Abstract
Background
Despite increasing evidence supporting the clinical utility of immune infiltration in the estrogen receptor-negative (ER-) subtype, the prognostic value of immune infiltration for ER+ disease is not well defined.
Methods
Quantitative immune scores of cell abundance and spatial heterogeneity were computed using fully automated hematoxylin & eosin-stain image analysis algorithm and spatial statistics for 1,178 postmenopausal patients with ER+ breast cancer treated with 5 years’ tamoxifen or anastrozole. Prognostic significance of immune scores was compared with Oncotype DX 21-gene recurrence score (RS), PAM50 risk of recurrence (ROR) score, IHC4, and clinical treatment score (CTS) available for 963 patients.
Results
Scores of immune cell abundance were not associated with recurrence-free survival. In contrast, high immune spatial scores indicating increased cell spatial clustering were associated with poor 10-year, early (0-5 year) and late (5-10 year) recurrence-free survival (Immune Hotspot LR-χ2 = 14.06, P < 0.001_for 0-10 year; LR-χ2 = 6.24, P = 0.01 for 0-5 year; LR-χ2 = 7.89, P = 0.005 for 5-10 year). The prognostic value of spatial scores for late recurrence was similar to that of IHC4 and RS in both populations, but was not as strong as other tests in comparison for recurrence across 10 years.
Conclusions
These results provide a missing link between tumor immunity and disease outcome in ER+ disease by examining tumor spatial architecture. The association between spatial scores and late recurrence suggests a lasting memory of pro-tumor immunity that may impact on disease progression and evolution of endocrine treatment resistance, which may be exploited for therapeutic advances.
Introduction
Estrogen receptor-positive (ER+) subtype accounts for about 80% of all breast cancers, the most common cancer in women. At diagnosis, the majority of ER+ patients have a good prognosis if treated with endocrine therapy. However, a subset of patients is at risk for disease recurrence and death, particularly after 5 years of adjuvant endocrine therapy. Differentiating these patients from low-risk patients who can safely avoid chemotherapy is a priority for breast cancer management (1). Currently available prognostic tests to predict risk in endocrine-treated patients include the widely used Oncotype DX 21-gene recurrence score (RS) (2), the PAM50 risk of recurrence (ROR) score (3), and the immunohistochemistry-based IHC4 test that is combined with the clinical treatment score (CTS) to integrate clinicopathological parameters (4). In particular, the amount of prognostic information provided for long-term (0-10 years) and late (beyond five years) recurrence varies across these tests (5, 6).
Immune infiltration is not explicitly accounted for in any of the above tests. Increasing evidence supports the role of tumor-infiltrating lymphocytes (TILs) in influencing disease progression and treatment response in breast cancer (7–10). Characterization of the nature of immune responses is key to understanding tumor immunity and empowering immunotherapy. However, the majority of reports focus on ER- and Human Epidermal growth factor receptor 2-positive (HER2+) breast cancers, where extensive immune infiltration is more common and immune scores were found to be highly predictive of survival and response to chemotherapy (8, 11–15). In contrast, there is a lack of definitive data on the prognostic value of immune scores in ER+ breast cancer following endocrine treatment (7, 16). A major reason for this is the absence of reproducible scoring methods to facilitate large-scale studies of ER+ breast cancer. This also limits the translation of immune scoring into clinical advances.
We have developed quantitative and reproducible approaches to score lymphocytic infiltration (LI) in breast cancer, based on fully automated image analysis of routinely generated hematoxylin & eosin (H&E)-stained histology sections. Quantitative immune scores of overall LI in tumors, as well as intra-tumor lymphocyte ratio (ITLR), were associated with good survival in ER- and ER-/HER2- breast cancer (17, 18). In addition, our automated image analysis scheme enables the study of complex spatial patterns of TILs (19). The spatial interactions among TILs and cancer cells generate complex ecological dynamics that can ultimately impact tumor progression and response to treatment (8, 20–22). In this study of 1,178 postmenopausal breast cancer patients with ER+ disease patients enrolled in the ATAC (Arimidex or Tamoxifen Alone or Combined) trial, our aims were to: 1) establish the prognostic value of H&E-based, quantitative scores of TIL abundance as well as spatial heterogeneity for 10-year, early (0-5 year) and late (5-10 year) recurrence after endocrine therapy; 2) compare their prognostic value with established prognostic tests including RS, ROR, IHC4 and CTS; 3) evaluate a new histology-based model that combines IHC4 and immune scores as a cost-effective biomarker.
Materials and Methods
Study population
Classical clinicopathologic factors (age, nodal status, tumor size, grade, randomized treatment) were collected from patients with ER+ primary breast cancer in the ATAC trial who were randomly assigned to either anastrozole or tamoxifen (23) (Supplementary Table 1). 1,178 eligible patients who did not receive chemotherapy and from whom H&E-stained slides from formalin-fixed, paraffin-embedded tissues were available were included (Figure 1). Of these, 963 patients were scored with prognostic scores including IHC4, RS, ROR46 and CTS (Table 1). 1,037 tumors were HER2-, 909 of which were scored with prognostic scores. A subset of 91 TransATAC samples were randomly selected and scored on H&E sections according to international recommendations (24) by a histopathologist (KN). Baseline demographics and clinical characteristics for all patients included in this analysis are provided in Supplementary Table 1. This study was approved by the South-East London Research Ethics Committee, and all patients included gave informed consent.
Figure 1. Consort diagram for the availability of samples for analysis from the ATAC (Arimidex, Tamoxifen Alone or Combined) trial.
ER, estrogen receptor; IHC, immunohistochemistry; PgR, progesterone receptor; ROR, Risk of Recurrence score; RS, recurrence score.
