Tumor samples were chosen from a retrospective cohort of 270 women with primary breast cancer including five women with bilateral tumors. These samples were collected as part of an observational longitudinal study from two tertiary cancer care hospitals in Bangalore, India between 2008 and 2013, and these women were followed-up for up to 9 years, with a total loss to follow-up of less than 5% and a median follow-up duration of more than 72 months. The study was approved by the ethical committee of both institutions, and informed consent was obtained from all the patients to use their tissue and blood sample for research. Information on clinical variables like age, grade, tumor size, lymph node status, stage of the disease with ER, progesterone receptor (PgR), and HER2 was obtained from their clinical records. Treatment information was obtained from clinical records of patients during follow-up. Endocrine therapy was recorded as tamoxifen or aromatase inhibitor, and chemotherapy regimens were noted for intake of anthracyclines or taxanes. Information on trastuzumab was recorded in HER2 positive patients whenever received. Formalin fixed paraffin embedded (FFPE) blocks from tumor tissue having more than 50% of the area of representative tumor were selected for the study.

Immunohistochemistry for AR was done on each of the tumor sections as per standard protocol using the Ventana Benchmark^XT staining system (Ventana Medical Systems, Tucson, AZ, USA). Briefly, 5 μm thick sections were fixed in hot air oven at 60°C for 60 min and loaded on to an IHC staining machine. De-paraffinization was performed using EZ Prep solution (Proprietary-Ventana reagent), and antigen retrieval was done using Cell Conditioning solution 1 (CC1) for 60 min. Primary antibody for AR (Clone AR 441, DAKO, dilution at 1:75) was added manually and incubated for 32 min at room temperature. Optiview DAB Detection Kit (Ventana Medical Systems) was used to visualize the signal, using DAB (3–3′diaminobenzidine) as the chromogen. Further, the sections were automatically counterstained with hematoxylin II (Ventana Medical Systems) for 12 min. The slides were removed from the autostainer, washed in de-ionized water, dehydrated in graded ethanol, cleared in xylene, and examined by microscopy. Appropriate positive and negative controls were run for each batch. Two pathologists scored the staining for AR protein independently and arrived at a final score. Nuclear staining in more >1% of tumor cells was considered as positive.

Total RNA was extracted using the Tri Reagent protocol according to manufacturer’s instructions (Sigma Aldrich # T9424) from two 20 µm sections from the selected tumor block. Briefly, tumor block was deparaffinized using heat, and then subjected to overnight digestion using proteinase K (Qiagen #19133). Quantitation of the RNA was done using the Qubit RNA BR (Broad-Range) Assay Kit (Invitrogen # Q10210) on a Qubit 2.0 Fluorometer (Invitrogen #Q32866). Then 500 ng of total RNA was reverse transcribed to cDNA using high capacity cDNA conversion kit from Thermofisher scientific (Cat # 4322171) as per manufacturer’s instruction.

Primers were designed for AR and ESR1 genes using primer 3 plus software and further validated on ensemble genome browser, NCBI blast and UCSC genome browser. The primers were synthesized by Juniper Life Sciences, Bangalore, India. The details of the primer sequences are given in the Supplementary Table 1 . For quantitative real time PCR (qPCR), 5 ng of cDNA template was used per reaction and performed in duplicate using SYBR^® Green on the LightCycler^® 480 II (Roche Diagnostics). Pre-incubation and initial denaturation of the template cDNA were performed at 95°C for 10 min, followed by amplification for 45 cycles at 95°C for 15 s and 60°C for 1 min. Cycles of threshold (Ct) values for the test genes were normalized to the mean Ct values of the three reference genes—ACTB, RPLP0, and PUM1 for each tumor sample which was normalized for varying abundance of transcripts. Relative normalized expression of test genes was calculated by ΔCT method. The methods used for nucleic acid extraction, quantitative PCR (qPCR), and selection of housekeeping genes (HKGs) and the quality control criteria for inclusion of samples in the analysis have been described in detail in our previous publication (21).

