Multiomics Machine Learning to Predict Neoadjuvant Chemotherapy Outcome and Relapse of Breast Cancer

BackgroundAberrant activation of anaplastic lymphoma kinase (ALK) signaling has been found to be involved in the tumorigenesis of multiple types of cancer. The aim of this study was to determine the role of this pathway in the pathogenesis of breast cancer.MethodsAn ALK pathway signature that we generated previously was used to compute the ALK pathway activity in 6381 breast cancer samples from 42 microarray datasets, and the associations between ALK pathway signature score and clinical variables were examined using logistic regression and survival analyses.ResultsOur results indicated that high ALK pathway activity was a significant risk factor for hormone receptor‐negative, high‐grade breast cancer in the 42 datasets. ALK pathway activity was positively associated with pathological complete response (pCR) in 15 datasets annotated with patient's neoadjuvant chemotherapy response information (overall odds ratio = 1.67, P < 0.01), and this association was more significant in HER2‐negative and grade 1&2 tumors than in HER2‐positive and grade 3 tumors. ALK pathway activity was also positively associated with recurrence risk in breast cancer patients from 30 datasets annotated with survival information (overall hazard ratio = 1.21, P < 0.01), particularly in patients with age > 50 years old, with positive lymph nodes, or with residual disease after neoadjuvant chemotherapy.ConclusionsALK may be involved in breast cancer tumorigenesis, and ALK pathway signature score may serve as a prognostic biomarker for breast cancer.

Androgen receptor (AR) antagonists are currently tested in multiple clinical trials for different breast cancer (BC) subtypes, which emphasizes the need for clarifying the role of AR in this type of cancer. Previous studies showed that AR expression was associated with a favorable prognosis in ER-positive BC. However, the true biological effect of AR signaling in BC is not clear. An AR pathway signature was generated to compute AR pathway activity in BCs (n=6439) from 46 microarray datasets. Associations of AR pathway activity and AR expression with BC prognosis were compared by survival analysis. AR pathway activity showed moderate positive and negative correlations with AR expression in HER2-positive and HER2-negative BCs, respectively. AR pathway activity increased while AR expression decreased in ER-negative BCs. Like ER and progesterone receptor (PR) expression, AR expression was also negatively associated with tumor grade, neoadjuvant response, and recurrence risk in BC. By contrast, AR pathway activity was positively, and more significantly, associated with these clinical features. Moreover, the AR pathway, but not AR expression, was significantly associated with recurrence risk in BC patients treated with endocrine therapy. These data suggest that, although AR expression probably reflects well-differentiated states of BC and is thus associated with favorable prognosis in BC, the biological effects of AR signaling confers worse outcomes in BC. Our findings encourage the continued evaluation of AR antagonists for BC treatment and support that AR pathway activity serves as a better prognostic factor than AR expression in BC.

Multidisciplinary considerations in the treatment of triple-negative breast cancer.

Numbers of epidemiologic studies assessing soy consumption and risk of breast cancer have yielded inconsistent results. We aimed to examine the association between soy isoflavones consumption and risk of breast cancer incidence or recurrence, by conducting a meta-analysis of prospective studies. We searched for all relevant studies with a prospective design indexed in PUBMED through September 1st, 2010. Summary relative risks (RR) were calculated using fixed- or random-effects models. Pre-specified stratified analyses and dose-response analysis were also performed. We identified 4 studies of breast cancer recurrence and 14 studies of breast cancer incidence. Soy isoflavones consumption was inversely associated with risk of breast cancer incidence (RR = 0.89, 95% CI: 0.79-0.99). However, the protective effect of soy was only observed among studies conducted in Asian populations (RR = 0.76, 95% CI: 0.65-0.86) but not in Western populations (RR = 0.97, 95% CI: 0.87-1.06). Soy isoflavones intake was also inversely associated with risk of breast cancer recurrence (RR = 0.84, 95% CI: 0.70-0.99). Stratified analyses suggested that menopausal status may be an important effect modifier in these associations. We failed to identify a dose-response relationship between total isoflavones intake and risk of breast cancer incidence. Our study suggests soy isoflavones intake is associated with a significant reduced risk of breast cancer incidence in Asian populations, but not in Western populations. Further studies are warranted to confirm the finding of an inverse association of soy consumption with risk of breast cancer recurrence.

