Associations between tumor immune response and prognosis in node-negative breast cancer patients in the randomized DBCG HYPO trial.

Background
Radiotherapy remains a cornerstone in the adjuvant treatment of breast cancer, substantially reducing the risk of local recurrence and improving long-term survival [1, 2]. In recent years, hypofractionated regimens have become widely adopted for early-stage breast cancer, demonstrating equivalent efficacy and safety compared to the previous standard 50 Gy (Gy)/25 fractions (fr) [3–7].
Beyond direct cytotoxic effect, preclinical studies have shown that radiotherapy can modulate the tumor microenvironment and elicit immunogenic cell death, potentially promoting systemic anti-tumor immune responses [8, 9]. Preclinical studies have demonstrated that radiotherapy-induced immune activation is dependent on dose and fractionation. Hypofractionated regimens may enhance antigen presentation and T-cell priming more effectively than conventional schedules, likely due to increased rates of immunogenic cell death [10, 11]. These immunological effects of radiotherapy have led to increased interest in integrating immune biomarkers into radiotherapy planning [12].
Tumor-infiltrating lymphocytes (TILs), particularly CD8+ cytotoxic T-cells, have gained attention as a promising prognostic biomarker in breast cancer, especially in more aggressive tumor subtypes [13, 14], and hold predictive information on pathological complete response after neoadjuvant chemotherapy [13] and response to immunotherapy [15, 16]. Only few studies have described the predictive value of TILs in terms of benefit of radiotherapy, and in most studies of TILs and systemic therapy, information on radiotherapy administration is not provided. TILs have, however, been described in a cohort of node-positive patients from the Danish Breast Cancer Group (DBCG) internal mammary node (IMN)2 study treated with adjuvant locoregional radiotherapy, where high TIL levels were associated with improved prognosis, particularly in estrogen receptor-negative (ER-) tumors [17].
Furthermore, in the DBCG82bc cohort of node-positive breast cancer patients randomized to postmastectomy radiotherapy, TILs and CD8+ T-cells were not only prognostic but also predictive of radiotherapy benefit in terms of improved distant tumor control and superior survival, indicating a systemic effect of radiotherapy [18, 19]. On the contrary, in the SweBCG91RT trial of patients with node-negative disease, treated with lumpectomy and randomized ±radiotherapy, patients with low TILs and CD8+ T-cell levels in their primary tumor appeared to benefit from radiotherapy in terms of local control, while patients with high TILs and high CD8+ T-cell levels did not show benefit from adjuvant radiotherapy [20, 21]. The diverging results from clinical studies question whether the prognostic and predictive role of immune infiltrates differs according to nodal status, reflecting differences in underlying risk profiles.
This study aimed to investigate the prognostic and predictive value of TILs and immune cell subsets in a contemporary treated, randomized cohort of node-negative, breast cancer patients treated with locoregional adjuvant radiotherapy with different fractionation schedules. Specifically, we aimed to evaluate: (1) whether immune infiltration in primary tumors is prognostic for outcomes in irradiated, node-negative patients, and (2) whether immune markers predict benefit from hypofractionated radiotherapy. We hypothesized that immune infiltration is prognostic in irradiated, node-negative breast cancer patients, dependent on ER status, and may predict differential benefit from 40 Gy/15 fr versus 50 Gy/25 fr.

Materials and methods

Patients
Tumor tissue for the retrospective biomarker analysis originated from patients, who had all been participants in the DBCG HYPO trial (ClincalTrials.gov identifier: NCT00909818) [3]. Among 1882 patients from Denmark, Norway, and Germany, 1329 Danish patients were selected for the present study. Patients had T1a-T2, N0-1(mi) disease, treated with breast-conserving surgery, adjuvant systemic treatment according to DBCG guidelines, and randomized to postoperative 50 Gy/25 fr versus 40 Gy/15 fr during 2009–2014. The entire DBCG HYPO trial is reported elsewhere [3].

Study design
The present study applied a case-cohort design to evaluate associations between immune markers and clinical outcomes. Events were loco-regional recurrence (LRR), distant recurrence (DR), breast cancer-specific mortality (BCM), or overall mortality (OM). Cases comprised all patients experiencing at least one of the events, whereas controls were drawn from a representative, randomly selected subcohort of event-free patients (controls) [17, 22, 23]. The numbers of patients were estimated using probability weights referring back to the full cohort. This design reduced the number of patient samples that needed to be collected and analyzed to 349 out of the original 1329 patients (Supplementary Figure S1). Because separate endpoints required different samplings and probability weights, small variations may occur in the reported number of patients across subgroups and endpoints. Supplementary Figure S2 provides an overview of the endpoint-specific subcohorts in which tumor blocks were successfully available for histopathological analysis.