The estimation of total testosterone in serum samples collected prior to surgery or following surgery of 169 breast cancer patients was done by a chemiluminescence based immunoassay method using the Abbott Architect ci8200 (Integrated) & i2000 (Immunoassay) instrument. In brief, the serum sample with a minimum volume of 300 µl was loaded onto the instrument. The sample was then transferred into multiple compartments where it is mixed, incubated, and washed. In the subsequent steps, the conjugate, pre-trigger and trigger solutions were added. The chemiluminescence emission was measured to determine the quantity of total testosterone in the serum sample. The result was calculated using a four parametric logistic curve fit data reduction method to generate a calibration curve.

Descriptive analysis was done to evaluate the cohort characteristics and distribution of the high and low AR/ER ratio groups. Difference in the clinical variables between high and low ratio groups was tested by independent Student’s t-test or Mann–Whitney U test for continuous variables, and chi-square test was done for categorical variables. Concordance between the AR transcript and protein was estimated by receiver operating characteristic (ROC) curve analysis. Kaplan–Meier survival curves and log rank tests were used to compare the disease-free and breast cancer specific survival between the high and low AR/ER ratio groups. Disease free survival (DFS) and breast cancer specific survival (BCSS) were calculated as the time from the date of first diagnosis to the time when a local or distant recurrence occurred and death due to disease, respectively. Patients with no event or had death due to non-breast cancer related causes were right censored. The prognostic importance of high AR/ER ratio in comparison to other clinicopathological characteristics was validated by both univariate and multivariate cox-proportional hazard analyses. All tests were two tailed, and P-value <0.05 was considered statistically significant. All statistical analyses were done on statistical software XLSTAT version 2019.4.2 and SPSS software version 20 (Chicago, IL).

A total of 270 patients were included in the study with a median age at first diagnosis of 56.2 years. Nearly 60% of the tumors were associated with spread to the regional lymph-nodes and half of women were at clinical stage 2 and a third stage 3. Less than 10% of the tumors were grade 1 with approximately half being grade 2; 68% were estrogen receptor positive, and 19% were HER2 positive. Clinical variables are shown in Table 1 .

Most of the patients (>95%) were treated with stage appropriate endocrine and chemotherapy as standard of care except those with stage IV disease who died due to disease before completion of therapy. Of ER positive patients 93% (50/54) received endocrine therapy and received stage appropriate chemotherapy as well. Similarly, in ER negative patients, more than 90% of patients received stage appropriate chemotherapy and one had defaulted. Only 15% (3/20) of the HER2 positive patients received trastuzumab while 95% of them received anthracycline and taxane based regimens as intensive chemotherapy.

Among the 270 patients, 88 women were less than or equal to 50 years at first diagnosis, and the median age of this subset was 43.1 years. In this subset, 60% of the tumors were lymph node positive, and nearly half of the tumors belonged to stage II. Ninety-five percent of the tumors were equally distributed between grades II and III. Sixty percent of the tumors were ER positive. Eighty percent of these patients were pre-menopausal (70/88), and the remainder had been diagnosed with breast cancer on average within 3 years of menopause. No significant difference in any of the clinical characteristics was observed when this subgroup was compared to the entire cohort ( Table 1 ).

Immunohistochemistry for AR could be successfully evaluated in 189 of the 275 tumors. Eighty six tumors were not evaluated either due to insufficient tissue or tumor content. An additional 12 tumors were excluded due to poor tissue preservation, and hence the final evaluation included only 177 tumors from 173 women. There were 59/173 women (34%) less than 50 years, and 114/173 women (66%) were >50 years of age.

Overall, 66/177 (37%) tumors had nuclear staining for AR. There was no difference in the distribution of AR protein by age groups (34% (20/59) in ≤50 and 39% (46/118) in >50 years age group). Of the 177 tumors, 120 were ER positive by IHC; 53/120 of these ER positive tumors were AR positive as well, and this proportion did not differ between the age groups [42% (16/38) vs 45% (37/82) in ≤50 years vs >50 years respectively]. Overall, only 30% (53/177) of the tumors were dual positive for both ER and AR. Of AR positive tumors, 80% were ER positive in all samples, and similar results were seen in both age groups.

