Search

Supplementary Figure from Hypoxia Drives Centrosome Amplification in Cancer Cells via HIF1α-dependent Induction of Polo-Like Kinase 4

Purpose:The purpose of this study is to predict risk of local recurrence (LR) in ductal carcinoma in situ (DCIS) with a new visualization and quantification approach using centrosome amplification (CA), a cancer cell–specific trait widely associated with aggressiveness.Experimental Design:This first-of-its-kind methodology evaluates the severity and frequency of numerical and structural CA present within DCIS and assigns a quantitative centrosomal amplification score (CAS) to each sample. Analyses were performed in a discovery cohort (DC, n = 133) and a validation cohort (VC, n = 119).Results:DCIS cases with LR exhibited significantly higher CAS than recurrence-free cases. Higher CAS was associated with a greater risk of developing LR (HR, 6.3 and 4.8 for DC and VC, respectively; P < 0.001). CAS remained an independent predictor of relapse-free survival (HR, 7.4 and 4.5 for DC and VC, respectively; P < 0.001) even after accounting for potentially confounding factors [grade, age, comedo necrosis, and radiotherapy (RT)]. Patient stratification using CAS (P < 0.0001) was superior to that by Van Nuys Prognostic Index (VNPI; HR for CAS = 6.2 vs. HR for VNPI = 1.1). Among patients treated with breast-conserving surgery alone, CAS identified patients likely to benefit from adjuvant RT.Conclusions:CAS predicted 10-year LR risk for patients who underwent surgical management alone and identified patients who may be at low risk of recurrence, and for whom adjuvant RT may not be required. CAS demonstrated the highest concordance among the known prognostic models such as VNPI and clinicopathologic variables such as grade, age, and comedo necrosis.

Supplementary Data

Purpose: Androgen receptor (AR) expression is absent in 40-90% of estrogen receptor (ER)-negative breast cancers. The prognostic value of AR in ER-negative patients and therapeutic targets for patients absent in AR remains poorly explored. Methods: We used an RNA-based multigene classifier to identify AR-low and AR-high ER-negative participants in the Carolina Breast Cancer Study (CBCS; n=669) and The Cancer Genome Atlas (TCGA; n=237). We compared AR-defined subgroups by demographics, tumor characteristics, and established molecular signatures [PAM50 risk of recurrence (ROR), homologous recombination deficiency (HRD), and immune response]. Results: AR-low tumors were more prevalent among Black (relative frequency difference (RFD) = +7%, 95% CI = 1% to 14%) and younger (RFD = +10%, 95% CI = 4% to 16%) participants in CBCS and were associated with HER2-negativity (RFD = -35%, 95% CI = -44% to -26%), higher grade (RFD = +17%, 95% CI = 8% to 26%), and higher risk of recurrence scores (RFD = +22%, 95% CI = 16.1% to 28%), with similar results in TCGA. The AR-low subgroup was strongly associated with HRD in CBCS (RFD = +33.3%, 95% CI = 23.8% to 43.2%) and TCGA (RFD = +41.5%, 95% CI = 34.0% to 48.6%). In CBCS, AR-low tumors had high adaptive immune marker expression. Conclusion: Multigene, RNA-based low AR expression is associated with aggressive disease characteristics as well as DNA repair defects and immune phenotypes, suggesting plausible precision therapies for AR-low, ER-negative patients.

