IRSN-23 Gene Diagnosis Predicts Response to Neoadjuvant Chemotherapy and Refines Breast Cancer Subtype Classification (Prospective Analyses)

doi:10.21203/rs.3.rs-4860616/v1

Download PDF

Research Article

IRSN-23 Gene Diagnosis Predicts Response to Neoadjuvant Chemotherapy and Refines Breast Cancer Subtype Classification (Prospective Analyses)

https://doi.org/10.21203/rs.3.rs-4860616/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Introduction:

This study evaluates the reproducibility of the IRSN-23 model, which classifies patients into highly chemotherapy-sensitive (Gp-R) or less-sensitive (Gp-NR) groups based on immune-related gene expression using DNA microarray analysis, and its impact on breast cancer subtype classification.

Methods:

Tumor tissues from 146 breast cancer patients receiving preoperative chemotherapy (paclitaxel-FEC) ± trastuzumab at Osaka University Hospital (OUH) were used to classify patients into Gp-R or Gp-NR using IRSN-23. The ability to predict a pathological complete response (pCR) was assessed and the results were validated with independent public datasets (N = 1,282).

Results:

In the OUH dataset, the pCR rate was significantly higher in the Gp-R group than in the Gp-NR group without trastuzumab (29 versus 1%, P = 1.70E-5). In all validation sets without anti-HER2 therapy, the pCR rate in the Gp-R group was significantly higher than that in the Gp-NR group. The pooled analysis of the validation set showed higher pCR rates in the Gp-R group than in the Gp-NR group, both without (N = 1103, 40 versus 12%, P = 2.02E-26) and with (N = 304, 49 versus 35%, P = 0.017) anti-HER2 therapy. Collaboration analyses of IRSN-23 and OncotypeDx or PAM50 could identify highly chemotherapy-sensitive groups and refine breast cancer subtype classification based on the tumor microenvironment (offensive factor - PAM50 and defensive factor - IRSN-23), and the immune subtype was correlated with a better prognosis after NAC.

Conclusions:

This study offers prospective analyses of IRSN-23 in predicting chemotherapy efficacy, showing high reproducibility. The findings indicate the clinical value of using IRSN-23 for refining breast cancer subtype classification, with implications for personalized treatment strategies and improved patient outcomes.

breast cancer

immune signature

neoadjuvant chemotherapy

pathological complete response

intrinsic subtype

Neoadjuvant chemotherapy (NAC) has been widely used in clinical practice for advanced breast cancer since its introduction in a clinical trial more than 40 years ago (1). The primary objectives of NAC are to enhance breast conservation rates and determine drug sensitivity (2). Recently, “response-guided therapy” has gained prominence, in which postsurgical drug treatment is adjusted based on the presence or absence of residual tumors (3, 4). This approach has been increasingly recognized for its significance and has been demonstrated in various trials. The pathological complete response (pCR) has been identified as the factor most strongly associated with prognosis in advanced breast cancer (5), particularly in HER2-positive (6) and triple-negative breast cancer (7). However, the pCR rate following NAC remains relatively low at 30–50% (2). Accurate prediction of treatment response before therapy is desirable; however, despite the investigation of several response predictors, a clinically appropriate level of testing has not yet been developed.

Previous studies (8–15) have attempted to predict the efficacy of anthracyclines and taxanes; however, these methods have not been adopted in clinical practice because of the insufficient accuracy and reproducibility of efficacy prediction. Consequently, there is a need for efficient mathematical combinations of sophisticated genetic markers to achieve highly accurate predictions. In 2014, we introduced the IRSN-23 model (immune-related 23 Gene Signature for NAC) (16), which classifies patients receiving anthracycline and/or taxane without anti-HER2 therapy into two groups based on the expression patterns of 23 probes (19 immune-related genes) using the Affymetrix Human Genome U133 Plus 2.0 chip, which is highly sensitive to chemotherapy (Gp-R, immune score (IS) > 0) and less sensitive to chemotherapy (Gp-NR, IS ≤ 0). Through retrospective validation using in-house data and a large number (n = 901) of public datasets, we demonstrated that the model can predict chemotherapy sensitivity with high accuracy and statistical significance for all breast cancer subtypes.

This study aimed to prospectively analyze newly collected data from Osaka University Hospital (OUH, n = 146) and over 1,282 freshly uploaded public data from the past decade after developing the IRSN-23 gene signature as an independent validation set. The primary objective was to demonstrate the reproducibility and clinical utility of the IRSN-23 model. The secondary objective was to evaluate the correlation between IRSN-23 and clinicopathological factors, including various immunohistochemical (IHC) staining and whole-exome sequencing (WES) results, and its impact on breast cancer subtypes. Our results indicated that IRSN-23 has high predictive accuracy for chemotherapy sensitivity and can reliably identify patients who are likely to benefit from NAC without anti-HER2 therapy (previous report (n = 901): Gp-R: 39% vs. Gp-NR: 11%, P = 4.98E-23; current report (n = 1,103): Gp-R: 40% vs. Gp-NR: 12%, P = 1.24E-26). Our study underscores the potential clinical application of IRSN-23 in identifying new breast cancer subtypes that are responsive to treatment and could contribute to the personalized treatment of patients with breast cancer.