Tumors which were positive for AR protein expression had significantly high levels of AR transcripts than AR negative (p = 0.003). Tumors in the >50 years group had higher levels of AR transcripts when compared to ≤50 age group (p = 0.021). ROC analysis showed only moderate concordance (AUC of 0.63, p = 0.07) between the transcript and the protein across all tumors. No difference in this concordance was observed when stratified by age groups.

The AR and ESR1 transcript levels were evaluated by real time PCR on all the 275 breast tumor samples. A significant positive correlation was observed between the AR and ESR1 transcript levels (Pearson’s r = 0.43, p < 0.0001). Relative normalized units of AR and ESR1 transcripts were used to calculate the AR/ER transcript ratio which ranged from 0.65 to 5.53. In the tumors of women ≤50 years of age, the ratio ranged from 0.65 to 3.53 with a median value of 1.46 and third quartile value of 1.75. In tumors from women over 50 years of age, the ratio ranged from 0.74 to 5.53 and had a median of 1.31 and third quartile of 1.57. Though the level of AR transcript was higher in >50 years group, the AR/ER ratio was significantly higher in tumors ≤50 years than in tumors >50 years (p = 0.005).

We further divided the tumors from women ≤50 years of age into high and low ratio groups based on the third quartile cut-off of 1.75, and 22/88 had high AR/ER ratio in this subset. Comparison of clinical characters between the high and low ratio groups showed higher preponderance of ER negative tumors in the high ratio group (68%, p = 0.001), and no significant differences were observed in other features like stage, grade, lymph node status, and tumor size as shown in Table 2 .

We first examined the prognostic ability of ER and AR independently both at protein and transcript levels in women ≤50 years by Kaplan–Meier survival analysis. No significant difference in survival was seen for ER protein (ER positive vs negative, mean survival 76.63 vs 76.98 months, log rank test p = 0.65) and ER transcript at mean cut-off (high vs low, mean survival time 81.76 vs 74.68 months, log rank test p = 0.75). Similarly, no difference in survival was seen with AR protein (AR positive vs negative, 75 vs 81.98 months, log rank test p = 0.18) or its transcript levels at mean cut-off (AR high vs low, 78.4 vs 76.7 months, log rank test p = 0.55).

Next, we examined the clinical significance of a higher AR/ER ratio in the age group of patients ≤50 years group by Kaplan–Meier survival analysis. As seen in the Figures 1A, B , both DFS and BCCS were significantly lesser in the high ratio group in comparison to low ratio group (mean survival time 64.9 vs 83.4 months, log rank test p = 0.01 for DFS and 56.99 vs 89.65 months, log rank test p = 0.003 for BCSS). We did not observe this difference in the survival in the >50 years of age group, though trends were indicative of better survival for low ratio group (mean survival time 66.9 vs 81.13 months, log rank test p = 0.1 for DFS).

Further, based on the ER protein expression by IHC, we divided the tumors into ER positive and negative and evaluated the prognostic significance of the AR/ER ratio independently within each category. Fifty-four (61%) of the 88 tumors were ER positive and women with tumors with higher ratio had significantly poorer survival when compared to the low ratio (mean survival time 41.8 vs 82.7 months, log rank test p = 0.007, Figure 2A ). As observed in all patients ≤50 years age group, no difference in survival was seen with either ER transcript or AR transcript levels alone within the ER positive tumors. A similar analysis within the ER negative (by IHC) category did not show any difference in the survival between the AR/ER high and low ratio groups (mean survival time 68 vs 72.65 months, log rank test p = 0.68)

To investigate the prognostic significance of the AR/ER ratio in patients ≤50 years of age group, Cox proportional hazard analysis was performed with other known prognostic variables like tumor size, grade, and lymph node status. Univariate analysis showed ( Table 3 ) prognostic significance of the high ratio with a hazard ratio of 2.6 (95% CI-1.0–6.5, p = 0.04) and 2.1 in multivariate analysis though not statistically significant (95% CI-0.8–5.8, p = 0.15). Similar analysis within ER positive tumors alone reconfirmed the prognostic relevance of the high AR/ER ratio with a significant hazard ratio of 4.6 (95% CI-1.35–15.37, p = 0.01) in univariate and 3.78 (95% CI-0.87–16.43, p=0.07) in multivariate analyses.