Histopathological analyses
Formalin-fixed, paraffin-embedded, treatment-naïve tumor samples from routine diagnostic procedures were collected. Stromal TILs were evaluated using Hematoxylin-Eosin (HE) stained sections performed on whole-slide sections and scored as semicontinuous variables using 5% intervals according to the International TILs Working Group guideline [24]. In 320 patients with available tumor blocks, tumors were categorized as high (≥30%) or low (<30%) TILs using a 30% cut-off value. This cut-off provides the most reliable interobserver agreement [19], and was used in other studies [14, 17, 19].
Areas with tumor epithelial cells and TILs were outlined on HE-sections. For 290 patients, a single 1.5 mm core was extracted from the annotated areas, targeting regions with representative immune infiltration, and embedded into a tissue microarray (TMA) using the TMA Grand Master (3DHISTECH Ltd., Budapest, Hungary) (Supplementary Figure S3). For eight patients lacking sufficient residual tumor for physical core extraction, virtual 1.5 mm TMA cores were delineated on scanned IHC-stained whole-slide sections using Visiopharm version 2023.09.5.15777 (Visiopharm A/S, Hørsholm, Denmark).
Multiplex immunohistochemistry (IHC) was performed on 4 μm TMA sections using antibodies targeting immune cells from the adaptive (CD8+ cytotoxic T-cell; CD4+ helper T-cell; FOXP3+ regulatory T-cells (Tregs)) and innate (CD68+ macrophages, CD11c+ dendritic cells (DCs)) immune system. FOXP3+ cells comprised both CD4+ and CD8+ T-cells, as well as cells lacking these markers. Pancytokeratin and myoepithelial marker p63 differentiated between invasive carcinoma, ductal carcinoma in situ (DCIS), and benign tissue structures. Staining protocols are detailed in Supplementary Tables S1-S2.
ER- and HER2-receptor status were obtained from routine diagnostic procedures. HER2 evaluation followed clinical guidelines at diagnosis [25–27]. ER positivity was defined as ≥1%. IHC-defined subtypes were constructed from ER and HER2 status.

Digital image analysis
TMA slides were scanned using NanoZoomer 2.0HT (Hamamatsu Photonics KK, Japan), aligned, and analyzed using Visiopharm. Automated digital image analysis was applied to scanned images of both physical and virtual cores to quantify single-marker expression, co-expression, and spatial distribution (Supplementary Figure S3). Immune cell detection was performed using convolutional neural networks (U-nets) trained with deep learning with manual slide-level annotations to recognize immune cells and remaining background. Regions of interest (ROIs) were automatically defined to capture invasive carcinoma while excluding DCIS and benign structures. Areas with necrosis or artefacts were manually removed from the analysis. For each ROI, cell-specific marker expression was measured as area fraction (µm2) within each ROI and classified into four spatial compartments: (1) tumor: tumor-epithelial islands, (2) tumor margin: interface between tumor epithelium and stroma, (3) stroma: interposed tumor-related stroma, or (4) total tumor area: the entire tumor area in the core. Area fractions for each marker (‘X’) in compartment (‘Y’) were calculated as:
Exploratory analysis determining optimal cut-offs, spatial compartments, and marker combinations have been described elsewhere (manuscript submitted).

Statistical analysis
Pearson correlation coefficients were used to assess pairwise relations between immune markers. Continuous area fractions were split into quartiles (Q1 = lowest expression, Q4 = highest expression). Primary endpoint was OM. DR was defined as recurrence outside the loco-regional area and BCM as death following DR. Competing risks included non-breast-cancer deaths for BCM and all-cause mortality for DR, and DR (if occurring within three months prior to LRR) and all-cause mortality for LRR. Follow-up time was calculated using reverse Kaplan-Meier.
Multivariable analyses were performed using flexible parametric survival models on the log cumulative hazard scale, incorporating a spline with one knot at the median event time. If models failed to converge, the spline was omitted. The analyses were adjusted for age, histological grade, ER and HER2 status. Adjusted hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated (Q1 was used as reference for multiplex analysis). Standardized cumulative incidence curves were generated from the models. As these curves are model-derived, traditional “number as risk” tables were not applicable. For further details on the survival models, standardized incidence curves, and their application in case-cohort design, see Özcan et al. [17].
Interactions were tested in multivariable analysis and evaluated using the p-value for the interaction term. Effect modification by immune marker levels on benefit from the two fractionation schedules was evaluated using joint Wald tests. All tests were two-sided, with p-values <0.05 considered statistically significant. Analyses were conducted in Stata version 18.5 (StataCorp LLC, College Station, TX, USA).