BackgroundProper stratification of recurrence risk in breast cancer is crucial for guiding treatment decisions. This study aims to predict the recurrence risk of breast cancer patients using a multimodal deep learning model that integrates multiple sequence MRI imaging features with clinicopathologic characteristics.MethodsIn this retrospective study, we enrolled 574 patients with non-metastatic invasive breast cancer from two Chinese institutions between September 2012 and July 2019. We developed a multimodal deep learning (MDL) model by constructing a multi-instance learning framework based on convolutional neural networks. We integrated imaging features from T2WI, DWI, and DCE-MRI sequences with clinicopathologic features for breast cancer recurrence risk stratification. Subsequently, the performance of the MDL model was evaluated using receiver operating characteristic (ROC) curves, the Hosmer–Lemeshow test, calibration curves, and decision curve analysis (DCA). Survival analysis was conducted with Kaplan–Meier survival curves to stratify breast cancer patients into high and low-recurrence risk groups. Time-dependent ROC curves were used to assess 3-year, 5-year, and 7-year recurrence-free survival (RFS) for breast cancer patients. Additionally, we performed differential and enrichment analyses on Oncotype DX genes. We correlated these genes with clinicopathologic features and deep-learning radiographic features using univariate Cox regression and Pearson correlation analysis.ResultsThe MDL model demonstrated good performance in predicting breast cancer recurrence risk and accurately differentiated between high- and low-recurrence risk groups, with an AUC as high as 0.915 (95% CI 0.8448–0.9856). The C-index of prediction models was 0.803 in the testing cohort. The AUCs for 5-year and 7-year RFS were 0.936 (95% CI 0.876–0.997) and 0.956 (95% CI 0.902–1.000) in the validation cohort. In the testing cohort, these AUCs were 0.836 (95% CI 0.763–0.909) and 0.783 (95% CI 0.676–0.891). This study found a significant correlation between Oncotype DX gene expression, clinicopathologic features, and deep-learning radiographic features (p < 0.05).ConclusionsThis study validated the robust predictive accuracy of the MDL model in identifying high- and low-risk groups for recurrence. The correlations identified between Oncotype DX genes, clinicopathologic features, and deep-learning radiographic features offer novel insights for future biomarker research in breast cancer.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13058-025-02129-z.

Could Modification of Lifestyle Factors Prevent Second Primary Breast Cancers?

We aimed to assess contralateral breast cancer (CBC) risk in patients with ductal carcinoma in situ (DCIS) compared with invasive breast cancer (BC). Women diagnosed with DCIS (N = 28,003) or stage I–III BC (N = 275,836) between 1989 and 2017 were identified from the nationwide Netherlands Cancer Registry. Cumulative incidences were estimated, accounting for competing risks, and hazard ratios (HRs) for metachronous invasive CBC. To evaluate effects of adjuvant systemic therapy and screening, separate analyses were performed for stage I BC without adjuvant systemic therapy and by mode of first BC detection. Multivariable models including clinico-pathological and treatment data were created to assess CBC risk prediction performance in DCIS patients. The 10-year cumulative incidence of invasive CBC was 4.8% for DCIS patients (CBC = 1334). Invasive CBC risk was higher in DCIS patients compared with invasive BC overall (HR = 1.10, 95% confidence interval (CI) = 1.04–1.17), and lower compared with stage I BC without adjuvant systemic therapy (HR = 0.87; 95% CI = 0.82–0.92). In patients diagnosed ≥2011, the HR for invasive CBC was 1.38 (95% CI = 1.35–1.68) after screen-detected DCIS compared with screen-detected invasive BC, and was 2.14 (95% CI = 1.46–3.13) when not screen-detected. The C-index was 0.52 (95% CI = 0.50–0.54) for invasive CBC prediction in DCIS patients. In conclusion, CBC risks are low overall. DCIS patients had a slightly higher risk of invasive CBC compared with invasive BC, likely explained by the risk-reducing effect of (neo)adjuvant systemic therapy among BC patients. For support of clinical decision making more information is needed to differentiate CBC risks among DCIS patients.