Patients and Samples

Prospective OUH Dataset

Tumor specimens were collected from patients with breast cancer (stages II-IV, n = 146) who received preoperative chemotherapy at the Osaka University Hospital (OUH) between 2010 and 2017. Gene expression analysis was performed using Affymetrix U133A Plus 2.0 DNA microarray since the last publication. The chemotherapy regimen involved 12 weekly doses of paclitaxel (80 mg/m²), followed by four cycles of 5-FU (500 mg/m²), epirubicin (75 mg/m²), and cyclophosphamide (500 mg/m²) every three weeks. The HER2-positive patients were treated with paclitaxel (PTX) and trastuzumab. Table 1 provides the details of the patients’ backgrounds. The pathological responses were assessed using surgical specimens obtained after chemotherapy. All patients provided informed consent before tumor biopsy, and the Ethics Committee approved the study (Approval Number: K14190). Approval No.K14190 is an observational study designed to develop a drug-sensitivity diagnostic method based on gene expression and mutation analyses of tumor tissues collected before preoperative chemotherapy. This study specifically focused on patients with stage I-IV breast cancer scheduled to receive preoperative chemotherapy. RNA extraction and DNA microarray analysis methods have been described previously (16).

Prospective Public Dataset

To ensure the validity of IRSN-23 predictions, we re-evaluated their accuracy using available datasets from the Gene Expression Omnibus (GEO) downloaded through January 2022, following the methodology described in our previous publication (16). We applied the following eligibility criteria: (i) total RNA expression data were obtained from DNA microarrays before treatment; (ii) the dataset had objective efficacy determination following treatment, and the results were retrieved; (iii) the dataset did not report IRSN-23 results; and (iv) the dataset did not include RNA extracted from formalin-fixed paraffin-embedded (FFPE) specimens. Previous studies have demonstrated that RNA extraction from FFPE tissues significantly alters immune-related gene expression. In contrast to our previous analysis (16), this study preoperatively included patients who received preoperative anti-HER2 therapy for breast cancer. The external validation set consists of 1282 cases from eight datasets (GSE25066, GSE28844, GSE66399, GSE140494, GSE21974, GSE130788, GSE4779, and GSE34138). Twenty-one cases were excluded due to unknown efficacy following chemotherapy, including 1261 cases for analysis. The patient characteristics are summarized in Table 1.

Experiment Design

Experiment 1: Validation of Reproducibility

This study aimed to validate the reproducibility of IRSN-23, a gene expression signature composed of 23 probes targeting 19 genes, to predict treatment efficacy in patients with cancer. To accomplish this, we evaluated the performance of IRSN-23 using microarray data from 146 patients, including 34 who received trastuzumab from our institution and 860 patients from GSE25066, GSE28844, GSE37946, GSE42822, GSE66399, and GSE140494, all of which utilized the Human Genome U133 Plus 2.0 platform, for external validation.

To assess the robustness of IRSN-23, we collected microarray data from 320 patients, including 110 from various platforms, such as GSE4779 using the Affymetrix Human X3P Array, GSE21974, and GSE34138 using RNA.Seq-014850 Whole Human Genome Microarray 4×44K G4112F, and GSE130788 using the Illumina Human WG-6 v3.0 expression bead chip and compared their accuracy.

Furthermore, IRSN-23 calculated via microarray was measured using RNA-Seq (n = 43), and their concordance rates were compared.

Experiment 2: Predictors of Chemotherapy Sensitivity and Clinical Utility

IRSN-23 and each factor (clinical factors and factors by IHC, TMB, and HRD by WES) were compared regarding sensitivity to chemotherapy. In addition, IRSN-23 IS collaborated with Oncotype Dx RS for ER + HER2 − subtype to investigate its clinical usefulness.

Experiment 3: Characterization of IRSN-23 and Its Impact on Breast Cancer Subtype Classification

In addition to IRSN-23 and clinicopathological factors, this study also investigated the presence or absence of various immune cells, including tumor-infiltrating lymphocytes (TILs), regulatory T cells (Foxp3-positive T cells), CD8 + T cells, and IL17F + cells (helper T cells), as well as the pre-treatment neutrophil-lymphocyte ratio (NLR), pre-treatment platelet-to-lymphocyte ratio (PLR), HRD score obtained from WES analysis, and TMB. The number of cases included in the study has increased since the previous report, with 117 training and validation cases reported in a previous report. Furthermore, we examined the effects of IRSN-23 on breast cancer subtype classification.

Measurement Methods for Each Factor

Immune-Related 23 Gene Signature for NAC Using DNA Microarray

In a previous study (16), we developed the IRSN-23, which consists of 23 probes (19 genes) and is now available on GitHub (https://github.com/SNlaboratory/irsn23), and using supervised analysis with DLDA (Diagonal Linear Discriminant Analysis) in the R package ‘supclust’ (https://github.com/cran/supclust), as shown in Supplementary Figure S1. Subsequently, we patented it (Publication of US20150066379A1). The previously reported IRSN-23 was directly adapted to evaluate the accuracy of Affymetrix microarray. When adapting the microarrays to Affymetrix X3P, Agilent, and Illumina, 16 of the 19 genes in the IRSN-23 were supported. Therefore, we adapted IRSN-23 coefficients recalculated within the training set with 16 genes. The expression data obtained from the raw data were mean-centered, and Z-scaling was adapted for datasets for which the raw data were unavailable. The IRSN-23 IS was calculated by Z-scaling as a pre-processing step by aggregating the expression data of 23 probes to 19 genes to fit different expression platforms such as RNA-Seq, where expression data are calculated on a per-gene basis. Pearson’s correlation coefficient between the IRSN-23 score (23 probes) as the original model and the IRSN-23 score (19 genes) as the IS in the OUH Internal Validation (n = 59) used in a previous report was 0.99.

TMB and HRD

TMB was calculated as the ratio of the total number of mutations detected to the total size of the coding region (SureSelect Human All Exon V6, Agilent Technologies, 357MB) (17). The HRD score was determined as the sum of three factors (NtAI, LST, and HRD-LOH) (17).

OncotypeDx and Intrinsic Subtype

The OncotypDx RS Code for Affymetrix microarrays was obtained from the ‘genefu’ R package (https://github.com/bhklab/genefu). PAM50 was also employed to classify breast cancer as luminal A (LumA), luminal B (LumB), HER2-enriched (HER2), or basal-like (basal) following a previously described method (18). Tumors were classified into clusters closest to each subtype, with the exception of the normal breast-like subtype.