PurposeMentoring postdocs is a shared responsibility and dynamic process that requires a mutual commitment between the faculty mentor and postdoc. The purpose of this study is to understand how minoritized science, technology, engineering, and mathematics (STEM) postdocs view and engage in mentoring exchanges with their faculty mentors. In the context of this study, minoritized postdocs include women, people of color, and individuals with international status; faculty mentors include postdocs’ Principal Investigators (PIs).Design/methodology/approachThree researchers and 31 data sources (i.e., interview transcripts) were used to construct the case. Researchers first deductively and independently coded the data sources using Molm’s (2006) social exchange framework to identify examples of direct, generalized, and productive mentoring exchanges. Researchers then used thematic analysis (Braun and Clarke, 2006) to identify emergent themes among coded examples of direct, generalized, and productive mentoring exchanges.FindingsData analyses revealed three emergent themes: (1.1) postdocs valued regular meetings and communication with mentors to clarify responsibilities and role expectations, (1.2) postdocs found more value in their interactions with junior faculty PIs who were flexible and open to innovative ideas, and (1.3) postdocs appreciated conversations about short- and long-term career goals and advice with mentors.Originality/valueFindings offer implications for faculty and postdocs’ approaches to mentoring relationships, and for approaches to cultivating supportive scholarly communities in STEM higher education. Recommendations include flexibility in research assignments, increased awareness of non-academic careers, and opportunities for informal interactions and intra/interdepartmental community building.

Quadruple negative breast cancer (QNBC), which refers to triple negative breast cancers (TNBCs) lacking androgen receptor (AR) expression, is now considered to be a highly aggressive BC subtype, with no actionable therapeutic targets. QNBC disproportionately afflicts and impacts patients of African descent. Kinesin family member C1 (KIFC1/HSET), a centrosome clustering protein that prevents cancer cells from undergoing centrosome amplification-induced apoptosis, has been reported to be upregulated in TNBCs and African-American (AA) TNBCs, compared to non-TNBCs and European-American (EA) TNBCs. Herein, we analyzed KIFC1 mRNA levels and their associations with clinical features and outcomes among AR-low and AR-high TNBC patients in three distinct publicly-available gene expression datasets and the Breast cancer gene expression database (bc-GenExMiner). We discovered that KIFC1 levels were significantly higher (p Citation Format: Nikita Wright, Yuan Yate-Ching, Tijana Talisman, Rama Natarajan, Mark LaBarge, Padmashree Rida, Victoria Seewaldt. Kinesin family member C1 (KIFC1/HSET) underlies aggressive disease course in androgen receptor-low triple negative breast cancers [abstract]. In: Proceedings of the 15th AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2022 Sep 16-19; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2022;31(1 Suppl):Abstract nr C065.

Black/African-American (AA) women, relative to their White/European-American (EA) counterparts, experience disproportionately high breast cancer mortality. Central to this survival disparity, Black/AA women have an unequal burden of aggressive breast cancer subtypes, such as triple-negative breast cancer (ER/PR-, HER2-wild type; TNBC). While TNBC has been well characterized, recent studies have identified a highly aggressive androgen receptor (AR)-negative subtype of TNBC, quadruple-negative breast cancer (ER/PR-, HER2-wildtype, AR-; QNBC). Similar to TNBC, QNBC disproportionately impacts Black/AA women and likely plays an important role in the breast cancer survival disparities experienced by Black/AA women. Here, we discuss the racial disparities of QNBC and molecular signaling pathways that may contribute to the aggressive biology of QNBC in Black/AA women. Our immediate goal is to spotlight potential prevention and therapeutic targets for Black/AA QNBC; ultimately our goal is to provide greater insight into reducing the breast cancer survival burden experienced by Black/AA women.

