This was a retrospective cohort study. The study protocol was approved by the Ethics Committee on Human Research, Faculty of Medicine, Prince of Songkla University (REC. 64-037-4-2). Patients who were diagnosed with TNBC between 2011 and 2019 at Songklanakarind Hospital, a university hospital in Southern Thailand, were included. The inclusion criteria were patients with primary TNBC whose tissue samples before adjuvant chemotherapy and/or radiation therapy were available in our institute. TNBC was identified when ER, PR, and HER2 staining was negative in the tissues using IHC and fluorescence/dual in situ hybridization assays. Evaluation of immunoreactivity of ER, PR, and HER2 followed the guidelines of the American Society of Clinical Oncology and College of American Pathologists (ASCO/CAP) [12–14].

Patient data, including age, clinical characteristics, pathological features, treatment, and follow-up information for recurrence, were retrieved from the electronic medical records of the hospital. The first episode of locoregional recurrence or distant metastasis was defined as a recurrence. Locoregional recurrence was defined as tumor recurrence in the chest wall/breast (ipsilateral or contralateral) and four lymphatic drainage regions of the breast (axillary, supraclavicular, subclavian, and internal mammary). Information regarding death was obtained from the Cancer Registry Unit of the faculty, which was updated biannually from the national death registration data.

Formalin-fixed, paraffin-embedded (FFPE) tissue samples and hematoxylin and eosin-stained histological slides were retrieved from the archives of the Department of Pathology. A Quick Ray^® manual tissue microarray (Unitma, Seoul, Korea) was used for TMA construction. Histological slides were reviewed, and two foci of the representative areas with viable and abundant tumor cells were selected and marked for each case. Then, the corresponding point in the respective FFPE block was marked and cut out using a 2-mm core needle and embedded within the recipient paraffin block. The recipient blocks were then heated at 40°C in a hot-air oven for 15 min. TMA was performed only for excisional biopsied tissue, whereas whole tissue sections were used for core needle biopsy specimens.

IHC was performed using an automated immunostaining system (BONDMAX; Leica Biosystems, Melbourne, Australia). Briefly, to begin with, the samples were rinsed with xylene for deparaffinization. The slides were rehydrated using increasing concentrations of alcohol and washed with phosphate-buffered saline. Antigens were retrieved and incubated with a bond peroxidase-blocking reagent, followed by primary antibodies against AR (1:200 dilution, Cell Signaling), CD8 (1:500, Cell Signaling), FOXC1 (1:200, Abcam), and DCLK1 (1:100, Cell Signaling Technology). A Bond Polymer Refining Detection Kit (Leica) was used for detection, and 3,3′-diaminobenzidine was used for immunodetection, followed by Mayer’s hematoxylin for counterstaining. We used prostate tissue and tonsil as positive controls for AR and CD8 expression, respectively. TNBC tissues which were tested strongly positive were used as positive control for DCLK1 and FOXC1 expression. The positive control tissues were put simultaneously on all of the examined slides.

Immunostaining was evaluated independently by two authors (M.L and P.T.) who were blinded from clinical data as the percentage of positive cells, irrespective of the signal intensity. For any discordant interpretation, the two authors would study the immunostaining slides together and discuss. Agreement on the positive or negative results of each stain was obtained.

For AR, FOXC1, and DCLK1 expression, the percentage of positive tumor cells over the total number of tumor cells on the slides was estimated. For CD8, stromal CD8⁺ tumor-infiltrating lymphocytes (TILs) were detected in five randomly selected fields (×200 magnification) using digitalized images of full tissue sections. The percentage of stromal CD8⁺ TILs was obtained based on the recommendation of the International TILs Working Group 2014 [15]; that is, the percentage of CD8+TILs was semi-quantitatively estimated over the area of stromal tissue within the borders of the invasive tumor, excluding the TILs outside the tumor boundary, such as tertiary lymphoid aggregation and tumor area with artifacts or necrosis.

For the TMA slides, the average percentage of the two cores was used for further analysis. The cutoff value of positive versus negative protein expression, according to Zhao et al. [10] was used. Tumors positive for AR, DCLK1, and FOXC1 were defined as tumors with ≥ 10% positive cells, whereas CD8-positive tumors were defined as ≥ 20% CD8⁺ TILs.

The classification method proposed by Zhao et al. [10] was used (Figure 1). Tumor with AR-positive (+) irrespective of other protein positivity was defined as LAR subtype, AR-negative (−)/CD8+ as IM subtype, AR-/CD8-/FOXC1+ as BLIS subtype, and AR-/CD8-/FOXC-/DCLK1+ as MES subtype. Tumors that could not be classified into any of these subtypes were designated as unclassifiable (UC).