Immunohistological Examination

Immunohistochemical assays for ER and progesterone receptor (PR) were conducted as previously described, with cutoff values of 1% for ER and PR (16). HER2 expression was analyzed via fluorescence in situ hybridization according to the ASCO/CAP 2018 guidelines (19), and TILs were evaluated using hematoxylin and eosin-stained sections according to the International TILs Working Group 2014 (20). A TILs of > 20% was defined as high and that of < 20% as low. Tumors were classified into four subtypes: ER + HER2 −, ER(+), and HER2(−); ER + HER2+, ER(+), and HER2(+); ER-HER2+, ER(−), and HER2(+); and ER-HER2 −, ER(−), and HER2(−). pCR for NAC was defined as the complete disappearance of invasive tumor cells in the breast regardless of the presence or absence of ductal carcinoma in situ in the breast and a negative lymph node.

Statistical Analysis

All statistical analyses were conducted using R software version 3.6.3. The association between various parameters was evaluated using the chi-square or Fisher’s exact test. The DeLong test was used to compare the area under the curve (AUCs). All tests were two-tailed, and P < 0.05 indicated statistical significance. Pearson’s correlation coefficients and the averaging method were employed for unsupervised cluster analysis, and the results are shown as heat maps. Gene set enrichment analysis (GSEA) was performed on genes with principal component loadings greater than 0.10 in absolute value. GSEA was performed using DAVID Functional Annotation Bioinformatics (https://david.ncifcrf.gov/). Kaplan–Meier (KM) survival analysis for distant recurrence-free survival for each breast cancer subtype was also performed. The log-rank test was used to compare the KM curves.

Reproducibility

Prospective Validation Datasets Without Anti-HER2 Therapy Using Affymetrix Human Genome U133 Arrays

The IRSN-23 signature was used in several independent datasets to classify breast cancer patients into Gp-R or Gp-NR. In the OUH prospective validation set (n = 112), patients were classified into the Gp-R (n = 42, 38%) or Gp-NR (n = 70, 63%) groups. ER + HER2- accounted for 74 cases, and only 4 of them achieved pCR. In the prospective analysis at OUH, there were a total of 13 cases (pCR% = 11.6%) of pCR in the group without anti-HER2 therapy. The Gp-R group exhibited a significantly higher pCR rate than the Gp-NR group (29% vs. 1%, P = 2.35E-05), as illustrated in Supplementary Figure S2A and detailed in Supplementary Table S1, which also shows the accuracy of the pCR predictions. Similarly, in the GSE25066 (n = 488), GSE28844 (n = 32), GSE42822 (n = 66), and GSE140494 (n = 91) datasets, IRSN-23 was used to classify patients into Gp-R or Gp-NR groups. In the GSE25066 cohort, the Gp-R group had a significantly higher pCR rate than the Gp-NR group (38% vs. 10%, P = 8.39E-13) (Supplementary Figure S2B and Supplementary Table S1). Additionally, in the GSE28844 cohort, the Gp-R group had a significantly higher pCR rate than the Gp-NR group (29% vs. 0%, P = 0.028) (Supplementary Figure S2C and Supplementary Table S1). Similarly, in the GSE42822 cohort, the Gp-R group exhibited a significantly higher pCR rate than the Gp-NR group (65% vs. 26%, P = 6.55E-03), as illustrated in Supplementary Figure S2D and detailed in Supplementary Table S1. The same trend was observed in the GSE140494 cohort (n = 91), in which the Gp-R group had a significantly higher pCR rate than the Gp-NR group (39% vs. 16%, P = 0.03), as shown in Supplementary Figure S2E and Supplementary Table S1.

Prospective External Validation Datasets Without Anti-HER2 Therapy Using Different Microarrays

A different DNA microarray platform was used to validate the reproducibility of the genes extracted from the IRSN-23. IRSN-23 was applied to GSE4779 (n = 102) of Affymetrix Human X3P arrays, and the pCR rate of Gp-R was higher than that of Gp-NR (54 vs. 24%, P = 3.53E-03), as shown in Supplementary Figure S3A and Supplementary Table S1. Consistent with these results, Agilent (GSE21974, n = 32) and Illumina (GSE34138, n = 178) DNA microarrays also showed higher pCR rates in the Gp-R group than in the Gp-NR group (P = 3.22E-03 and P = 3.63E-05, respectively) (Supplementary Figure S3B-C and Supplementary Table S1).

Datasets of NAC with Anti-HER2 Therapy

In patients receiving NAC with anti-HER2 therapy, there were no significant differences in pCR rates between the Gp-R and Gp-NR groups in multiple datasets: the OUH prospective validation set (n = 34, Gp-R 78% vs. Gp-NR 63%, P = 0.46), GSE37946 (n = 50, Gp-R 64% vs. Gp-NR 50%, P = 0.55), GSE42822 (n = 25, Gp-R 14% vs. Gp-NR 61%, P = 0.073), and GSE66399 (n = 88, Gp-R 43% vs. Gp-NR 23%, P = 0.059) (Supplementary Figure S4). In addition, in GSE130788 with anti-HER2 therapy (n = 110), there was no significant difference in pCR rates between Gp-R and Gp-NR (52% vs. Gp-NR 35%, P = 0.084) in the pre-treatment samples. Interestingly, however, when Gp-R patients reconfirmed with IRSN-23 using samples obtained 2-weeks after treatment were transferred to the baseline Gp-R group, the pCR rate in the Gp-R group was significantly higher than that of Gp-NR (P = 0.025). These results indicate a remarkable change in the immune microenvironment captured by IRSN-23 within 2-weeks of the start of anti-HER2 therapy. Refer to Supplementary Figure S4E-F and Supplementary Table S1 for details.