Background: Neoadjuvant chemotherapy (NAC) is a critical component of breast cancer treatment; however, patient responses and long-term prognoses vary significantly. Accurately predicting post-NAC prognosis is essential for guiding individualized treatment plans. This study aims to develop a deep learning-based prediction model to analyze the correlation between NAC efficacy and long-term outcomes in breast cancer patients, providing a new approach to identifying high-risk populations. Objective: To construct a deep learning model that integrates multi-dimensional clinical and pathological parameters to predict recurrence and metastasis risk in breast cancer patients following NAC, thereby facilitating personalized treatment strategies. Methods: A retrospective analysis was conducted on 832 breast cancer patients who received NAC at our hospital from 2013 to 2022. Comprehensive clinical, pathological, and molecular subtype data including：pre- and post-NAC tumor characteristics, Ki-67 index, lymph node status, lymphovascular invasion, and Miller-Payne grading were collected. A Multi-layer Perceptron (MLP) based deep learning model was developed, incorporating ensemble learning strategies to integrate multi-modal prediction results. The model’s performance in assessing recurrence and metastasis risk was evaluated across different breast cancer subtypes. Results: The analysis identified key prognostic factors, including tumor size reduction, post-NAC lymph node status, Ki-67 index, lymphovascular invasion, and Miller-Payne grading. The MLP model achieved AUC values of 0.86 (95% CI: 0.82-0.93) for HER2+,0.82 (95% CI: 0.70-0.92)for triple-negative breast cancer, and 0.76 (95% CI: 0.66-0.82) for HR+/HER2−. The model successfully stratified high-risk subgroups with significant differences in prognosis, providing valuable insights for clinical decision-making. Conclusions: The deep learning-based prediction model developed in this study effectively assesses the prognostic risk of breast cancer patients after NAC. Its clinical application holds potential for optimizing individualized treatment and follow-up strategies, ultimately improving patient outcomes.