Triple-negative breast cancer (TNBC) surpasses other BC subtypes as the most challenging to treat due to its lack of traditional BC biomarkers. Nearly 30% of TNBC patients express the androgen receptor (AR), and the blockade of androgen production and AR signaling have been the cornerstones of therapies for AR-positive TNBC. However, the majority of women are resistant to AR-targeted therapy, which is a major impediment to improving outcomes for the AR-positive TNBC subpopulation. The hypoxia signaling cascade is frequently activated in the tumor microenvironment in response to low oxygen levels; activation of the hypoxia signaling cascade allows tumors to survive despite hypoxia-mediated interference with cellular metabolism. The activation of hypoxia signaling networks in TNBC promotes resistance to most anticancer drugs including AR inhibitors. The activation of hypoxia network signaling occurs more frequently in TNBC compared to other BC subtypes. Herein, we examine the (1) interplay between hypoxia signaling networks and AR and (2) whether hypoxia and hypoxic stress adaptive pathways promote the emergence of resistance to therapies that target AR. We also pose the well-supported question, “Can the efficacy of androgen-/AR-targeted treatments be enhanced by co-targeting hypoxia?” By critically examining the evidence and the complex entwinement of these two oncogenic pathways, we argue that the simultaneous targeting of androgen biosynthesis/AR signaling and hypoxia may enhance the sensitivity of AR-positive TNBCs to AR-targeted treatments, derail the emergence of therapy resistance, and improve patient outcomes.

Purpose Triple-negative breast cancer (TNBC) is the most deadly form of breast cancer (BC) today. TNBC treatment is fraught with challenges because of the extensive interpatient heterogeneity in clinical behavior and scarcity of stratifying biomarkers and actionable targets. Women of African ancestry face a disproportionate burden resulting from this disease, which affects them earlier and more aggressively and has a higher propensity to spread and resist conventional treatments. A much higher proportion of Nigerian patients with BC have TNBC compared with patients with BC in the United States and Europe. Methods This article spotlights Nigeria as an example of a nation wherein genetic and nongenetic spheres of influence intersect to affect the prevalence of this disease, the scale of its challenge, and its toll. Results Studies have illuminated the inherently different tumor biology of Nigerian TNBCs, which show distinct genetic variants and gene expression patterns compared with European or European-American TNBCs. Parallels are apparent between TNBC phenotypes among African Americans and Nigerians, implicating the common thread of shared genetic ancestry between these populations. Reproductive, lifestyle, socioeconomic, and cultural factors also shape TNBC outcomes in Nigeria, as do resource constraints in Nigerian health care and research sectors. Conclusion Increasing our understanding of how these factors contribute to poorer outcomes among Nigerian women may uncover valuable insights and strategies in alleviating the TNBC burden in many countries of the world and help reduce the racial disparity in BC-related outcomes here in the United States. Importantly, this review also highlights collaborative global and local initiatives that converge expertise and resources to advance research on effective management of TNBC in diverse populations.

The enigma of why some premalignant or pre-invasive breast lesions transform and progress while others do not remains poorly understood. Currently, no radiologic or molecular biomarkers exist in the clinic that can successfully risk-stratify high-risk lesions for malignant transformation or tumor progression as well as serve as a minimally cytotoxic actionable target for at-risk subpopulations. Breast carcinogenesis involves a series of key molecular deregulatory events that prompt normal cells to bypass tumor-suppressive senescence barriers. Kinesin family member C1 (KIFC1/HSET), which confers survival of cancer cells burdened with extra centrosomes, has been observed in premalignant and pre-invasive lesions, and its expression has been shown to correlate with increasing neoplastic progression. Additionally, KIFC1 has been associated with aggressive breast tumor molecular subtypes, such as basal-like and triple-negative breast cancers. However, the role of KIFC1 in malignant transformation and its potential as a predictive biomarker of neoplastic progression remain elusive. Herein, we review compelling evidence suggesting the involvement of KIFC1 in enabling pre-neoplastic cells to bypass senescence barriers necessary to become immortalized and malignant. We also discuss evidence inferring that KIFC1 levels may be higher in premalignant lesions with a greater inclination to transform and acquire aggressive tumor intrinsic subtypes. Collectively, this evidence provides a strong impetus for further investigation into KIFC1 as a potential risk-stratifying biomarker and minimally cytotoxic actionable target for high-risk patient subpopulations.