Pooled Analysis of the current Independent Validation Sets and Meta-analyses without Anti-HER2 Therapy in All Previous and Current datasets

The results of the pooled analyses showed that the pCR rates in the independent OUH datasets and public datasets without anti-HER2 therapy were significantly higher in the Gp-R group (40% (170/424] ) than in the Gp-NR group (12% (83/679) (P = 2.02E-26), as presented in Supplementary Figure S5A and Supplementary Table S1. Similarly, in patients treated with anti-HER2 therapy, a significant difference was observed in the pCR rate between the Gp-R (49%, 61/120) and Gp-NR (35%, 64/184) groups (P = 0.017), as shown in Supplementary Fig. 5B and Supplementary Table S1.

Furthermore, in a meta-analysis combining all datasets from previous and current independent validation sets, Gp-R was more likely to achieve pCR than Gp-NR (OR 5.24 (4.00–6.85)) in all breast cancer subtypes not treated with anti-HER2 therapy (Fig. 1A). Consistent with previous reports, Gp-R was a significant predictor of pCR in all cohorts, although each cohort received different treatments. When comparing subtypes and treatments, Gp-R was more likely to achieve pCR than Gp-NR in ER + HER2-, ER ± HER2+, ER-HER2- breast cancer subtypes, anthracycline, and anthracycline-taxane therapy, with odds ratios of 4.49 (2.92–6.89), 3.91 (2.04–7.48), 3.23 (2.24–4.66), 5.72 (3.68–8.90), 4.92 (3.79–6.37) respectively, as shown in Fig. 1B. When the relationship between IRSN-23 IS changes and pCR was confirmed by local regression and Loess regression fitting, it was found that a lower IS was less likely to result in pCR and a higher IS was more likely to result in pCR, although the trend varied by subtype (Fig. 1C).

Predictors of Chemotherapy Sensitivity and Clinical Utility

Univariate analyses for pCR in the OUH datasets without trastuzumab (n = 229) showed that several factors were significantly associated with pCR, including IRSN-23 Gp-R (OR = 50.11, P = 1.12E-08), Oncotype Dx RS > 25 (OR = 8.29, P = 1.63E-04), ER-negative (OR = 7.92, P = 1.34E-08), PgR-negative (OR = 6.65, P = 2.29E-06), \(\:\text{H}\text{R}\text{D}\:\left(\text{h}\text{o}\text{m}\text{o}\text{l}\text{o}\text{g}\text{o}\text{u}\text{s}\:\text{r}\text{e}\text{c}\text{o}\text{m}\text{b}\text{i}\text{n}\text{a}\text{t}\text{i}\text{o}\text{n}\:\text{d}\text{e}\text{f}\text{i}\text{c}\text{i}\text{e}\text{n}\text{c}\text{y}\right)\ge\:42\) (OR = 6.00, P = 0.003), \(\:\text{K}\text{i}67\ge\:20\text{\%}\) (OR = 3.65, P = 0.004), cT 1 or 2 (OR = 3.15, P = 0.049), TILs-positive (OR = 3.04, P = 0.005), age<50 (OR = 2.61, P = 0.017), and HER2-positive (OR = 2.29, P = 0.045), as shown in Fig. 2A. In the analysis limited to ER+HER2- breast cancer, IRSN-23 Gp-R (OR = 28.54, P = 0.009), \(\:\text{H}\text{R}\text{D}\ge\:42\) (OR = 20.14, P = 0.006), and Oncotype Dx RS > 25 (OR = 18.74, P = 0.014) were significantly associated with pCR (Fig. 2B).

We also investigated the collaboration between the OncotypeDx RS (recurrence score) and IRSN-23 IS in ER + HER2 − breast cancer. The most NAC-sensitive group was ‘\(\:\text{I}\text{S}>-25+\text{R}\text{S}>25\)’ (pCR%: 20% (79/405), followed by ‘\(\:\text{I}\text{S}>-25+\text{R}\text{S}<25\)’ (pCR%: 7% (12/165)). The other two groups were less sensitive, with pCR% of 5% (9/192) for ‘\(\:\text{I}\text{S}\le\:-25+\text{R}\text{S}\le\:25\)’ and 4% (3/70) for ‘\(\:\text{I}\text{S}\le\:-25+\text{R}\text{S}>25\)’, as shown in Fig. 2C. Interestingly, despite RS>25 being more likely to be eligible for chemotherapy, the IRSN-23 \(\:\text{I}\text{S}\le\:-25\) group was less sensitive to NAC than the IRSN-23 \(\:\text{I}\text{S}>-25\) group (‘\(\:\text{I}\text{S}>-25+\text{R}\text{S}>25\)’: 20% vs. ‘\(\:\text{I}\text{S}\le\:-25+\text{R}\text{S}>25\)’: 4%, P = 3.28E-03). The logistic regression model between OncotypeDx RS and IRSN-23 IS (RI-Score: recurrence and immune score), and their confidence intervals are shown in Supplementary Figure S6. Additionally, the ROC curve analysis for pCR in ER+HER2- breast cancer for all previous and current validation sets demonstrated that the RI-Score was significantly more accurate in predicting pCR than OncotypeDx RS alone (area under the curve (AUC): RI-Score = 0.753 vs. OncotypeDx RS = 0.703, P = 0.004), as presented in Fig. 2D.

