+\u0026thinsp;T cell function through collagen deposition and metabolic reprogramming of the breast cancer microenvironment. Nat Cancer 5(7):1045\u0026ndash;1062\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAxelrod ML, Cook RS, Johnson DB, Balko JM (2019) Biological Consequences of MHC-II Expression by Tumor Cells in Cancer. Clin Cancer Res 25(8):2392\u0026ndash;2402\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMostafa AA, Codner D, Hirasawa K, Komatsu Y, Young MN, Steimle V et al (2014) Activation of ERα Signaling Differentially Modulates IFN-γ Induced HLA-Class II Expression in Breast Cancer Cells. PLoS ONE 9(1):e87377\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang H, Zhou J, Chen H, Li J, Zhang C, Jiang X et al (2021) The immunomodulatory effects of endocrine therapy in breast cancer. J Experimental Clin Cancer Res 40(1):19\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEgelston CA, Guo W, Tan J, Avalos C, Simons DL, Lim MH et al (2022) Tumor-infiltrating exhausted CD8\u0026thinsp;+\u0026thinsp;T cells dictate reduced survival in premenopausal estrogen receptor\u0026ndash;positive breast cancer. JCI Insight. ;7(3)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRozenblit M, Blenman K, Harigopal M, Reisenbichler E, Singh K, Qing T et al (2022) PD-L1 protein expression in relation to recurrence score values in early-stage ER\u0026thinsp;+\u0026thinsp;breast cancer. Breast Cancer Res Treat\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChervoneva I, Peck AR, Sun Y, Yi M, Udhane SS, Langenheim JF et al (2023) High PD-L2 Predicts Early Recurrence of ER-Positive Breast Cancer. JCO Precision Oncol. (7):e2100498\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCardoso F, McArthur HL, Schmid P, Cort\u0026eacute;s J, Harbeck N, Telli ML, et al. LBA21 KEYNOTE-756:Phase III study of neoadjuvant pembrolizumab (pembro) or placebo (pbo)\u0026thinsp;+\u0026thinsp;chemotherapy(chemo), followed by adjuvant pembro or pbo\u0026thinsp;+\u0026thinsp;endocrine therapy (ET) for early-stage high-risk ER+/HER2\u0026ndash; breast cancer. Annals of Oncology. 2023;34:S1260-S1\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNanda R, Liu MC, Yau C, Shatsky R, Pusztai L, Wallace A et al (2020) Effect of Pembrolizumab Plus Neoadjuvant Chemotherapy on Pathologic Complete Response in Women With Early-Stage Breast Cancer: An Analysis of the Ongoing Phase 2 Adaptively Randomized I-SPY2 Trial. JAMA Oncol 6(5):676\u0026ndash;684\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWolf DM, Yau C, Campbell M, Glas A, Barcaru A, Mittempergher L et al (2025) Immune Subtyping Identifies Patients With Hormone Receptor\u0026ndash;Positive Early-Stage Breast Cancer Who Respond to Neoadjuvant Immunotherapy (IO): Results From Five IO Arms of the I-SPY2 Trial. JCO Precision Oncol. (9):e2400776\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDixon-Douglas J, Virassamy B, Clarke K, Hun M, Luen SJ, Savas P et al (2024) Sustained lymphocyte decreases after treatment for early breast cancer. npj Breast Cancer 10(1):94\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark YH, Lal S, Lee JE, Choi Y-L, Wen J, Ram S et al (2020) Chemotherapy induces dynamic immune responses in breast cancers that impact treatment outcome. Nat Commun 11(1):6175\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMichaud M, Martins I, Sukkurwala AQ, Adjemian S, Ma Y, Pellegatti P et al (2011) Autophagy-Dependent Anticancer Immune Responses Induced by Chemotherapeutic Agents in Mice. Science 334(6062):1573\u0026ndash;1577\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGalluzzi L, Humeau J, Buqu\u0026eacute; A, Zitvogel L, Kroemer G (2020) Immunostimulation with chemotherapy in the era of immune checkpoint inhibitors. Nat Reviews Clin Oncol 17(12):725\u0026ndash;741\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMonteran L, Ershaid N, Doron H, Zait Y, Scharff Ye, Ben-Yosef S et al (2022) Chemotherapy-induced complement signaling modulates immunosuppression and metastatic relapse in breast cancer. Nat Commun 13(1):5797\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF et al (2015) Prospective Validation of a 21-Gene Expression Assay in Breast Cancer. N Engl J Med 373(21):2005\u0026ndash;2014\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSparano JA (2018) Prognostic gene expression assays in breast cancer: are two better than one? npj Breast Cancer 4(1):11\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTolaney SM, Garrett-Mayer E, White J, Blinder VS, Foster JC, Amiri-Kordestani L et al (2021) Updated Standardized Definitions for Efficacy End Points (STEEP) in Adjuvant Breast Cancer Clinical Trials: STEEP Version 2.0. J Clin Oncol 39(24):2720\u0026ndash;2731\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMerritt CR, Ong GT, Church SE, Barker K, Danaher P, Geiss G et al (2020) Multiplex digital spatial profiling of proteins and RNA in fixed tissue. Nat Biotechnol 38(5):586\u0026ndash;599\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBankhead P, Loughrey MB, Fern\u0026aacute;ndez JA, Dombrowski Y, McArt DG, Dunne PD et al (2017) QuPath: Open source software for digital pathology image analysis. Sci Rep 7(1):16878\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchmidt U, Weigert M, Broaddus C, Myers G (eds) Cell Detection with Star-Convex Polygons. Medical Image Computing and Computer Assisted Intervention \u0026ndash; MICCAI 2018; 2018 2018//; Cham: Springer International Publishing\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman L (2001) Random Forests. Mach Learn 45(1):5\u0026ndash;32\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLynch SM, Russell NM, Barron S, Wang C-JA, Loughman T, Dynoodt P, et al. Prognostic value of the 6-gene OncoMasTR test in hormone receptor\u0026ndash;positive HER2-negative early-stage breast cancer: Comparative analysis with standard clinicopathological factors. European Journal of Cancer. 2021;152:78\u0026ndash;89\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGauthier J, Wu QV, Gooley TA (2020) Cubic splines to model relationships between continuous variables and outcomes: a guide for clinicians. Bone Marrow Transplant 55(4):675\u0026ndash;680\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVickers AJ, van Calster B, Steyerberg EW (2019) A simple, step-by-step guide to interpreting decision curve analysis. Diagn Prognostic Res 3(1):18\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7726652/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7726652/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDespite a strong 5\u0026ndash;10 year prognosis, in ER\u003csup\u003e+\u003c/sup\u003eHER2\u003csup\u003e\u0026minus;\u003c/sup\u003e breast cancer relapses are common beyond 10 years. Genomic assays employed to determine recurrence risk at diagnosis are still ambiguous for a significant proportion of patients with Intermediate risk (i.e. Oncotype Dx Recurrence Score (RS) 16\u0026ndash;25). We demonstrate that macrophage and T-helper cells facilitate and impede cytotoxic T-cells, are associated with extracellular matrix remodelling and M2-like (SPP1) genes, and cytotoxicity (GZMA, GZMB, PRF1), checkpoints (LAG3, PD-1, PD-L2), and exhaustion (TOX) genes respectively, consistent with restrained T-cell activation. High cytotoxic T-cell density was associated with poorer 15-year iDFS in the High RS (p\u0026thinsp;=\u0026thinsp;0.017, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.1). Critically, in the randomised Intermediate RS, a treatment-biomarker interaction indicated chemotherapy inferiority at higher stromal CD8 density (ΔLR-χ\u003csup\u003e2\u003c/sup\u003e: 7.36, p\u0026thinsp;=\u0026thinsp;0.007), which was validated orthogonally on whole-resection specimens from the same cohort (ΔLR-χ\u003csup\u003e2\u003c/sup\u003e: 7.48, p\u0026thinsp;=\u0026thinsp;0.006). Using the candidate biomarker suggested a potential treatment change for up to 50% of the Intermediate RS.\u003c/p\u003e","manuscriptTitle":"Spatial-Immune Multi-omics Refines Prognostication in Early-Stage Estrogen Receptor-Positive Breast Cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-12 14:40:53","doi":"10.21203/rs.3.rs-7726652/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"10f51f7f-8622-4c18-8ede-47514c8b652f","owner":[],"postedDate":"October 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":55958424,"name":"Biological sciences/Cancer/Breast cancer"},{"id":55958425,"name":"Health sciences/Oncology/Cancer/Tumour biomarkers"},{"id":55958426,"name":"Biological sciences/Cancer/Cancer microenvironment"}],"tags":[],"updatedAt":"2026-05-07T08:13:14+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-12 14:40:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7726652","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7726652","identity":"rs-7726652","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}