Supplementary Fig 20. (A) The Harrellâ s c-statistic and 95% confidence interval for the 8-feature model and common clinopathological variables in the validation cohort. (B) The Akaike Information Criterion (AIC) comparing the fit of a null model (no variables), the 8-feature model, and a model composed of the common clinopathological variables (Margins status, necrosis, radiation, age, and size). The lower the AIC value the better the model fits the recurrence data. (PDF 224 kb)

Supplementary Figure S4. An example of the statistical moments obtained from full slide analysis. For each window for an annotated class the distribution of all texture features was computed. From each of these distributions, the mean, standard deviation, skew, and kurtosis was calculated and input as individual components of the full slide feature list. (PDF 285 kb)

Supplementary Figure S13. Impact of clinical features on model performance when clinical variables are concatenated with the 8 features of the recurrence classifier, within a random forest model. Averaged out-of-bag feature importance (and 95% confidence intervals) from 100 models shows that clinical features do not contribute positively to the overall performance of the model. Feature importance (i.e., how heavily the model relies on each given feature for the output prediction) is defined as the change in prediction error when the values of those variables are permuted (to, in effect, break the relationship between the feature and the model outcome) across out-of-bag observations. Hence larger error changes correspond to more vital variables. Insert: Average cross-validated accuracy and hazard ratios of models built with and without clinical variables show (yes/no) significant differences. (PDF 257 kb)

Supplementary Table 3. Features extracted from class-annotated virtual/digital slides. The texture feature distribution statistics constitute the majority of evaluated features as they include the mean, standard deviation, skew, and kurtosis for each of the 166 textural features within each of the 5 annotated classes. (PDF 333 kb)

Supplementary Table S4. The distribution of baseline characteristics between patients who experienced ipsilateral recurrences versus those that did not in the training cohort. The Ď 2 p value signifies significant difference in proportions. (PDF 538 kb)

Supplementary Figure S17. Mean values for the continuous metrics obtained when using A) the class probability, or proportion of recurrence voting trees, using the original random forest model and B) the output of a random survival forest trained with the 8 selected features. The astrix (*) represents groups with significant (p

Supplementary Table S7. Comparison of multiple machine learning algorithms to select the best model (and its associated features) for the recurrence classifier. â No annotationâ indicates the performance of a random forest model built without considering classes obtained from the first annotation step. Optimized models reflect performance after selection of optimal set of features. For each ML model, the model accuracy and high-risk group hazard ratio upon using either the full feature set or the optimized feature set, are shown. (PDF 364 kb)

Supplementary Table S5. Distribution of baseline characteristics between patients who experienced recurrence versus those that did not in the validation cohort. The Ď 2 p-value signifies significant difference in proportions. (PDF 538 kb)

Supplementary Figure S2. Summary of the methodology for model development. (A) The slide annotation classifier was developed using a random selection of slides within the training cohort. The ground truth regions were preprocessed and color deconvoluted so that texture features could be extracted from the hematoxylin distributions. Five-fold cross validation was performed to determine the modelâ s classification ability after which the training set was augmented through rotation and transposition of ground truth regions and input into the final annotation classifier (red box fill). (B) To develop the recurrence classifier the training slides were first annotated through the trained annotation classifier (red box fill). The fully class-annotated slides had whole slide features extracted and selected to identify the set of features that differed most significantly between patients who recurred and recurrence-free patients. The performance of these features within a classifier was determined through 5-fold cross validation, and the full training cohort was used to train a recurrence classifier (gold box fill). (C) The prognostic value of the pipeline was confirmed on a validation cohort. Both the previously-trained annotation classifier and recurrence classifier were applied towards the patient samples in this validation cohort, and the resulting stratification of patients was evaluated. (PDF 403 kb)

Supplementary Table S8. Feature characteristics of the final 8-feature recurrence classification model. The significance shown is based on the t-test for each feature between patients who experienced recurrence within 10 years and those that did not. The misclassification cost is computed sequentially (for e.g., the misclassification cost for feature 3 is the cost for a model which includes features 1, 2 and 3). SFTA: Segmentation-based Fractal Texture Analysis, GLRL: Grey Level Run Length, GLCO: Grey Level Co-Occurrence. (PDF 357 kb)