Characterization of IRSN-23 and Its Impact on Breast Cancer Subtype Classification

The association between IRSN-23 Gp-R and clinicopathological factors was previously reported, but we re-examined this in additional cases. We found that age > 50 years (P = 1.37E-04), HG3 (P = 0.002), Ki67 ≥ 20% (P = 8.64E-05), ER-negative (P = 1.93E-13), and PgR-negative (P = 2.99E-10) were correlated with Gp-R. TILs-positive (P = 2.97E-04) was correlated with CD8 + lymphocyte-positive (P = 0.037) and Foxp3 + lymphocyte-positive cells as a marker for regulatory T cells (P = 0.012) but not with the ratio of lymphocytes in the peripheral blood (neutrophil-lymphocyte rate (NLR) (P = 0.425) and platelet-to-lymphocyte ratio (PLR) (P = 0.803)). We also observed strong correlations between TMB (Tumor mutation burden) ≥ 2 (P = 5.26E-04) and HRD ≥ 42 (P = 0.009) obtained by WES in the Gp-R group (Supplementary Figure S7).

Refining Breast Cancer Subtype Classification Using PAM50 and IRSN-23

Figure 3 shows unsupervised clustering results with PAM50 and IRSN-23 genes as ‘PAMIR’ (see Supplementary Figure S8 for clustering results with PAM50 only). The Basal subtype generally exhibits high IRSN-23 gene expression (\(\:\text{B}\text{a}\text{s}\text{a}{\text{l}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\)). The HER2, Luminal A (LumA), and Luminal B (LumB) subtypes appear to be divided into high (\(\:\text{H}\text{E}\text{R}{2}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\), \(\:\text{L}\text{u}\text{m}{\text{A}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\), and \(\:\text{L}\text{u}\text{m}{\text{B}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\)) and low (HER2, LumA, and LumB) groups based on IRSN-23 gene expression, respectively. The high-expression groups (\(\:\text{H}\text{E}\text{R}{2}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\), \(\:\text{L}\text{u}\text{m}{\text{A}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\), and \(\:\text{L}\text{u}\text{m}{\text{B}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\)) had higher pCR rates than the low-expression groups (HER2, LumA, and LumB). \(\:\text{L}\text{u}\text{m}{\text{A}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\) and \(\:\text{L}\text{u}\text{m}{\text{B}}^{\text{i}\text{m}\text{m}\text{u}\text{n}\text{e}}\) were divided into multiple clusters, but their HRD and TMB statuses and PGR, KRT14, and KRT17 expression levels were markedly different, suggesting the possibility of further division in the future.

Principal component analysis of the PAMIR gene expression was also performed. Figure 4A-B shows the localization of the newly classified subtypes by PAMIR, in addition to the PAM50 classification. Gene set enrichment analysis (GSEA) with genes having an absolute component loading value of \(\:\ge\:0.1\) for each principal component was performed. Compared to PAM50 and PAMIR, the first principal component changed from ‘cell cycle’, ‘oocyte meiosis’, and ‘chemical carcinogenesis’ to those plus immune-related pathways (‘viral protein interaction with cytokine and cytokine receptor‘, ‘cytokine-cytokine receptor interaction’, ‘chemokine signaling pathway’, and ‘Toll-like receptor signaling pathway’). The second principal component remained in the estrogen signaling and breast cancer pathways (Supplementary Table S2 and Supplementary Table S3). PAMIR clarifies the distance between subtypes, and the more effective treatment groups are shifted to the right in Fig. 4C. IRSN-23 was predicted to surround these groups (Fig. 4D and Supplementary Figure. S9). PAMIR-pCR predictions for all Affymetrix array data used in previous and current studies could classify all PAM50 subtypes into chemotherapy high- and low-sensitivity groups, as shown in Fig. 5 and Supplementary Figure S10.

Prognosis after NAC

Prognostic data after NAC was available for 769 patients from the OUH, GSE16446, and GSE25066 datasets. The median follow-up period for the survival analysis was 44 months, with a total of 174 events observed. Distant recurrence-free survival rate (DRFS) was analyzed based on the PAMIR breast cancer classification. In this analysis, LumA and LumB were grouped as LumA/B; similarly, LumA-immune and LumB-immune were grouped as LumA/B-immune, whereas HER2 and HER2-immune and basal and basal immune were considered individually. Comparatively, Basal-immune and LumA/B-immune cells showed significantly better DRFS than basal and luminal A/B (P = 0.023 and P = 0.035, respectively), as shown in Fig. 6.

Many chemotherapy sensitivity prediction models have been developed, but only a few have been prospectively validated. One of the rare prospectively examined and reported chemotherapy-sensitivity-prediction models is DLDA30, which Tabchy Adel et al. reported could predict sensitivity to weekly paclitaxel and fluorouracil, doxorubicin, and cyclophosphamide (T/FAC) but had low prediction accuracy for FAC. They noted that the prediction model was influenced by treatment (12). Nearly a decade ago, we reported the high accuracy of the IRSN-23 in predicting sensitivity to chemotherapy. Since then, numerous chemotherapy-sensitivity-prediction models have emerged. In 2018, the accuracy of IRSN-23 was prospectively validated by investigators from different institutions in a phase 3 TEX trial that examined the efficacy of epirubicin and paclitaxel with or without capecitabine in recurrent breast cancer. Foukakis et al. compared IRSN-23 with other diagnostic models, such as ESR1, PIK3CA, and TP53, and confirmed a linear correlation between IRSN-23 IS and the four-month shrinkage rate of the tumor at the metastatic site, especially in ER-positive, luminal-type breast cancer, demonstrating that it is one of the most accurate predictive models (21). In this prospective validation of the IRSN-23, we applied it to our data and public datasets. Moreover, the accuracy of IRSN-23 was evaluated in patients receiving anti-HER2 therapy, which was not applicable in a previous study. In recent years, ER + HER2- breast cancers have been determined by OncotypeDx RS as an indication for chemotherapy based on the risk of recurrence (22); therefore, collaboration with OncotypeDx was also evaluated in terms of its clinical utility value.