Results

Patients
Median age at diagnosis was 59 (range, 53–66) years. Of all patients, 48% (n = 600) received 50 Gy/25 fr, while 52% (n = 642) were randomized to 40 Gy/15 fr. Median follow-up time was 7.9 (interquartile range, 6.8–9.2) years, and patients were observed until March 1, 2020. During this period, LRR occurred in 32 patients, 38 had DR, while 37 died of breast cancer, and 84 died from any cause.
Table 1 shows clinicopathological characteristics of the patients by TILs status. High TILs were found in 18% of the patients. High TILs were associated with younger age, higher T- and N-stage, ductal carcinomas, higher grade, ER+, and HER2 + status.

Immune cell composition
Immune cell marker analyses were based on total area fractions categorized into quartiles. Among the immune cell populations, CD11c+ cells were the most abundant (median 3.31%), whereas FOXP3+ cells were the least abundant (median 0.20%) (Supplementary Figure S4). Immune cells were primarily localized in the stromal compartment, with only sparse presence in the tumor epithelial regions, and ER- tumors exhibited higher levels of immune infiltrations (Supplementary Figure S5).
Supplementary Figure S6 presents pairwise correlations between immune markers based on continuous area fractions in the total tumor area. T-cell markers were strongly correlated, with the highest correlation between CD8+ and CD4+ cells (Pearson correlation coefficient (r)= 0.78), followed by CD8+ and FOXP3+ (r = 0.74), suggesting co-infiltration of cytotoxic and regulatory T-cells. CD11c+ DCs showed moderate correlations with T-cell markers (r = 0.56–0.73), indicating partial overlap between adaptive and innate immune components. CD68+ macrophages correlated weakly with T-cells, with the lowest correlation with CD8+ (r =  0.32).

Prognostic value of TILs
Among all patients, no statistically significant associations were observed between TILs and prognosis (Fig. 1). For OM, patients with high TILs had an adj. HR of 0.76 (95% CI 0.29-2.00), corresponding to 9-year OM-risks of 6% vs. 8% (Fig. 1A). Similar trends were observed for BCM, DR and LRR, but only very few events occurred (Fig. 1B-D).

When stratified by ER status, the adjusted HRs suggested lower OM among patients with ER-/high TILs tumors compared to ER-/low TILs tumors (adj. HR 0.43 (0.09–2.03)), corresponding to 9-year OM-risks of 7% vs. 16%. Among the ER+ group, high TILs showed a numerically higher OM-risks (8% vs. 6%, adj. HR 1.22 (0.38–3.95)). The interaction test between ER status and TILs was not statistically significant (p = 0.09) (Fig. 2A-B). Similar trends were observed for BCM and DR (Fig. 2C-F), however, the number of events was small. Due to few events, LRR could not be analyzed with ER stratification. Given the low number of events and wide confidence intervals, along with non-significant interaction tests, the ER-stratified analyses should be regarded as exploratory and hypothesis-generating.

Prognostic value of immune subsets
When evaluating immune cell subsets, higher CD8+ and CD68+ cell levels were associated with numerically lower OM in the full cohort without ER stratification (Fig. 3A and D). For CD8+ T-cells, patients with highest infiltration had an absolute reduction in OM-risk of 7% (risk estimates for Q1 and Q4: 11% vs. 4%, respectively) (Fig. 3A). For CD4+ and FOXP3+ T-cells, the prognostic value was less pronounced (Fig. 3B-C). For CD68+ macrophages, a marked but non-significant OM-risk reduction between Q1 and Q3 of 11% (Fig. 3D), while CD11c+ DCs demonstrated the greatest numerical risk reduction between Q1 and Q2 (absolute risk difference 6%) (Fig. 3E).

Stratifying by ER status revealed the same overall patterns for immune subsets as in the TILs analysis. Although the association between higher immune cell levels and reduced mortality appeared numerically more pronounced in ER- tumors than in ER+ tumors, none of the interaction tests reached statistical significance (Fig. 4). For CD8+ T-cells, the 9-year cumulative OM-risk decreased from 49% in Q1 to 6% in Q4 (absolute risk-difference 43%, HR Q4: 0.06 (0.01–0.47)) for patients with ER- tumors, while the risk difference was 5% between Q3 vs. Q4 in the ER+ group (Fig. 4A-B). A similar trend appeared for FOXP3+ Tregs, with a reduced OM-risk of 46% observed for patients with ER- tumors, while no difference among quartiles for patients with ER+ tumors was observed (Fig. 4E-F). For CD68+ macrophages, a similar dose-response relationship between higher quartiles and reduced OM-risk was observed for patients with ER- tumors, while reduction in survival was modest for patients with ER+ tumors (Fig. 4G-H). CD4+ and CD11c+ levels for patients with ER- tumors included outliers and small subgroups, resulting in several high HRs with wide CIs (Figs. 4C-D and I-J). These results should be interpreted with caution. Figure 5 summarizes the adjusted HRs for overall mortality across TILs and immune markers.