Supplementary Figure S11. Stratification of patients in training cohort using standard clinical variables. Cross validated Kaplan-Meier curves of patient outcomes (Recurrence-free survival, RFS) stratified based on (A) tumor size, (B) patient age, (C) comedo necrosis status, and (D) Nottingham grade. Significance is measured through the log-rank test. (PDF 253 kb)

Supplementary Figure S15. (A) Cross validated Kaplan-Meier curves of patients within the training cohort stratified by the trained recurrence classifier and using only invasive recurrence as an event. Significance is measured through the log-rank test. (B) Univariate and multivariate Cox regression analysis comparing the influence of common clinicopathological variables alongside the 8-feature recurrence risk prediction model for invasive recurrence-free survival, on the training set. (PDF 340 kb)

Supplementary Figure S7. Effect of sample size used for ground truth annotation on cross-validated accuracy. Average k-fold accuracy of annotation prediction versus number of ground truth regions. Shaded bands represent 95% confidence intervals. (PDF 229 kb)

Supplementary Figure S6. Recurrence distributions of the 159 patients in the training/test cohort, ordered according to earliest censored time or time of recurrence to last follow-up. Red points indicate a recurrence at the last follow up date while green points specify censoring. (PDF 501 kb)

Supplementary Figure S3. Example of region smoothing using a mode (class appearing most often) filter. In this example the middle tile was originally classified as a lymphocyte-dense region. The surrounding neighbors though, were predominantly classified as cancer; thus, the middle tile had its class changed to cancer. While this example showed the mode depending on each tileâ s predicted class, our model actually uses the mode of tree predictions of surrounding neighbors to adjust the middle tile classification. (PDF 342 kb)

Supplementary Figure S1. An example of the Graphical User Interface (GUI) developed to allow for ground truth annotations used for classifier training. Through this interface, a user will select regions representative of each class, from which the program will apply feature extraction from multiple 50x50 pixel windows within that region. (PDF 429 kb)

Supplementary Figure S5. Schematic of the logic used to translate risk category of patient slides to patient risk. Patients who possessed multiple resection slides were put into a high-risk subgroup if any of their slides were classified as high-risk by the recurrence classifier. (PDF 328 kb)

Supplementary Table S9. Univariate cox regression analysis of the impact that continuous metrics can have on both the training (through combining the cross-validation test sets) and validation cohorts. The random survival forest (RSF) was a new model trained with the 8 selected features whereas the RF class probability reflects the continuous score obtained from counting the proportion of trees voting for â recurrenceâ in the classification model. (PDF 399 kb)

Supplementary Figure S9. Combination of features produces optimal stratification. (A) Optimally stratifying patients by feature #3 provides little individual prognostic benefit. However, if patients are first split by feature #1, followed by feature #3 (B), a very significant survival difference can be observed between the high- and low-risk groups (C). (PDF 257 kb)

Supplementary Figure S8. (A) The cumulative density function (CDF) of feature significance, noted by the t-test p-values, versus maximum follow-up (FU) time explored. Using 10-year recurrence, 37% of whole slidefeatures were significantly (0.05) different between patients who developed recurrence by 10 years versus those that remained recurrence-free. (B) Within this 10-year follow-up recurrence distinction, the significant feature distribution by class difference is shown in a radar plot, with the max fill (blood vessel features) indicating 39% of the filtered total significant features. (PDF 232 kb)

Supplementary Figure S19. Kaplan-Meier curves of patients, without discordant slides, within the validation cohort stratified by the trained recurrence classifier model. Significance is measured through the log-rank test. (PDF 199 kb)

Supplementary Methodology. (PDF 700 kb)