IRSN-23 prospectively predicted pCR in patients who received NAC without anti-HER2 therapy under the same conditions in which the model was constructed. IRSN-23 Gp-R was significantly more sensitive to chemotherapy in all cohorts that met the criteria used in the previous and current studies (Fig. 1). In a previous study (16), pooled analysis of 901 cases in a public database for all subtypes without anti-HER2 therapy showed a pCR rate of 40% in the Gp-R group vs. 11% in the Gp-NR group (P = 4.98E-23). In the pooled analysis of 1,103 patients with all subtypes not receiving anti-HER2 therapy, the pCR rate for the Gp-R group was 40 vs. 12% for the Gp-NR group (P = 2.02E-26), which was also consistent with the results of the current validation study. IRSN-23 has a predictive ability similar to other microarray platforms, which are inherently difficult to adapt. In conjunction with this study, we tested the transfer of IRSN-23 from microarray to RNA-Seq. IRSN-23 exhibited a highly significant correlation coefficient (Pearson correlation coefficient r = 0.98) in microarray and RNA-Seq analyses of the corresponding samples. Only 1 of the 43 cases showed a discrepancy in IRSN-23 diagnosis (Supplementary Figure S11). This high reproducibility may indicate the measurability of the IRSN-23.

Unexpectedly, the IRSN-23 signature did not demonstrate a distinctive ability to predict NAC susceptibility to concurrent anti-HER2 therapy, although a barely significant difference was observed in the pooled analysis. This result was consistently confirmed in all cohorts included in the study, suggesting that despite numerous investigations into predictors of response to anti-HER2 therapy, to the best of our knowledge, no factor has demonstrated a reproducible predictor of response to chemotherapy, except for HER2 expression and its amplification (23). This may indicate that the tumor microenvironment undergoes significant alterations before and after anti-HER2 therapy (24). However, our study suggests that the tumor microenvironment may be more sensitive to anti-HER2 therapy than NAC. Notably, reanalysis of the Gp-R group in the 2-week post-treatment sample showed that patients who were determined to have Gp-R after anti-HER2 therapy were more likely to achieve pCR, even if they had Gp-NR at baseline (Supplementary Figure S4F). IRSN-23 primarily consists of immune-only genes, particularly those associated with lymphocyte infiltration, suggesting a significant immune involvement. This finding may be related to antibody-dependent cell-mediated cytotoxicity, a known therapeutic effect of HER2-targeted therapy (25). Therefore, while IRSN-23 may not be a reliable predictor of response to neoadjuvant chemotherapy in combination with anti-HER2 therapy at baseline, its utility may be enhanced when used to monitor changes in the tumor microenvironment during the early phases of treatment. Further research is needed to validate this approach and determine the optimal timing for IRSN-23 assessment to improve its predictive accuracy and clinical applicability in HER2-positive breast cancer patients.

Analysis of the IRSN-23 gene signature and clinicopathological factors revealed a strong correlation between the Gp-R group and more malignant factors, including HG3 and Ki67 ≥ 20, HRD ≥ 42, and TMB ≥ 2. Moreover, the Gp-R group correlated with the infiltration of CD8-positive T cells and regulatory T cells, suggesting the recognition of tumor antigens by the immune system. Our previous study did not assess the relationship between tumors, peripheral blood lymphocytes, and IRSN-23. The present study showed no correlation between tumor lymphocytes and peripheral blood lymphocytes. Although the NLR has been repeatedly reported as a predictor of eribulin efficacy (26), the local immune response may play a major role in operable breast cancer. With respect to the biological mechanisms underlying the association between IRSN-23 and chemotherapy responsiveness, Fig. 4C shows that the shift to the right is associated with higher pathologic complete response (pCR), and Fig. 4D shows higher IRSN-23 IS. Genes associated with this shift include CX3CR1, CXCL10, CXCL11, CXCL9, and IL6ST, which are related to Cytokine-cytokine receptor interaction and Cytokine-cytokine receptor interaction pathways (Table S2). These genes are considered representative functional components of IRSN-23.

A recent report by Edlund et al. revealed that most of the 20-gene chemotherapy sensitivity prediction models they developed consisted of immune-related genes, with some gene crossover with the IRSN-23 gene signature (27). They also explained that a high cutoff value is one of the reasons why genetic diagnosis has not been clinically applied. After confirming the reproducibility of the IRSN-23 gene signature, we re-evaluated our previous findings, which were evaluated using binary values of Gp-R (IRSN-23 IS > 0) and Gp-NR (IRSN-23 IS ≤ 0) as continuous values of IRSN-23 IS. Collaboration with Oncotype Dx RS in ER + HER2- breast cancer analysis revealed that IRSN-23 IS and Oncotype Dx RS were linearly distributed and mutually predictive of the response to NAC.

Pooled analyses performed in previous and present studies showed that IRSN-23 can predict pCR in all breast cancer subtypes, with significant differences. Furthermore, IRSN-23 may have a prognostic significance after NAC treatment. This indicates that IRSN-23 goes beyond the chemotherapy susceptibility predictive diagnosis of breast cancer and predicts the existence of subtypes that are more likely to respond to chemotherapy. Perou et al. proposed an intrinsic subtype classification (28), which categorizes breast cancer into five types based on biological characteristics: LumA, LumB, HER2, Basal, and normal breast-like. Breast cancer classification using the PAM50 is currently the basis for breast cancer (18). However, our proposed PAMIR as a new subgroup classification includes tumor factors as offensive factors and immune responses on the host side as defensive factors. This approach differs from previous classifications that focused solely on the potential nature of the tumor. In 2022, Wolf et al. also suggested that breast cancers may form new subgroups through immune responses in their analysis of the I-SPY2 neoadjuvant trial (NCT01042379) (29). As a sophisticated factor in immune responses, IRSN-23 supports our hypothesis that host immunity is a factor in the new subtype classification, as shown in Fig. 7.