Table 1. Immune scores and prognostic scores*.
| Type and Name | Definition | Reference |
|---|---|---|
| Immune abundance scores (based on H&E) | ||
| LI: Lymphocytic Infiltrate | Fraction of lymphocytes in all cells. Cells were identified using automated histology image analysis. | Yuan et al., 2012 (17) |
| ITLR: Intra-Tumor Lymphocyte Ratio | The number of intra-tumor lymphocytes normalized by the number of cancer cells. Intra-tumor lymphocytes were identified as the cluster of lymphocytes to be the closest to cancer cells based on their spatial proximity in unsupervised clustering. | Yuan et al., 2015 (18) |
| ATLR: Adjacent-Tumor Lymphocyte Ratio | The number of adjacent-to-tumor lymphocytes normalized by the number of cancer cells. Adjacent-to-tumor lymphocytes were identified as the intermediate cluster of lymphocytes based on their spatial proximity to cancer cells in unsupervised clustering. | |
| DTLR: Distal-Tumor Lymphocyte Ratio | The number of distal-tumor lymphocytes normalized by the number of cancer cells. Distal-tumor lymphocytes were identified as the cluster of lymphocytes to be the furthest away from cancer cells based on their spatial proximity in unsupervised clustering. | |
| Immune spatial scores (H&E) Cancer Hotspot | Fraction of tissue displaying spatial clustering of cancer cells. Tumor regions with spatial clustering of cancer cells were identified using Getis-Ord hotspot analysis, which assigned statistical significance of difference between observed local cancer cell density and global mean. | Nawaz et al., 2015 (13) |
| Immune Hotspot | Fraction of tissue displaying spatial clustering of immune cells. Tumor regions with spatial clustering of immune cells were identified in a similar way as cancer hotspots. | |
| Immune-Cancer Hotspot | Fraction of tissue displaying spatial clustering of immune and cancer cells simultaneously, in other words, tumor regions that are both immune hotspots and cancer hotspots. | |
| Prognostic scores IHC4 (IHC) | Immunohistochemistry-based score as a combination of ER, PgR, HER2, Ki67 expression, previously derived using TransATAC samples. | Cuzick et al., 2011 (4) |
| CTS (clinical) | Clinical treatment score that consider node status, size, grade, age and treatment, previously derived using TransATAC samples. | Cuzick et al., 2011 (4) |
| RS (Molecular) | Oncotype DX 21-gene recurrence score based on RNA expression of 21 prespecified Oncotype DX genes. | Paik et al., 2004 (2) |
| ROR (Molecular) | PAM50 risk of recurrence score that combine molecular signatures with clinical information on tumor size. | Nielsen et al., 2010 (3) |
Unless specified otherwise, the scores were predefined for validation in external cohorts.
H&E: hematoxylin & eosin, IHC: immunohistochemistry.
H&E image analysis and validation
We curated a digital database of H&E histology slides for TransATAC and applied our histology image analysis pipeline (17) (Supplementary Figure 1A). In brief, the image analysis pipeline exploits the nuclear morphological differences among cancer cells, lymphocytes, and stromal cells to differentiate them in H&E histological tissue sections. Cancer cell nuclei are generally large in size and demonstrate greater variability in appearance as compared to lymphocyte and stromal cell nuclei; lymphocyte nuclei are typically small, round and homogeneously basophilic, and nuclei of stromal cells including fibroblasts and endothelial cells are more elongated. The pipeline consisted of four stages: (1) unsupervised segmentation of the nuclei; (2) supervised classification of individual cell nuclei into cancer, lymphocyte, other cell nuclei, and artefacts; (3) kernel smoothing to correct local sporadic errors; and (4) a hierarchical multi-resolution model fitting to identify cancer cell clusters to further improve classification accuracy. The classifier was previously validated in METABRIC to have an overall accuracy of 90.1% and a high correlation between image analysis and pathological scores in the entire cohort (17).
To evaluate the accuracy of our image analysis pipeline for TransATAC, a test set of 627 cells randomly sampled from 3 images were annotated by a pathologist (DNR) blinded to image analysis results (Supplementary Figure 1A). Accuracy for identifying the three cell types was: cancer cell 93.8%, lymphocyte 87.9% and stromal cell 84.2% (Supplementary Figure 1B). The balanced accuracy as the average for sensitivity and specificity for the three cell types were 0.864 for cancer, 0.839 for lymphocyte and 0.876 for stromal cell (Supplementary Figure 1C). On average 217,101 cancer cells (±178677.5), 25,956 lymphocytes (±35365.21) and 161,341 stromal cells (±91862.59) were identified in each TransATAC whole-section sample, with a total of 525,718,198 cells identified in the whole cohort.
Automated scoring of immune cell abundance and spatial heterogeneity
Immune cell abundance scores include Lymphocytic Infiltration (LI) which summarizes the fraction of lymphocytes in all cells in the tumor section, Intra-Tumor Lymphocyte Ratio (ITLR), Adjacent-to-Tumor Lymphocyte Ratio (ATLR) and Distal-To-Tumor Lymphocyte Ratio (DTLR) which are defined as the number of specific types of lymphocytes normalized by the number of cancer cells (17, 18). Intra-Tumor Lymphocytes (ITLs), Adjacent-to-Tumor Lymphocytes (ATLs) and Distal-To-Tumor Lymphocytes (DTLs) were identified using unsupervised clustering based on their spatial proximities to cancer cells, which was quantified using a kernel density method on the distribution of cancer cells.
Immune cell spatial scores quantify the amount of spatial clusters, or hotspots, formed by lymphocytes and/or cancer cells within the section (Figure 2) (13). Getis-Ord spatial analysis was carried out to identify tumor regions with statistically significant spatial clustering of immune cells, i.e., immune hotspots. This means that, the frequency of immune cells appearing at these locations is greater than expected by chance given the distribution of all cells in the entire tumor section and that, importantly, the difference between the actual and expected value is statistically significant. A p-value was computed for each spatial region, which can be used to determine statistical significance using a significance level of 0.05. Previously, tumors with a high amount of regions that are both immune hotspots and cancer hotspots (immune-cancer hotspots) were found to have good prognosis in ER- cancer (13). Here for comparison in ER+ breast cancer, Cancer Hotspot and Immune Hotspot scores that examine only one type of cell at a time were also included.