Introduction: In triple-negative breast cancer (TNBC), both response to neoadjuvant chemotherapy (NAC) and the degree of pre-treatment (pre-tx) tumor immune infiltration as defined by tumor-infiltrating lymphocytes (TILs) are prognostic. Improving NAC response prediction in early TNBC would provide the opportunity to consider adjustments to the NAC regimen prior to initiating therapy. Breast magnetic resonance imaging (MRI) enables noninvasive whole-tumor measurement of microenvironment features. We investigated the value of pre-tx MRI metrics in addition to TILs for the prediction of NAC response in early TNBC patients. Methods: Women with Stage I-III TNBC who underwent pre-tx clinical breast MRI and NAC at our institution (2005-2019) were retrospectively identified. Response to NAC was noted, with pathologic complete response (pCR) defined as no residual invasive cancer present within the breast. When tissue was available, diagnostic biopsy was used to quantify pre-tx TILs as deciles from 10-100% by a breast pathologist. Patients underwent pre-tx breast MRI on either a 1.5T or 3T scanner including diffusion-weighted (DW-) and dynamic contrast-enhanced (DCE-) MRI. From DCE-MRI, tumor longest diameter (LD) and T stage (1-4), as well as contrast kinetics including percent enhancement (PE) at 2 minutes post-contrast and signal enhancement ratio (SER) were determined. Tumor peak PE and peak SER (representing the highest mean PE and SER, respectively, for 3×3 voxel subregions) and functional tumor volume (FTV, tumor volume exhibiting PE ≥ 50%) were calculated. Mean apparent diffusion coefficient (ADC) was calculated from DW-MRI. TIL levels and imaging features were compared between pCR and non-pCR groups by Wilcoxon rank sum test and performance for prediction of pCR was evaluated using areas under the curve (AUC) measures from receiver operating characteristic curve analysis. Results: 115 TNBC patients (median age: 49, range 26-79 years) were evaluated, of which 45 (39%) achieved pCR. The majority received an anthracycline-containing regimen. Pre-tx TILs (evaluated in N=60 with available biopsy specimens) ranged from 10% to 80% (median, 10%) and were significantly higher in pCR vs. non-pCR patients (p = 0.02, AUC = 0.63). Pre-tx lesion size on imaging was predictive of response (Table 1), with pCR patients having significantly lower LD (p &amp;lt; 0.01, AUC = 0.68) and FTV (p = 0.01, AUC = 0.67). Peak PE was also associated with response, significantly lower in pCR patients (p = 0.04, AUC = 0.62), while SER and ADC were not (p &amp;gt; 0.05). Stratifying by T stage, both peak PE (p = 0.03) and FTV (p = 0.05) were predictive of response in T1/T2 patients, while no imaging metrics reached significance in T3/T4 patients. Discussion: In a large cohort of TNBC patients undergoing NAC, measures of tumor size and immune infiltration were strongly predictive of NAC response. Preliminary results suggest that baseline peak PE and FTV are associated with NAC response, particularly in earlier stage TNBC patients. These findings support the utility of imaging and TILs assessments to predict response in TNBC and potentially guide NAC regimens for improved outcomes. Future work will extend these analyses to assess the value of changes in imaging metrics over the course of NAC to predict response. Acknowledgments: NIH P30CA015704, R01CA248192, and Roger E. Moe Fellowship, ASCO/CCF Hayden Family Foundation Young Investigator Award in Breast Cancer Summary of imaging characteristics across TNBC cohort. Values indicate mean (standard deviation).Whole cohort. N=115T1/T2. N=80T3/T4. N=35pCR. N=45non-pCR. N=70pAUCpCR. N=39non-pCR. N=41pAUCpCR. N=6non-pCR. N=29pAUCLongest dimension (mm)34 (15)50 (27)&amp;lt;0.010.6829 (11)31 (10)0.410.5561 (6)76 (20)0.060.75FTV (cm3)14.7 (17.0)31.2 (43.4)0.010.6710.4 (11.0)19.0 (20.2)0.050.6446.1 (24.5)52.0 (58.0)0.720.57Peak SER1.77 (0.26)1.78 (0.23)0.960.501.76 (0.26)1.79 (0.23)0.690.531.91 (0.19)1.78 (0.26)0.270.68Peak PE (%)234 (61)268 (69)0.040.62237 (61)275 (71)0.030.66247 (70)256 (63)1.000.5ADC(×10-3 mm2/s)1.19 (0.29)1.31 (0.36)0.180.581.16 (0.28)1.24 (0.30)0.390.561.40 (0.30)1.4 (0.43)0.900.48 Citation Format: Anum S Kazerouni, Laura C. Kennedy, Michael Hirano, Bonny Chau, Debosmita Biswas, Shaveta Vinayak, Matthew J. Nyflot, Habib Rahbar, Suzanne Dintzis, Savannah C. Partridge. Associations of baseline breast MRI metrics and immune infiltration with chemotherapy response in triple negative breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr PD11-08.