Predictive value of immune cells on benefit from fractionation
No statistically significant predictive interaction was observed between TILs levels and fractionation scheme in relation to OM (test for interaction, p =  0.50) (Fig. 6). Similarly, none of the immune cell subsets showed predictive value for fractionation regimen (p-values ranged from 0.25 to 0.99, figures not shown).

Discussion
This study investigated the prognostic and potential predictive value of TILs and immune cell subsets in a well-defined, randomized cohort of irradiated, node-negative breast cancer patients treated with 50 Gy/25 fr versus 40 Gy/15 fr. We observed trends indicating that immune infiltration may be associated with better prognosis, with the strongest effects observed among patients with ER- tumors. Additionally, we did not find a predictive value of immune infiltration on benefit from fractionation. Taken together, these findings may further extend the prognostic relevance of immune infiltration beyond high-risk, node-positive patients, suggesting that TILs and immune cell subsets may hold prognostic information in early-stage, node-negative patients. However, the modest number of events led to wide confidence intervals and non-significant results, and the estimates should be interpreted with some caution. The observed associations should therefore be considered as hypothesis-generating and require confirmation in larger studies. However, the overall patterns support the tumor immune landscape as a general prognostic marker across different risk groups.
Although statistical significance was not reached, the observed trends align with our previous findings, indicating that low TILs levels were associated with poorer OM among patients with ER- tumors, whereas no prognostic value of TILs was evident in the ER+ group [17, 19]. These findings suggest a potential interaction between immune infiltration and ER status, with TILs appearing to be more prognostic in more immunogenic, non-luminal subtypes [13, 14]. In our previous study, however, the prognostic value of TILs within ER+ disease appeared most evident in high-grade tumors [17]. Together, these observations suggest the presence of an aggressive ER+ subgroup where immune infiltration could have distinct prognostic effects, highlighting the need for refined risk stratification within luminal disease and consideration of immune profiling to identify patients who may benefit from tailored radiotherapy strategies or combination treatments. The proportion of high TILs (18%) may appear high for a node-negative cohort, but is explained by enrichment of ER- and grade 3 tumors, whereas prevalence was low in grade 1 disease, indicating that TILs distribution follows tumor biology as previously described [28, 29]. By including node-negative patients, our findings are compatible with a prognostic impact of TILs independent of nodal status. Methodological consistency with the DBCG IMN2 study, including study design, immune marker assessment, TILs cut-off, and endpoint definitions, strengthens comparability. Both cohorts were diagnosed and treated within the same quite recent time frame, enhancing external validity. As screening increasingly detects early-stage disease [30], the DBCG HYPO cohort likely represents the majority of contemporary clinical cases, underscoring the translational and clinical relevance of the findings.
Analysis of individual immune subsets largely confirmed the subtype-dependent patterns observed for TILs, with the strongest associations in ER- tumors. CD8+ T-cells showed a clear dose-response relationship, although with wide CIs, in agreement with prior studies demonstrating a favorable prognostic impact in non-luminal disease [31–33]. FOXP3+ Tregs and CD68+ macrophages also tended toward improved outcomes in ER- tumors, whereas most previous studies have associated higher levels of these cells with adverse prognosis, although findings remain heterogeneous across breast cancer subtypes [34–37]. For FOXP3+ cells, this discrepancy may reflect biology-dependent immunological effects, as the prognostic impact of FOXP3+ infiltration appears to vary by tumor subtype. Prior studies have reported no effect or adverse associations in luminal tumors [36], whereas a favorable association between FOXP3+ Tregs and prognosis has been described in ER- or highly immunogenic tumors [37]. For CD68+ macrophages, the observed associations may possibly be explained by methodological differences, as CD68 is a pan-macrophage marker capturing both pro-inflammatory M1-like and immunosuppressive M2-like macrophage subsets, whereas adverse associations in prior studies have often been driven by M2-specific markers [34]. By contrast, CD4+ and CD11c+ infiltration did not demonstrate consistent prognostic associations, and the presence of outliers and small subgroups in ER- tumors likely contributed to imprecise estimates with wide confidence intervals.
The divergent predictive value of TILs on radiotherapy benefit reported in the DBCG82bc and SweBCG91RT studies [18–21] may reflect differences in baseline prognosis and recurrence rates, which in turn shape distinct immune landscape [13]. In contrast, the present DBCG HYPO cohort, despite sharing comparable baseline prognosis with the SweBCG91RT cohort, demonstrated trends of prognostic associations more in line with those observed in the high-risk cohorts [17–19], supporting a robust association between immune infiltrations, ER status, and outcomes after radiotherapy across different risk profiles.