Figure 2. Illustration of the pipeline for identifying spatial hotspots with visual examples, and the Kaplan-Meier estimates for 10-year recurrence according to immune spatial scores in the validation set, split into two groups using cutoffs selected in the training set.
A) An example of TransATAC H&E image and corresponding map of identified cancer and immune cells. Scale bar illustrates 2.5mm. B-D) Visual examples of hotspots and Kaplan-Meier curves illustrating survival associations with immune spatial scores. Scale bar illustrates 35μm. Kaplan-Meier curves were calculated and tested for equality using the log-rank test. The numbers of patients at risk in each group at various time points are given below each graph. All statistical tests were two-sided. HR: hazard ratio (95% confidence interval).
For validation of the inter-correlations among immune scores, H&E-stained, whole-section images of 743 ER+, treatment-naïve primary tumors from the METABRIC study (25) were analyzed using the same methods. All patient specimens were obtained with appropriate ethical approval from the relevant institutional review boards.
Statistical analyses
Our primary objective was to assess whether immune scores had statistically significant prognostic information for predicting 10-year recurrence in postmenopausal women with breast cancer given either tamoxifen or anastrozole monotherapy but not chemotherapy. Secondary analyses included determining the prognostic ability of immune scores in predicting early (0-5 year) and late recurrences (5-10 year), in patients divided into subgroups by HER2 status, and the additional prognostic information provided by tests in multivariable comparisons including age (<65, ≥65 years), nodal status (0,1-3,4+), tumor size (≤1cm, >1 to ≤2cm, >2to ≤3cm, >3 cm), centrally read grade (poor, intermediate, well differentiated), and randomized treatment (anastrozole v tamoxifen). Hazard Ratios are for a change in 1 SD in the overall dataset to compare the effect size between different immune scores. The contribution of each of the variables was evaluated by the change in likelihood ratio χ2 (LR-χ2; 1 df, significance level χ2=3.84) in three ways: by univariate analyses, as an addition to a model containing only the clinical variables, and as a difference in LR-χ2 when the variable was added to the IHC4 score. Sample splitting was used in which the immune score was dichotomized by using half the data as the training set, and then the cut-off points for each score were evaluated in the remaining half of the data as the validation set. Since the use of optimizing cut-offs may lead to overestimation of prognostic power, dichotomized variables were only used for analysis presented in Figure 2B-D, and continuous variables were used elsewhere. For measuring correlation among immune scores and with pathological TIL scores, Spearman’s correlation was used. All statistical analyses were performed using STATA version 13.1 or R version 3.3.1. All statistical tests were two-sided and a P value of less than 0.05 was considered statistically significant.
Results
Correlations among immune scores
Four immune abundance scores (overall Lymphocytic Infiltration LI, Intra-Tumor Lymphocyte Ratio ITLR, Adjacent-to-Tumor Lymphocyte Ratio ATLR and Distal-To-Tumor Lymphocyte Ratio DTLR) and three spatial scores (Immune Hotspot, Cancer Hotspot, Immune-Cancer Hotspot) were calculated based on fully automated histology image analysis on whole section slides (Figure 2, Table 1, Supplementary Figure 2). There was a strong, negative correlation between ITLR and DTLR (r = -0.888, Table 2), indicating that lymphocytes either infiltrate into close contact with cancer cells, or largely stay in the stromal area. Correlations among spatial scores were also strong (r = 0.502-0.796), suggesting that spatial clustering of cancer cells and lymphocytes tends to co-occur in the same tumors. These data were further validated in the METABRIC cohort (25) (n = 743, Table 2). We then compared the automated scores to a pathologist’s TIL score (24) in a subset of 91 TransATAC samples. Overall, a weak correlation between the pathologist’s score and all automated scores was found (r < 0.260), with the highest correlation observed between TILs scoring and DTLR (r = 0.259) (Supplementary Figure 3).
Table 2. Correlations among immune scores and clinical variables in TransATAC and METABRIC*.
| Dataset / variable | ITL | ATL | DTL | LI | Cancer Hotspot | Immune Hotspot | Immune-Cancer Hotspot | Age | Grade | Node | Size | Treatment |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TransATAC | ||||||||||||
| ITL | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| ATL | -0.317 | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| DTL | -0.888 | -0.154 | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| LI | -0.01 | 0.381 | -0.174 | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| Cancer Hotspot | -0.422 | 0.309 | 0.289 | 0.167 | -- | -- | -- | -- | -- | -- | -- | -- |
| Immune Hotspot | -0.395 | 0.409 | 0.213 | 0.476 | 0.796 | -- | -- | -- | -- | -- | -- | -- |
| Immune-Cancer Hotspot | -0.408 | 0.203 | 0.327 | 0.341 | 0.502 | 0.75 | -- | -- | -- | -- | -- | -- |
| Age | 0.01 | 0.018 | -0.019 | -0.036 | 0.145 | 0.077 | 0.033 | -- | -- | -- | -- | -- |
| Grade | -0.168 | 0.126 | 0.114 | 0.144 | 0.268 | 0.287 | 0.243 | 0.128 | -- | -- | -- | -- |
| node | 0.007 | 0.212 | -0.018 | 0.213 | 0.096 | 0.096 | 0.107 | 0.188 | 0.099 | -- | -- | -- |
| Size | 0.021 | -0.012 | -0.016 | -0.051 | 0.231 | 0.194 | 0.155 | 0.247 | 0.171 | 0.322 | -- | -- |
| Treatment | 0.028 | -0.03 | -0.014 | 0.044 | -0.023 | 0.021 | 0.048 | -0.002 | -0.014 | 0.007 | 0.024 | -- |
| METABRIC | ||||||||||||
| ITL | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| ATL | -0.276 | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| DTL | -0.83 | -0.306 | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| LI | 0.077 | 0.199 | -0.191 | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| Cancer Hotspot | -0.331 | 0.035 | 0.307 | 0.062 | -- | -- | -- | -- | -- | -- | -- | -- |
| Immune Hotspot | -0.319 | 0.16 | 0.223 | 0.29 | 0.849 | -- | -- | -- | -- | -- | -- | -- |
| Immune-Cancer Hotspot | -0.253 | 0.051 | 0.221 | 0.152 | 0.796 | 0.824 | -- | -- | -- | -- | -- | -- |
ITLR: Intra-Tumor Lymphocyte Ratio, ATLR: Adjacent-to-Tumor Lymphocyte Ratio, DTLRL: Distal-To-Tumor Lymphocyte Ratio, LI: Lymphocytic Infiltration.