This study has several limitations. Firstly, the performance of IRSN-23 varied across different datasets and subtypes, which may be influenced by the specific characteristics of each dataset, such as patient demographics and treatment protocols. Secondly, some cohorts, like GSE28844, had small sample sizes (n = 32), reducing the findings' statistical power and reliability. This highlights the need for validation in larger and more diverse cohorts. Lastly, our findings are based on the specific datasets used in this study, limiting our results' generalizability. Future studies will aim to include larger and more diverse clinical trial datasets to validate further and improve the accuracy and reliability of IRSN-23. While the current study demonstrates the potential prognostic value of the PAMIR classification, comparing its performance with established prognostic markers or signatures is essential. Future studies will focus on this comparison to further validate and enhance the clinical relevance of the PAMIR.

In conclusion, our study prospectively validated the accuracy of IRSN-23, a diagnostic model based solely on immune-related genes, for predicting chemotherapy sensitivity. We found that IRSN-23 was a reliable predictor of response to chemotherapy. Furthermore, we demonstrated that IRSN-23 could be a factor in breast cancer subtype classification as a host defense mechanism against tumor-derived attack factors. Notably, IRSN-23 can be measured on different platforms and is expected to maintain its predictive performance. The analysis codes for the IRSN-23, RI-Score, and PAMIR are publicly available. We anticipate that accumulating more data will contribute to the clinical application of these models in the future.

Funding

This research was partly supported by JSPS KAKENHI (Grant Numbers: 16K10456, 18H02870, and 20K08926).

Author Contribution

The authors made the following contributions. YS: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Writing - original draft; SS: Methodology, Software, Supervision, Validation, Visualization, Writing - Review & Editing; YN: Funding acquisition, Project administration, Visualization, Writing - Review & Editing; KH: Investigation; MS: Writing - Review & Editing; TT: Writing - Review & Editing; HM: Software, Methodology, Supervision, Visualization, Writing - Review & Editing; KS: Writing - Review & Editing, Supervision.

Acknowledgement

The authors sincerely acknowledge Dr. Shinzaburo Noguchi (S.N.), former professor at the Department of Breast and Endocrine Surgery, Graduate School of Medicine, Osaka University, for his contribution to the design of this study and interpretation of the significance of the results. It should be mentioned with gratitude that this study would not have been completed without Dr. S.N.’s inspiration and supervision during his term as professor.

Data Availability

The OUH expression data for 263 patients (training set (n = 58), validation set (n = 59), and this prospective dataset (n = 146)) are available as GSE230881 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE230881) from NCBI’s Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/).