Although we did not identify differential immune effects between the fractionation schedules, the biological impact of dose and fractionation remains an active area of research. The present analyses were limited to conventional and moderate hypofractionated RT (50 Gy/25 fr vs. 40 Gy/15 fr), and whether immune markers may interact similarly with modern ultra-hypofractionated schedules (e.g. 26 Gy/5 fractions) remains to be explored in future clinical studies. A clinical study of breast cancer patients compared 50 Gy/25 fr vs. 40.3 Gy/13 fr schedules, reporting that hypofractionation preserved higher lymphocyte counts, particularly CD4+ T-cells, while conventional radiotherapy was associated with prolonged lymphodepletion. However, these findings reflect systemic rather than intratumoral immunity and do not directly capture tumor microenvironment-specific immune alterations [38].
Preclinical studies indicate that immune effects of radiotherapy depend on dose and fractionation. Experimental models suggest that hypofractionated radiotherapy can enhance antigen presentation and T-cell activation compared with more protracted schedules or single ablative doses, particularly when combined with immunotherapy [10, 11, 39, 40]. However, these findings derive from murine models with intact tumors and should be interpreted with caution, as they may not directly reflect immune dynamics in the adjuvant clinical setting.
In this study, immune infiltration was measured in surgically removed primary tumors, with adjuvant radiotherapy delivered to residual breast tissue, in which minimal or no tumor burden is expected to remain. In contrast, most preclinical studies have investigated fractionation and dose effects on intact tumors in vivo, where a higher mutational load may elicit stronger immune responses, limiting direct clinical translation. Preclinical data further suggest that tumor control after radiotherapy may depend on the survival and reactivation of pre-existing tumor-specific T-cells, particularly tissue-resident memory T-cells, which appear more radioresistant than circulating T-cells [41]. Thus, it can be hypothesized that hypofractionated radiotherapy may exert immunostimulatory effects by reactivating T-memory cells in residual disease or peripheral sites, where they expand into effector T-cells such as CD8+ T-cells [42]. Future clinical studies should aim to incorporate functional immune profiling (e.g. exhausted vs. memory phenotypes) to better characterize the immune landscape and its relevance to radiotherapy-induced immune responses.
In this study, immune profiling was performed using single 1.5 mm TMA cores sampled from tumor regions with immune infiltration. This targeted approach was chosen to characterize immune cell composition but may not fully capture intratumoral spatial heterogeneity or spatial immunophenotypes (e.g. immune-desert, immune-excluded, or immune-inflamed phenotypes) [43], which may influence interpretation of the immune infiltration. This is a recognized limitation when using TMAs for immune profiling in breast cancer [31, 44].
This study is limited by the modest number of events and the small effective sample size in subgroup analyses, including patients with high TILs, which reduced statistical power and resulted in wide confidence intervals. Within the case-cohort design, subgroup and interaction analyses may be subject to increased variability, especially when the number of events is low, limiting the precision and interpretability of HR estimates. Consequently, all effect estimates should be regarded as non-significant and exploratory. Furthermore, the absence of a non-irradiated control group precluded direct evaluation of immune markers as predictors of radiotherapy benefit. While definitive conclusions cannot be drawn, our findings indicate that immune infiltration may be useful for prognostic risk stratification after radiotherapy in node-negative patients. External validation in larger cohorts of node-negative breast cancer patients and prospective evaluation is required before future clinical implementations.

Conclusion
This study suggests that biomarkers of immune infiltration, including TILs, CD8+, FOXP3+ T-cells, and CD68+ macrophages, may withhold prognostic information in node-negative, early-stage irradiated breast cancer patients, with trends that appear more pronounced in patients with ER- tumors. However, given the limited number of events and small subgroup sizes, these findings should be considered hypothesis-generating trends rather than definitive effect estimates. We observed no statistically significant evidence of a predictive value of immune markers for radiotherapy fractionation, suggesting that pre-existing anti-tumor immunity did not appear to modify the effect of fractionation schedules. Taken together, these observations support a potential prognostic role of pre-existing anti-tumor immune composition and outcome in node-negative breast cancer patients, while emphasizing the need for validation in larger studies before future clinical implications can be considered.

Supplementary Information