Prognostic value of immune scores
None of the immune abundance scores provided significant prognostic information for recurrence (p > 0.1). In contrast, high spatial scores were associated with significantly poor recurrence-free survival across 10 years in the univariate analysis (n = 1,178, Immune Hotspot LR-χ2 = 14.06, P < 0.001; Table 3). When dichotomized, immune spatial scores were also prognostic (Immune Hotspot training set n = 589, p = 0.01, Hazard Ratio HR = 1.88 and 95% Confidence Interval CI = [1.16-3.07]; validation set n = 589, p = 0.002, HR = 2.21 [1.35-3.63], Figure 2, Supplementary Figure 4). In addition, immune spatial scores were significantly prognostic for early (0-5 year) and late recurrence (5-10 year) (Immune Hotspot 0-5 year: LR-χ2 = 6.24, P = 0.01; 5-10 year: LR-χ2 = 7.89, P = 0.005; Table 3). In the multivariate analysis adjusted for clinical variables as expressed by the CTS including node status, tumor size, grade, age and treatment, spatial scores remained prognostic for all three time windows, except for Cancer Hotspot for early recurrence and Immune-Cancer Hotspot for late recurrence (Table 3).
Table 3. Comparison of prognostic value of immune scores, classical clinical variables, and prognostic scores tests in univariate and multivariate analyses adjusted for nodal status, grade, size, age and treatment (all CTS components).
| Group | n | No. of Recurrences | Univariate | Multivariate* | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Immune Hotspot | Cancer Hotspot | Immune-Cancer Hotspot | Immune Hotspot | Cancer Hotspot | Immune-Cancer Hotspot | |||||||||||||||
| HR(95%CI) | LR-χ2 | P† | HR(95%CI) | LR-χ2 | P† | HR(95%CI) | LR-χ2 | P† | HR(95%CI) | LR-χ2 | P† | HR(95%CI) | LR-χ2 | P† | HR(95%CI) | LR-χ2 | P† | |||
| All patients | ||||||||||||||||||||
| 0-5 years | 1178 | 103 | 1.23 (1.06-1.43) | 6.24 | 0.01 | 1.31 (1.12-1.53) | 9.83 | 0.002 | 1.30 (1.14-1.49) | 11.46 | <0.001 | 1.13 (0.97-1.32) | 4.56 | 0.03 | 1.20 (1.02-1.41) | 2.36 | 0.12 | 1.20 (1.05-1.38) | 5.85 | 0.0156 |
| 5-10 years | 1019 | 113 | 1.26 (1.09-1.47) | 7.89 | 0.005 | 1.30 (1.11-1.52) | 9.34 | 0.002 | 1.24 (1.05-1.47) | 5.45 | 0.01 | 1.16 (1.00-1.35) | 3.61 | 0.05 | 1.21 (1.03-1.42) | 5.11 | 0.02 | 1.14 (0.97-1.35) | 2.27 | 0.13 |
| 0-10 years | 1178 | 216 | 1.25 (1.23-1.39) | 14.06 | <0.001 | 1.30 (1.17-1.46) | 19.16 | <0.001 | 1.28 (1.15-1.42) | 16.72 | <0.0001 | 1.15 (1.03-1.28) | 5.91 | 0.01 | 1.20 (1.08-1.35) | 9.59 | 0.002 | 1.17 (1.06-1.30) | 7.77 | 0.0053 |
| HER2-negative patients | ||||||||||||||||||||
| 0-5 years | 1037 | 78 | 1.23 (1.04-1.46) | 4.63 | 0.03 | 1.33 (1.11-1.59) | 8.50 | 0.003 | 1.26 (1.06-1.49) | 5.56 | 0.01 | 1.12 (0.94-1.34) | 1.49 | 0.22 | 1.21 (1.01-1.46) | 3.78 | 0.05 | 1.14 (0.96-1.36) | 2.00 | 0.16 |
| 5-10 years | 909 | 102 | 1.33 (1.14-1.54) | 10.9 | 0.001 | 1.39 (1.18-1.62) | 13.77 | <0.001 | 1.33 (1.12-1.57) | 8.63 | 0.003 | 1.23 (1.06-1.42) | 6.31 | 0.01 | 1.30 (1.11-1.53) | 9.09 | 0.002 | 1.22 (1.04-1.44) | 4.95 | 0.026 |
| 0-10 years | 1037 | 180 | 1.28 (1.15-1.44) | 15.10 | <0.001 | 1.36 (1.21-1.53) | 22.17 | <0.001 | 1.29 (1.14-1.45) | 13.98 | <0.001 | 1.18 (1.05-1.32) | 7.01 | 0.008 | 1.26 (1.12-1.42) | 12.49 | <0.001 | 1.18 (1.05-1.32) | 6.33 | 0.012 |
Adjusted for node, grade, tumor size, age, treatment. CI=confidence interval; HR=hazard ratio; CTS=clinical treatment score
Likelihood Ratio χ2, P-value two-sided [Please name the statistical test used to calculate the P values and say if it was two-sided.]
None of the patients in this study received trastuzumab if their tumors were HER2+, as is current practice. In the HER2- population (n=1,037), again spatial scores but not abundance scores were prognostic for all time windows (Table 3). In the multivariate analysis, spatial scores were prognostic for all time windows except for Immune Hotspot and Immune-Cancer Hotspot for early recurrence (Table 3). We henceforth focused on spatial scores only.