Supplementary Table S6. Additional confusion matrix performance metrics for the annotation classifier. (PDF 210 kb)

Supplementary Figure S18. Kaplan-Meier curves of slides within the validation cohort stratified by the trained recurrence classifier model. Significance is measured through the log-rank test. (PDF 201 kb)

Supplementary Equation 1. Density Distance Statistic. Statistic comparing the size (A) and distance (D) between all (sum) cancer (i) areas (connected regions) and either immune-rich or blood vessel (BV) areas (j), normalized (divided) by the total cancer area. (PDF 296 kb)

Supplementary Figures and Tables (DOCX 811Â kb)

Supplementary Table S2. Optical density matrix. This matrix is used to deconvolute RGB H&E images into greyscales of each layer whose intensity correlated with stain absorbance. (PDF 732 kb)

Supplementary Table S2. Optical density matrix. This matrix is used to deconvolute RGB H&E images into greyscales of each layer whose intensity correlated with stain absorbance. (PDF 732 kb)

Supplementary Equation 1. Density Distance Statistic. Statistic comparing the size (A) and distance (D) between all (sum) cancer (i) areas (connected regions) and either immune-rich or blood vessel (BV) areas (j), normalized (divided) by the total cancer area. (PDF 296 kb)

Supplementary Figure S21. Cross validated Kaplan-Meier curves of patients within the validation cohort, developed by combining the testing sets for a cross validated iteration. (A) Recurrence classifier model used on slides from patients who received adjuvant radiation and (B) Patients who were treated with BCS alone. Significance is measured through the logrank test. (PDF 221 kb)

Supplementary Figures and Tables (DOCX 811Â kb)

Supplementary Figure S14. Cross validated Kaplan-Meier curves of patients within the training cohort, developed by combining the testing sets for a cross validated iteration. (A) The recurrence classifier model used with Grade 3 patientsâ slides only. (B) The recurrence classifier model used with Grade 1 and 2 patientsâ slides only. (C) Recurrence classifier used on slides from patients who received adjuvant radiation and (D) Recurrence classifier used on slides taken from patients treated with BCS alone. (PDF 282 kb)

Supplementary Figure S23. (A) Kaplan-Meier curves showing robust stratification of patients in the validation cohort into high-risk of recurrence and low-risk of recurrence subgroups and using only DCIS recurrence as an event. (B) Univariate and multivariate Cox regression analysis comparing the influence of common clinicopathological variables alongside the 8-feature recurrence risk prediction model for DCIS recurrence-free survival, on the validation set. (PDF 325 kb)

Supplementary Figure S10. Kaplan-Meier curves of patients, without discordant slides, within the training cohort stratified by the trained recurrence classifier model. Significance is measured through the log-rank test. (PDF 209 kb)

Supplementary Table S1. Optical density matrix. This matrix is used to deconvolute RGB H&E images into greyscales of each layer whose intensity correlated with stain absorbance. (PDF 206 kb)

Supplementary Figure S22. Cross validated Kaplan-Meier curves of patients within the validation cohort, developed by combining the testing sets for a cross validated iteration. (A) Recurrence classifier model used on slides from patients who received adjuvant radiation and (B) Patients who were treated with BCS alone. Significance is measured through the logrank test. (PDF 221 kb)

Supplementary Figure S14. Cross validated Kaplan-Meier curves of patients within the training cohort, developed by combining the testing sets for a cross validated iteration. (A) The recurrence classifier model used with Grade 3 patientsâ slides only. (B) The recurrence classifier model used with Grade 1 and 2 patientsâ slides only. (C) Recurrence classifier used on slides from patients who received adjuvant radiation and (D) Recurrence classifier used on slides taken from patients treated with BCS alone. (PDF 282 kb)

Supplementary Table S1. Optical density matrix. This matrix is used to deconvolute RGB H&E images into greyscales of each layer whose intensity correlated with stain absorbance. (PDF 206 kb)