Statistical analysis was performed using R program version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria). Categorical variables were described by the number of observations and percentages. Continuous variables are represented as mean ± standard deviation (SD). The correlation between clinicopathological characteristics and the TNBC subtypes was calculated using the Fisher exact or Pearson’s chi-square test, as appropriate. Disease-free survival (DFS) was defined as the time interval from the date of diagnosis to the date of first recurrence (local or distant), the date of last known follow-up, or the date of death for those who had no recurrence (censored data). Overall survival (OS) was defined as the time interval from the date of diagnosis to the date of death or the date of the last follow-up if the patients were alive (censored data). OS and DFS were determined using Kaplan-Meier analysis. The Cox proportional hazard regression model was used to determine the prognostic significance of variables. For this analysis, the missing values of clinical stage, chemotherapy and radiotherapy were imputed and replaced by random sampling imputation. IHC-based TNBC subtype and significant variables from univariate analysis were selected for multivariate analysis. Statistical significance was set at a two-sided p-value < 0.05.

During the study period, 198 patients with primary TNBC for whom tissue blocks were available, from the Department of Pathology, Songklanagarind Hospital, between 2011 and 2019 were involved. Three patients had no follow-up information, and a total of 195 TNBC patients were included in the analysis. The mean age of the patients in the cohort was 52.3 years (range: 25–92 years). The patients were grouped into two categories, based on their ages, namely, <50 and ≥50 years, for further analysis. Pathologically, most tumors were invasive ductal carcinomas (invasive breast carcinoma of no special type) (93.3%) and histologic grade 3 (81.5%). At the time of diagnosis, approximately half of the patients were in stage II (47.4%), followed by stage III (27.3%) of the disease. Majority of patients had received mastectomy (61%). Surgery was not performed in 8 patients as they died shortly after biopsy or unfit for surgery. Seventeen patients did not receive chemotherapy due to various reasons including death before treatment (8), denied treatment (3), clinically unfit (3), and non-chemotherapy option for stage I disease (2 patients). Data on treatments were missing in some patients as they transferred to receive treatments at other hospitals.

Representative IHC for all proteins is shown in Figure 2. According to IHC-based subtypes, BLIS was the most common subtype (103 cases, 52.82%), followed by LAR (37 cases, 18.97%), IM (34 cases, 17.44%), and MES (1 case, 0.51%). Twenty patients (10.26%) could not be classified and were categorized as UC. The clinicopathological characteristics of the patients according to TNBC subtypes are presented in Table 1. No characteristics, except age, were significantly varied between the subtypes. The BLIS subtype was found majorly in younger patients (mean: 49.6 years) as opposed to the other subtypes (mean: 51–57.7 years). The only one patient with MES subtype was 30 years old with clinical stage III and histologic type of invasive ductal carcinoma, grade 2. The patient received chemo-radiation therapy.

The median follow-up time was 40.95 months (interquartile range (IQR): 23.48–89.22 months). By the end of the study (December 2021), 63 patients had experienced at least one episode of recurrence, and 66 patients died. The 5-year rate of DFS was 64.7% (95% CI: 0.576–0.726), and OS was 65.0% (95% CI: 0.581–0.728). The Kaplan–Meier estimates of DFS between different IHC-based subtypes did not show any variation (p = 0.28), but the OS was significantly different (p = 0.034) (Figure 3). MES subtype was not included in the Kaplan-Meier curves because there was only one case. Patients with the IM subtype had a better OS than those with other subtypes, while the survival curves of the patients with other subtypes overlapped. The patient with MES experienced no recurrence and remained alive until the end of the study.

Tables 2, 3 present univariate and multivariate Cox regression analyses of DFS and OS, respectively. Age, clinical stage, histologic type, lymphovascular invasion, and radiotherapy were significant factors in the univariate analysis of DFS. However, only age and clinical stage remained significant in the multivariate model.

In the Cox regression analysis for OS, clinical stage, chemotherapy, and IHC-based subtype were significant factors. Specifically, compared to IM, LAR (HR: 5.05, 95% CI: 1.46–17.4) and BLIS (HR: 4.72, 95% CI: 1.46–15.24) were significantly associated with poorer OS in univariate analysis. However, only BLIS remained significant in the multivariate model (HR: 3.29, 95% CI: 1.01–10.72).