Comparison of immune spatial scores with RS, IHC4, ROR, and CTS
The prognostic value of spatial scores for late recurrence (5-10 year) is similar to that of IHC4 and RS in both the overall population (Immune Hotspot: LR-χ2 = 6.93, IHC: LR-χ2 = 6.75, RS: LR-χ2 = 6.79) and the HER2- population (Immune Hotspot LR-χ2 = 9.80, IHC4 LR-χ2 = 10.87, RS LR-χ2 = 7.78, Figure 3). None of these scores, however, added to IHC4 and RS for early recurrence (∆LR-χ2 ≤ 3.84) or was as prognostic as ROR and CTS in any time window (Figure 3). We then examined the additional prognostic value of spatial scores to IHC4 and RS for 0-10 year and late recurrence. Immune-Cancer Hotspot provided statistically significant prognostic value when added to IHC4 and RS for years 0-10 whereas Immune Hotspot also added prognostic value to IHC4 and RS for late recurrence (∆LR-χ2 > 3.84, Table 4). Cancer Hotspot, on the other hand, added prognostic information to IHC4 and RS in both time windows (Table 4). In the HER2- population, Immune Hotspot and Immune-Cancer Hotspot added statistically significant prognostic information to IHC4 and RS in both time windows (Table 4). Again, Cancer Hotspot added statistically significant information to IHC4 and RS in both time windows (Table 4). However, none of the spatial scores provide significant prognostic information beyond that of CTS or ROR in the HER2- or the overall population (∆LR-χ2 ≤ 3.84).
Figure 3. Barplots of likelihood scores for immune spatial and prognostic scores as well as combination of IHC4 and each immune spatial score (IHC4+I) for 0-10 and 5-10 year time window in A) overall population and B) HER2-population.
Kaplan-Meier curves were calculated and tested for equality using the log-rank test. The numbers of patients at risk in each group at various time points are given below each graph. All statistical tests were two-sided.
Table 4. Additional prognostic value of immune spatial scores to IHC4 and RS in all patients and Her2- subgroup.
| All patients | Immune Spatial Score [Correct?] | |||||
| 0-10 (N=963) | 5-10 (N=824) | |||||
| HR (95%CI) | ΔLR-χ2 | P* | HR (95%CI) | ΔLR-χ2 | P* | |
| Cancer Hotspot to IHC4 | 1.22 (1.08-1.37) | 9.72 | 0.002 | 1.24 (1.04-1.47) | 5.57 | 0.02 |
| Cancer Hotspot to RS | 1.25 (1.11-1.41) | 12.46 | <0.001 | 1.27 (1.07-1.49) | 6.97 | 0.008 |
| Immune Hotspot to IHC4 | 1.13 (1.01-1.27) | 3.93 | 0.05 | 1.19 (1.02-1.40) | 4.10 | 0.04 |
| Immune Hotspot to RS | 1.14 (1.02-1.28) | 4.50 | 0.03 | 1.21 (1.03-1.42) | 4.71 | 0.03 |
| Immune-Cancer Hotspot to IHC4 | 1.17 (1.04-1.31) | 5.87 | 0.02 | NS | NS | NS |
| Immune-Cancer Hotspot to RS | 1.15 (1.02-1.29) | 4.76 | 0.03 | NS | NS | NS |
| Her2 subgroup | 0-10 (N=848) | 5-10 (N=733) | ||||
| HR (95%CI) | ΔLR-χ2 | P* | HR (95%CI) | ΔLR-χ2 | P* | |
| Cancer Hotspot to IHC4 | 1.26 (1.11-1.43) | 11.73 | <0.001 | 1.32 (1.12-1.57) | 9.31 | 0.002 |
| Cancer Hotspot to RS | 1.29 (1.14-1.46) | 14.28 | <0.001 | 1.37 (1.16-1.61) | 11.82 | <0.001 |
| Immune Hotspot to IHC4 | 1.14 (1.02-1.30) | 4.56 | 0.03 | 1.23 (1.05-1.44) | 5.71 | 0.02 |
| Immune Hotspot to RS | 1.15 (1.02-1.30) | 4.84 | 0.03 | 1.26 (1.08-1.47) | 7.12 | 0.008 |
| Immune-Cancer Hotspot to IHC4 | 1.17 (1.04-1.33) | 5.64 | 0.02 | 1.24 (1.04-1.48) | 4.95 | 0.03 |
| Immune-Cancer Hotspot to RS | 1.13 (0.99-1.29) | 3.22 | 0.07 | 1.23 (1.03-1.47) | 4.69 | 0.03 |
Likelihood Ratio χ2, P-value two-sided; CI=confidence interval; HR=hazard ratio; RS=Oncotype DX 21-gene recurrence score.
IHC4+Immune spatial score
To evaluate a new biomarker based entirely on histology slides, we sought to determine the prognostic value of a combined model of IHC4 and spatial scores for 0-10 year and late recurrence. The prognostic value of IHC4 and RS are similar in the two-time windows and populations (∆LR-χ2 ≤ 3.84). For predicting recurrence across 10 years, Cancer Hotspot combined with IHC4 achieved a better performance than RS alone in both cohorts (overall population ∆LR-χ2= 9.23; in HER2-: ∆LR-χ2 = 6.54, Figure 3). For predicting late recurrence, prognostic value higher than RS was observed for IHC4 combined with any of the spatial score in both populations, except for Immune-Cancer Hotspot in the overall population (Figure 3). However, none of the combined scores outperformed CTS and ROR. Finally, IHC4 combined with spatial scores added prognostic value to CTS, similar to that achieved by the RS but lower than ROR in the overall population (∆LR-χ2: IHC4+Immune Hotspot = 22.64, IHC4+Immune-Cancer Hotspot = 22.96, IHC4+Cancer Hotspot = 23.38, RS = 23.53, ROR 29.18).