Advances in Screening, Diagnosis, and Treatment of Breast Cancer

XPF -673C>T variation is associated with the susceptibility to breast cancer

BackgroundGrowing evidence supports the modulatory role of human gut microbiome on neoadjuvant chemotherapy (NAC) efficacy. However, the relationships among the gut microbiome, tumor-infiltrating lymphocytes (TILs), and NAC response for breast cancer (BC) patients remain unclear. We thus proposed this preliminary study to investigate the relationship between gut microbiome and BC patients’ responses to NAC treatment as well as underlying mechanisms.MethodsPrior to receiving NAC, the fecal metagenome collected from 23 patients with invasive BC was analyzed. Patients were subsequently assigned to the NAC non-effectual group and the NAC effectual group based on their response to NAC. The peripheral T lymphocyte subset counts were examined by flow cytometry methods. CellMinor analysis was employed to explore the relationship between CD4 mRNA expression and the reaction of tumor cells to NAC drugs.ResultsThe gut microbiomes of the NAC non-effectual group showed characteristics of low diversity with low abundances, distinct metagenomic composition with decreased butyrate-producing and indolepropionic acid-producing bacteria, and increased potential pathobionts compared with the NAC effectual group. The combination of Coprococcus, Dorea, and uncultured Ruminococcus sp. serves as signature bacteria for distinguishing NAC non-effectual group patients from the NAC effectual group. The absolute numbers of CD4+ and CD8+ TIL infiltration in tumors in the NAC non-effectual group were significantly lower than those in the effectual group. Similar findings were reported for the CD4+ T lymphocytes in the peripheral blood (p’s < 0.05). NAC effectual-related signature bacteria were proportional to these patients’ CD4+ T lymphocyte counts in peripheral blood and tumors (p’s < 0.05). CellMinor analysis showed that the CD4 mRNA expression level dramatically climbed with increased sensitivity of tumor cells to NAC drugs such as cyclophosphamide, cisplatin, and carboplatin (p’s < 0.05).ConclusionsThe composition of the gut microbial community differs between BC patients for whom NAC is effective to those that are treatment resistant. The modulation of the gut microbiota on host CD4+ T lymphocytes may be one critical mechanism underlying chemosensitivity and NAC pathologic response. Taken together, gut microbiota may serve as a potential biomarker for NAC response, which sheds light on novel intervention targets in the treatment of NAC non-effectual BC patients.

Integration of a Radiosensitivity Molecular Signature Into the Assessment of Local Recurrence Risk in Breast Cancer

Changes in breast cancer risk associated with different volumes, doses, and techniques in female Hodgkin lymphoma patients treated with supra-diaphragmatic radiation therapy

6085 Background: In HPV-positive oropharyngeal squamous cell carcinoma (OPSCC), patients with good response to neoadjuvant chemotherapy (NAC) exhibit superior prognosis. Accurate prediction of NAC response allows for NAC candidate selection and personalized treatment de-intensification in HPV-positive OPSCC. In this study, we aimed to apply baseline magnetic resonance (MR) radiomic features to predict NAC response and prognosis. Methods: Pre-treatment MR images and clinical data of 131 patients with HPV-positive OPSCC were retrieved from Fudan University Shanghai Cancer Center. Radiomic features of both oropharyngeal lesions and metastatic nodes were extracted on T2WI and contrast-enhanced T1WI sequence. Patients were divided into training cohort (n=47), prospective validation cohort (n=49) and real-world validation cohort (n=35). Following radiomic feature selection, a linear support vector machine (SVM) model was built and validated for NAC response prediction. Nomograms that combined radiomics and clinical characteristics were then developed to predict survival outcomes. The performance of response models was assessed by the area under the curve (AUC), accuracy, sensitivity, specificity and prognostic models were measured by C-index. RNA-seq and proteomic data were further leveraged and compared to interpret the molecular features underlying radiomic signatures with differential NAC response. Results: For NAC response prediction, the fusion model with both oropharyngeal and nodal radiomic signatures on multi-sequence MR images achieved encouraging performance to predict good responders in the training cohort (AUC 0.89, 95% CI, 0.79-0.95) and prospective validation cohort (AUC 0.71, 95% CI, 0.59-0.83). For prognosis prediction, radiomics-based nomograms exhibited satisfactory discriminative ability between low-risk and high-risk patients in training cohort and two validation cohorts (PFS, C-index: 0.85, 0.76 and 0.83; OS, C-index: 0.79, 0.76 and 0.87). An exploratory analysis in the prospective validation cohort showed that de-intensified radiotherapy after NAC in low-risk patients yielded 100% in both PFS and OS. Furthermore, expression analysis unveiled distinct molecular phenotypes in relation to NAC response, where poor responders had predominantly enhanced keratinization while good responders were featured by stronger innate and adaptive immune response. Conclusions: The MR-based radiomic models and subsequent prognostic models efficiently discriminate among patients with different NAC response and survival risk. This study provides a new strategy for patient selection in HPV-positive OPSCC that are suitable for personalized de-intensification. Integration of radiomics in future trials is warranted.