To check if the results were recapitulated in other cohorts, we accessed the TCGA dataset (https://www.cancer.gov/tcga). This dataset had a total number of 1,082 breast cancer patients, of which 322 patients were ≤50 years. The AR/ER ratio of the transcripts of AR and ESR1 was calculated and ranged from 0.02 to 4.07. A third quartile cut-off of the ratio at 0.88 was used to divide the tumors into high and low ratio groups. Comparison of clinical characters between the high and low ratio groups showed significantly different distribution of the ER status (p = 0.03) between the two groups, and no significant differences were observed in other features as shown in Supplementary Table 2 . Kaplan–Meir survival analysis performed in the patients ≤50 years showed that the patients with high AR/ER ratio had a significantly poorer disease free survival than the low ratio tumors (mean survival time 74.8 vs 157.3 months, log rank test p = 0.003) similar to what we had seen in our cohort ( Supplementary Figure 1 ). The Cox proportional hazard analysis showed prognostic significance of the high AR/ER ratio with a hazard ratio of 2.8 (95% CI-1.4–5.8, p = 0.005) in the univariate analysis and a significant hazard ratio of 3.2 (95% CI-1.4–7.3, p = 0.006) in the multivariate analysis ( Supplementary Table 3 ).

Further, we performed similar analysis within ER positive tumors within TCGA; 218/322 were ER positive by IHC. Kaplan–Meir survival analysis in this subgroup showed patients with high AR/ER ratio had a significant poorer disease-free survival than the low ratio tumors (mean survival time 83 vs 163 months, log rank test p = 0.037), similar to our results ( Figure 2B ). Cox proportional hazard analysis showed prognostic significance of the high AR/ER ratio with a hazard ratio of 2.9 (95% CI-1.0–8.0, p = 0.046) in the univariate analysis and hazard ratio of 5.96 (95% CI-1.7–20.2, p = 0.004) in the multivariate analysis.

We also attempted to validate our results in the METABRIC dataset (22) (details in the Supplementary Data ). As seen in our cohort, tumors with high ratio of AR/ER showed poorer survival than the low ratio tumors in both DFS (mean survival time 107.2 vs 142 months, log rank test p = 0.001) and BCCS (mean survival time 101.5 vs 137.5 months, log rank test p = 0.001) in the ≤50 years age group ( Supplementary Figures 2A, B ). Comparison between clinical variables between high and low ratio groups is shown in Supplementary Table 4 . Cox proportional hazard analysis showed prognostic significance of the high AR/ER ratio with a hazard ratio of 1.8 (95% CI-1.25–2.52, p = 0.001) in the univariate analysis and hazard ratio of 1.2 (95% CI-0.82–1.78, p=0.33) in the multivariate analysis, though not statistically significant.

To examine the association between circulating testosterone and AR expression in breast tumors, we estimated the total serum testosterone levels in patients at first diagnosis by chemiluminescence method. Among the total 270 patients, adequate serum was available in 169 patients (63 women were ≤50 years and 106 women were >50 years) for estimation of total testosterone level. The testosterone level ranged from 0.13 to 2.43 ng/ml with the mean value of 0.25 ng/m. No significant difference was observed in the circulating total testosterone levels between women with AR positive versus negative tumors (p = 0.847) and between patients ≤50 years and >50 years (p = 0.42). Women ≤50 years, with tumors having high AR/ER ratio tumors had significant high levels of testosterone compared to women with tumors with a low AR/ER ratio (p = 0.02). In contrast, high levels of circulating testosterone were observed in the patients >50 years with tumors having a low AR/ER ratio (p = 0.002) as shown in Figures 3A, B .