Discussion
In this study, we aimed to establish the prognostic value of immune scores for recurrence in ER+ breast cancer patients treated with anastrozole or tamoxifen. While immune response and immunotherapy for ER- diseases have been under the spotlight, correlation of TILs with outcomes in ER+ disease is less clear (16), with many studies reporting the lack of significant prognostic association (11, 15, 26, 27). In line with these reports, we did not find prognostic value in immune abundance scores that only account for the amount of TILs in the entire histology section or in specific tumor regions including intra-tumor, adjacent-tumor, and distal-tumor, as scored by digital histology slide image analysis. In contrast, our immune scores based on the spatial heterogeneity of TILs were highly prognostic, particularly for late recurrence after 5 years of endocrine therapy. This suggests a lasting memory of tumor immunity on disease progression and evolution of treatment resistance in ER+ cancer. Such spatial heterogeneity may reflect spatial distribution patterns of different immune cell subsets. Intra-tumor heterogeneity of cancer cells may also in turn influence immune spatial distribution through cytokine secretion and neoantigen presentation. While the biological mechanisms remain to be investigated, our finding has significant clinical implication, which suggests that TILs have been previously overlooked in ER+ diseases due to the lack of in-depth analysis of TILs on the tissue spatial organization level. Our findings highlight the importance of examining not just cell abundance but also spatial patterns that can be indicative of immune functional phenotypes and disease prognosis.
In contrast to our observation in ER- tumors (13), high Immune-Cancer Hotspot score, indicating increased spatial clustering in immune and cancer cells, correlated with poor prognosis in ER+ breast cancer. However, this is consistent with our previous finding that immune gene signature was associated with poor response to endocrine therapy in a neoadjuvant setting (28, 29). The difference may be due to immune composition and functionality in the two subtypes, and mechanisms by which immune response contributes to hormonal therapy resistance (30, 31). In a recent study of immune composition in 7,270 breast cancers, higher fraction of T-regulatory cells and M2 macrophages and lower fraction of M1 macrophages was found in ER+ compared with ER- cancers (12). Therefore, compared with ER- subtype, the immune landscape of ER+ subtype is characterized by increased T-cell regulation and macrophage polarization towards pro-tumorigenic M2, which is consistent with our results.
Strengths of this study include the large patient cohort with long-term follow-up systematically collected in a well-documented clinical trial, well-characterized samples that enables a direct comparison with established biomarkers, and fully automated and reproducible methods for immune scoring. These allowed us to evaluate quantitative immune scoring based entirely on H&E-stained tumor slides, which are readily generated as part of clinical routine. Comparison of the automated immune scores with a pathologist’s score following recommendations for TIL evaluation in breast cancer (24) in a subset of samples showed a weak correlation overall. This is, however, unsurprising when one considers that the automated scores include regions of the tumor that are excluded on pathological evaluation. Indeed, the latter includes stromal TILs only, which are assessed as percentage surface area that is inflammatory as opposed to fibroblastic (25). In contrast, the automated scores count either the absolute number of lymphocytes in relation to cancer cells, or the frequency of (co-)clustering of immune and cancer cells. Thus, the information gleaned from these methods potentially provides different biological information about the interaction between these cell types, and, as demonstrated herein, can provide valuable prognostic information. Developing a histology-based test as such thus has the advantages of cost effectiveness and general applicability. The prognostic value of immune scores for late recurrence is similar to that of IHC4 and RS which is the most widely used test for residual risk of recurrence following surgery and endocrine. Although they did not add prognostic value to CTS, in almost all occasions immune scores add prognostic information to the individual components that make up CTS, including node status, size, grade, age and treatment. Since CTS was developed and optimized using TransATAC samples (4), the independent prognostic value of immune scores remains to be validated in further endocrine adjuvant therapy studies with homogeneous treatments such as POETIC (32). In addition, the strong prognostic value of CTS and ROR could be partly explained by the use of tumor size in their calculation, which is a highly prognostic factor. Furthermore, IHC4 combined with immune score is statistically significantly more prognostic than RS, particularly for late recurrence. The clinical utility of IHC4+ immune scores as a combined histology-based test is well worth exploring.
Weaknesses of our study include the use of a single histology slide per tumor such that intra-tumor heterogeneity cannot be fully addressed, and the lack of immune markers that may provide further insights for immune functions in ER+ disease and better predictors due to little residual tissue available. In addition, whether our methods can sufficiently address challenges arising from variability in factors including fixation, staining, and acquisition in a clinical setting, need to be evaluated before implementation. Our results also only apply to women who are chemotherapy-free.
Furthermore, our study also provides relevant information for new treatment strategy in the high-risk ER+ population identified by immune scores. Our findings support different immunosuppressive mechanisms in the ER+ and ER-subtypes, and in light of these results call for the development of novel cancer therapeutics targeting the pathways that reverse these mechanisms specifically for ER+ disease. This may also help explain why anti-PD1 checkpoint inhibition, despite demonstrating activity as monotherapy in early phase trials in ER+ breast cancer, had low response rates compared with triple-negative breast cancer and highly variable among trials (33, 34). Further, we speculate that immune scores may be useful as predictive biomarkers for immunotherapy, given the limited clinical utility of PDL1 expression in guiding patient selection (35).
In summary, enabled by fully automated image analysis of histology sections, our study provided an additional dimension to the analysis of immune functional phenotype in breast cancer. Spatial data provided by histology, once quantitatively analyzed, will aid the identification of clinically relevant features, potentially yielding predictions more powerful than measurements of cell abundance that ignore the spatial context.
Supplementary Material
Acknowledgments
Funding
This work was supported by the Royal Marsden National Institutes of Health Biomedical Research Centre grant A105. MD and YY acknowledge support by the Royal Marsden National Institutes of Health Biomedical Research Centre. MD acknowledges a Breast Cancer Now grant (CTR-Q4-Y1). JC and IS acknowledge a CRUK grant (C569/A16891). YY acknowledges support by CRUK (C45982/A21808) and Wellcome Trust (105104/Z/14/Z).
Notes
The funder had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.
Acknowledgement and disclosure: We thank Andrew Dodson and Daniel Nava Rodrigues for technical and pathological support. M. Dowsett has received commercial research grants and speakers bureau honoraria from AstraZeneca.