De Lena M, Varini M, Zucali R, Rovini D, Viganotti G, Valagussa P, et al. Multimodal treatment for locally advanced breast cancer. Result of chemotherapy-radiotherapy vs. chemotherapy-surgery. Cancer Clin Trials. 1981;4:229–36.
Asaoka M, Gandhi S, Ishikawa T, Takabe K. Neoadjuvant chemotherapy for breast cancer: Past, present, and future. Breast Cancer: Basic and Clinical Research. Volume 14. London, England: SAGE Publications Sage UK; 2020. p. 1178223420980377.
Von Minckwitz G, Huang C-S, Mano MS, Loibl S, Mamounas EP, Untch M, et al. Trastuzumab emtansine for residual invasive HER2-positive breast cancer. New Engl J Med Mass Med Soc. 2019;380:617–28.
Schmid P, Cortes J, Pusztai L, McArthur H, Kümmel S, Bergh J, et al. Pembrolizumab for early triple-negative breast cancer. New Engl J Med Mass Med Soc. 2020;382:810–21.
Nekljudova V, Loibl S, von Minckwitz G, Schneeweiss A, Glück S, Crane R, et al. Trial-level prediction of long-term outcome based on pathologic complete response (pCR) after neoadjuvant chemotherapy for early-stage breast cancer (EBC). Contemp Clin Trials Elsevier. 2018;71:194–8.
Schettini F, Pascual T, Conte B, Chic N, Brasó-Maristany F, Galván P, et al. HER2-enriched subtype and pathological complete response in HER2-positive breast cancer: A systematic review and meta-analysis. Cancer Treat Reviews Elsevier. 2020;84:101965.
Huang M, O’Shaughnessy J, Zhao J, Haiderali A, Cortés J, Ramsey SD, et al. Association of pathologic complete response with long-term survival outcomes in triple-negative breast cancer. Cancer Res AACR. 2020;80:5427–34.
Chang JC, Wooten EC, Tsimelzon A, Hilsenbeck SG, Gutierrez MC, Elledge R, et al. Gene expression profiling for the prediction of therapeutic response to docetaxel in patients with breast cancer. Lancet Elsevier. 2003;362:362–9.
Ayers M, Symmans W, Stec J, Damokosh A, Clark E, Hess K, et al. Gene expression profiles predict complete pathologic response to neoadjuvant paclitaxel and fluorouracil, doxorubicin, and cyclophosphamide chemotherapy in breast cancer. J Clin Oncol Am Soc Clin Oncol. 2004;22:2284–93.
Hatzis C, Pusztai L, Valero V, Booser DJ, Esserman L, Lluch A, et al. A genomic predictor of response and survival following taxane-anthracycline chemotherapy for invasive breast cancer. Jama Am Med Association. 2011;305:1873–81.
Farmer P, Bonnefoi H, Anderle P, Cameron D, Wirapati P, Becette V, et al. A stroma-related gene signature predicts resistance to neoadjuvant chemotherapy in breast cancer. Nature Medicine. Volume 15. New York: Nature Publishing Group US; 2009. pp. 68–74.
Tabchy A, Valero V, Vidaurre T, Lluch A, Gomez H, Martin M, et al. Evaluation of a 30-gene paclitaxel, fluorouracil, doxorubicin, and cyclophosphamide chemotherapy response predictor in a multicenter randomized trial in breast Cancer30-gene t/FAC response predictor in breast cancer. Clin Cancer Res AACR. 2010;16:5351–61.
Lee JK, Coutant C, Kim Y-C, Qi Y, Theodorescu D, Symmans WF, et al. Prospective comparison of clinical and genomic multivariate predictors of response to neoadjuvant chemotherapy in breast cancer. Clin Cancer Res AACR. 2010;16:711–8.
Rodriguez AA, Makris A, Wu M, Rimawi M, Froehlich A, Dave B et al. DNA repair signature is associated with anthracycline response in triple negative breast cancer patients. Breast Cancer Research and Treatment. Springer; 2010;123:189 – 96.
Naoi Y, Kishi K, Tanei T, Tsunashima R, Tominaga N, Baba Y, et al. Prediction of pathologic complete response to sequential paclitaxel and 5-fluorouracil/epirubicin/cyclophosphamide therapy using a 70-gene classifier for breast cancers. Cancer Wiley Online Libr. 2011;117:3682–90.
Sota Y, Naoi Y, Tsunashima R, Kagara N, Shimazu K, Maruyama N, et al. Construction of novel immune-related signature for prediction of pathological complete response to neoadjuvant chemotherapy in human breast cancer. Annals of Oncology. Volume 25. Oxford University Press; 2014. pp. 100–6.
Kim SJ, Sota Y, Naoi Y, Honma K, Kagara N, Miyake T, et al. Determining homologous recombination deficiency scores with whole exome sequencing and their association with responses to neoadjuvant chemotherapy in breast cancer. Translational Oncol Elsevier. 2021;14:100986.
Parker JS, Mullins M, Cheang MC, Leung S, Voduc D, Vickery T, et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol Am Soc Clin Oncol. 2009;27:1160.
Wolff AC, Hammond MEH, Allison KH, Harvey BE, Mangu PB, Bartlett JM, et al. Human epidermal growth factor receptor 2 testing in breast cancer: American society of clinical oncology/college of american pathologists clinical practice guideline focused update. Volume 142. Archives of Pathology & Laboratory Medicine. the College of American Pathologists; 2018. pp. 1364–82.
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 Oncol Elsevier. 2015;26:259–71.
Foukakis T, Lövrot J, Matikas A, Zerdes I, Lorent J, Tobin N, et al. Immune gene expression and response to chemotherapy in advanced breast cancer. Br J Cancer Nat Publishing Group. 2018;118:480–8.
Sparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF, et al. Adjuvant chemotherapy guided by a 21-gene expression assay in breast cancer. New Engl J Med Mass Med Soc. 2018;379:111–21.
Nami B, Maadi H, Wang Z. Mechanisms underlying the action and synergism of trastuzumab and pertuzumab in targeting HER2-positive breast cancer. Cancers. Volume 10. MDPI; 2018. p. 342.
Esteva FJ, Yu D, Hung M-C, Hortobagyi GN. Molecular predictors of response to trastuzumab and lapatinib in breast cancer. Nat Reviews Clin Oncol Nat Publishing Group UK Lond. 2010;7:98–107.
Clynes RA, Towers TL, Presta LG, Ravetch JV. Inhibitory fc receptors modulate in vivo cytoxicity against tumor targets. Nat Med Nat Publishing Group. 2000;6:443–6.
Miyoshi Y, Yoshimura Y, Saito K, Muramoto K, Sugawara M, Alexis K, et al. High absolute lymphocyte counts are associated with longer overall survival in patients with metastatic breast cancer treated with eribulin—but not with treatment of physician’s choice—in the EMBRACE study. Breast Cancer Springer. 2020;27:706–15.
Edlund K, Madjar K, Lebrecht A, Aktas B, Pilch H, Hoffmann G, Hengstler. J. G.Gene expression–based prediction of neoadjuvant chemotherapy response in early breast cancer: results of the prospective multicenter EXPRESSION trial. Clin Cancer Res. 2021;278:2148–58.
Perou CM, Sørlie T, Eisen MB, Van De Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits of human breast tumours. Nat Nat Publishing Group UK Lond. 2000;406:747–52.
Wolf DM, Yau C, Wulfkuhle J, Brown-Swigart L, Gallagher RI, Lee PRE, et al. Redefining breast cancer subtypes to guide treatment prioritization and maximize response: Predictive biomarkers across 10 cancer therapies. Cancer Cell Elsevier. 2022;40:609–23.

Table 1 is available in the Supplementary Files section.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

IRSN-23 Gene Diagnosis Predicts Response to Neoadjuvant Chemotherapy and Refines Breast Cancer Subtype Classification (Prospective Analyses)

Status:

Version 1

Abstract

Figures

Introduction

Methods

Patients and Samples

Prospective OUH Dataset

Prospective Public Dataset

Experiment Design

Experiment 1: Validation of Reproducibility

Experiment 2: Predictors of Chemotherapy Sensitivity and Clinical Utility

Experiment 3: Characterization of IRSN-23 and Its Impact on Breast Cancer Subtype Classification

Measurement Methods for Each Factor

Immune-Related 23 Gene Signature for NAC Using DNA Microarray

TMB and HRD

OncotypeDx and Intrinsic Subtype

Immunohistological Examination

Statistical Analysis

Results

Reproducibility

Prospective Validation Datasets Without Anti-HER2 Therapy Using Affymetrix Human Genome U133 Arrays

Prospective External Validation Datasets Without Anti-HER2 Therapy Using Different Microarrays

Datasets of NAC with Anti-HER2 Therapy

Predictors of Chemotherapy Sensitivity and Clinical Utility

Characterization of IRSN-23 and Its Impact on Breast Cancer Subtype Classification

Refining Breast Cancer Subtype Classification Using PAM50 and IRSN-23

Prognosis after NAC

Discussion

Conclusions

Declarations

Funding

Author Contribution

Acknowledgement

Data Availability

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1