Footnotes
Author contribution: MD and YY designed the experiments. AH, IS, KN and JC performed the analyses. AH, IS and YY wrote the manuscript. All authors have approved the manuscript.
References
- 1.Dowsett M, Goldhirsch A, Hayes DF, Senn HJ, Wood W, Viale G. International Web-based consultation on priorities for translational breast cancer research. Breast cancer research : BCR. 2007;9(6):R81. doi: 10.1186/bcr1798. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. The New England journal of medicine. 2004;351(27):2817–26. doi: 10.1056/NEJMoa041588. [DOI] [PubMed] [Google Scholar]
- 3.Nielsen TO, Parker JS, Leung S, Voduc D, Ebbert M, Vickery T, et al. A comparison of PAM50 intrinsic subtyping with immunohistochemistry and clinical prognostic factors in tamoxifen-treated estrogen receptor-positive breast cancer. Clinical cancer research : an official journal of the American Association for Cancer Research. 2010;16(21):5222–32. doi: 10.1158/1078-0432.CCR-10-1282. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Cuzick J, Dowsett M, Pineda S, Wale C, Salter J, Quinn E, et al. Prognostic value of a combined estrogen receptor, progesterone receptor, Ki-67, and human epidermal growth factor receptor 2 immunohistochemical score and comparison with the Genomic Health recurrence score in early breast cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2011;29(32):4273–8. doi: 10.1200/JCO.2010.31.2835. [DOI] [PubMed] [Google Scholar]
- 5.Sestak I, Dowsett M, Zabaglo L, Lopez-Knowles E, Ferree S, Cowens JW, et al. Factors predicting late recurrence for estrogen receptor-positive breast cancer. Journal of the National Cancer Institute. 2013;105(19):1504–11. doi: 10.1093/jnci/djt244. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Sestak I, Dowsett M, Ferree S, Baehner FL, Cuzick J. Retrospective analysis of molecular scores for the prediction of distant recurrence according to baseline risk factors. Breast cancer research and treatment. 2016;159(1):71–8. doi: 10.1007/s10549-016-3868-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Savas P, Salgado R, Denkert C, Sotiriou C, Darcy PK, Smyth MJ, et al. Clinical relevance of host immunity in breast cancer: from TILs to the clinic. Nature Reviews Clinical Oncology. 2016;13(4):228–41. doi: 10.1038/nrclinonc.2015.215. [DOI] [PubMed] [Google Scholar]
- 8.Denkert C, von Minckwitz G, Brase JC, Sinn BV, Gade S, Kronenwett R, et al. Tumor-Infiltrating Lymphocytes and Response to Neoadjuvant Chemotherapy With or Without Carboplatin in Human Epidermal Growth Factor Receptor 2–Positive and Triple-Negative Primary Breast Cancers. Journal of Clinical Oncology. 2015;33:983–91. doi: 10.1200/JCO.2014.58.1967. [DOI] [PubMed] [Google Scholar]
- 9.Denkert C, Loibl S, Noske A, Roller M, Müller BM, Komor M, et al. Tumor-associated lymphocytes as an independent predictor of response to neoadjuvant chemotherapy in breast cancer. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2010;28:105–13. doi: 10.1200/JCO.2009.23.7370. [DOI] [PubMed] [Google Scholar]
- 10.Gu-Trantien C, Loi S, Garaud S, Equeter C, Libin M, de Wind A, et al. CD4(+) follicular helper T cell infiltration predicts breast cancer survival. The Journal of clinical investigation. 2013;123:2873–92. doi: 10.1172/JCI67428. Epub 2013/06/20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Loi S, Sirtaine N, Piette F, Salgado R, Viale G, Van Eenoo F, et al. Prognostic and predictive value of tumor-infiltrating lymphocytes in a phase III randomized adjuvant breast cancer trial in node-positive breast cancer comparing the addition of docetaxel to doxorubicin with doxorubicin-based chemotherapy: BIG 02-98. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2013;31:860–7. doi: 10.1200/JCO.2011.41.0902. Epub 2013/01/24. [DOI] [PubMed] [Google Scholar]
- 12.Bense RD, Sotiriou C, Piccart-Gebhart MJ, Haanen JB, van Vugt MA, de Vries EG, et al. Relevance of Tumor-Infiltrating Immune Cell Composition and Functionality for Disease Outcome in Breast Cancer. Journal of the National Cancer Institute. 2017;109(1) doi: 10.1093/jnci/djw192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Nawaz S, Heindl A, Koelble K, Yuan Y. Beyond immune density: critical role of spatial heterogeneity in estrogen receptor-negative breast cancer. Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc. 2015 doi: 10.1038/modpathol.2015.133. [DOI] [PubMed] [Google Scholar]
- 14.Dieci MV, Criscitiello C, Goubar A, Viale G, Conte P, Guarneri V, et al. Prognostic value of tumor-infiltrating lymphocytes on residual disease after primary chemotherapy for triple-negative breast cancer: a retrospective multicenter study. Annals of oncology : official journal of the European Society for Medical Oncology / ESMO. 2014;25:611–8. doi: 10.1093/annonc/mdt556. Epub 2014/01/10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Loi S, Michiels S, Salgado R, Sirtaine N, Jose V, Fumagalli D, et al. Tumor infiltrating lymphocytes are prognostic in triple negative breast cancer and predictive for trastuzumab benefit in early breast cancer: Results from the FinHER trial. Annals of Oncology. 2014;25:1544–50. doi: 10.1093/annonc/mdu112. [DOI] [PubMed] [Google Scholar]
- 16.Luen S, Virassamy B, Savas P, Salgado R, Loi S. The genomic landscape of breast cancer and its interaction with host immunity. Breast. 2016;29:241–50. doi: 10.1016/j.breast.2016.07.015. [DOI] [PubMed] [Google Scholar]
- 17.Yuan Y, Failmezger H, Rueda OM, Ali HR, Graf S, Chin S-FF, et al. Quantitative image analysis of cellular heterogeneity in breast tumors complements genomic profiling. Science translational medicine. 2012;4:157ra43. doi: 10.1126/scitranslmed.3004330. Epub 2012/11/24. [DOI] [PubMed] [Google Scholar]
- 18.Yuan Y. Modelling the Spatial Heterogeneity and Molecular Correlates of Lymphocytic Infiltration in Triple-Negative Breast Cancer. Journal of the Royal Society: Interface. 2015;12 doi: 10.1098/rsif.2014.1153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Yuan Y. Spatial Heterogeneity in the Tumor Microenvironment. Cold Spring Harbor perspectives in medicine. 2016;6(8) doi: 10.1101/cshperspect.a026583. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Demaria S, Kawashima N, Yang AM, Devitt ML, Babb JS, Allison JP, et al. Immune-mediated inhibition of metastases after treatment with local radiation and CTLA-4 blockade in a mouse model of breast cancer. Clinical cancer research : an official journal of the American Association for Cancer Research. 2005;11:728–34. Epub 2005/02/11. [PubMed] [Google Scholar]
- 21.Fridman WH, Pages F, Sautes-Fridman C, Galon J. The immune contexture in human tumours: impact on clinical outcome. Nature reviews Cancer. 2012;12:298–306. doi: 10.1038/nrc3245. Epub 2012/03/16. [DOI] [PubMed] [Google Scholar]
- 22.Gentles AJ, Newman AM, Liu CL, Bratman SV, Feng W, Kim D, et al. The prognostic landscape of genes and infiltrating immune cells across human cancers. Nature medicine. 2015 doi: 10.1038/nm.3909. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dowsett M, Cuzick J, Wale C, Forbes J, Mallon EA, Salter J, et al. Prediction of risk of distant recurrence using the 21-gene recurrence score in node-negative and node-positive postmenopausal patients with breast cancer treated with anastrozole or tamoxifen: a TransATAC study. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2010;28(11):1829–34. doi: 10.1200/JCO.2009.24.4798. [DOI] [PubMed] [Google Scholar]
- 24.Salgado R, Denkert C, Demaria S, Sirtaine N, Klauschen F, Pruneri G, et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Annals of oncology : official journal of the European Society for Medical Oncology / ESMO. 2014 doi: 10.1093/annonc/mdu450. mdu450. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature. 2012;486:346–52. doi: 10.1038/nature10983. Epub 2012/04/24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Dieci MV, Criscitiello C, Goubar a, Viale G, Conte P, Guarneri V, et al. Prognostic value of tumor-infiltrating lymphocytes on residual disease after primary chemotherapy for triple-negative breast cancer: a retrospective multicenter study. Annals of oncology : official journal of the European Society for Medical Oncology / ESMO. 2014;25:611–8. doi: 10.1093/annonc/mdt556. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ali HR, Provenzano E, Dawson SJ, Blows FM, Liu B, Shah M, et al. Association between CD8+T-cell infiltration and breast cancer survival in 12 439 patients. Annals of Oncology. 2014;25(8):1536–43. doi: 10.1093/annonc/mdu191. [DOI] [PubMed] [Google Scholar]
- 28.Dunbier AK, Ghazoui Z, Anderson H, Salter J, Nerurkar A, Osin P, et al. Molecular profiling of aromatase inhibitor-treated post-menopausal breast tumors identifies immune-related correlates of resistance. Clinical cancer research : an official journal of the American Association for Cancer Research. 2013 doi: 10.1158/1078-0432.CCR-12-1000. Epub 2013/03/16. [DOI] [PubMed] [Google Scholar]
- 29.Gao Q, Patani N, Dunbier AK, Ghazoui Z, Zvelebil M, Martin LA, et al. Effect of aromatase inhibition on functional gene modules in estrogen receptor-positive breast cancer and their relationship with antiproliferative response. Clinical cancer research : an official journal of the American Association for Cancer Research. 2014;20(9):2485–94. doi: 10.1158/1078-0432.CCR-13-2602. [DOI] [PubMed] [Google Scholar]
- 30.Wei C, Cao Y, Yang X, Zheng Z, Guan K, Wang Q, et al. Elevated expression of TANK-binding kinase 1 enhances tamoxifen resistance in breast cancer. Proceedings of the National Academy of Sciences of the United States of America. 2014;111(5):E601–10. doi: 10.1073/pnas.1316255111. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Jansen MP, Foekens JA, van Staveren IL, Dirkzwager-Kiel MM, Ritstier K, Look MP, et al. Molecular classification of tamoxifen-resistant breast carcinomas by gene expression profiling. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2005;23(4):732–40. doi: 10.1200/JCO.2005.05.145. [DOI] [PubMed] [Google Scholar]
- 32.Dowsett M, Smith I, Robertson J, Robison L, Pinhel I, Johnson L, et al. Endocrine therapy, new biologicals, and new study designs for presurgical studies in breast cancer. Journal of the National Cancer Institute - Monographs. 2011:120–3. doi: 10.1093/jncimonographs/lgr034. [DOI] [PubMed] [Google Scholar]
- 33.Dirix LY, Takacs I, Nikolinakos P, Jerusalem G, Arkenau HT, Hamilton EP, et al. Avelumab (MSB0010718C), an anti-PD-L1 antibody, in patients with locally advanced or metastatic breast cancer: A phase 1b JAVELIN solid tumor trial. Cancer research. 2016;76 doi: 10.1007/s10549-017-4537-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Rugo HS, Delord JP, Im SA, Ott PA, Piha-Paul SA, Bedard PL, et al. Preliminary efficacy and safety of pembrolizumab (MK-3475) in patients with PD-L1 positive, estrogen receptor-positive (ER+)/HER2-negative advanced breast cancer enrolled in KEYNOTE-028. Cancer research. 2016;76 [Google Scholar]
- 35.Patel SP, Kurzrock R. PD-L1 Expression as a Predictive Biomarker in Cancer Immunotherapy. Molecular cancer therapeutics. 2015;14(4):847–56. doi: 10.1158/1535-7163.MCT-14-0983. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.