Cell‐by‐cell quantification of the androgen receptor in benign and malignant prostate leads to a better understanding of changes linked to cancer initiation and progression

Abstract

The androgen receptor (AR) plays a crucial role in the development and homeostasis of the prostate and is a key therapeutic target in prostate cancer (PCa). The gold standard therapy for advanced PCa is androgen deprivation therapy (ADT), which targets androgen production and AR signaling. However, resistance to ADT develops via AR‐dependent and AR‐independent mechanisms. As reports on AR expression patterns in PCa have been conflicting, we performed cell‐by‐cell AR quantification by immunohistochemistry in the benign and malignant prostate to monitor changes with disease development, progression, and hormonal treatment. Prostates from radical prostatectomy (RP) cases, both hormone‐naïve and hormone‐treated, prostate tissues from patients on palliative ADT, and bone metastases were included. In the normal prostate, AR is expressed in >99% of luminal cells, 51% of basal cells, and 61% of fibroblasts. An increase in the percentage of AR negative (%AR−) cancer cells along with a gradual loss of fibroblastic AR were observed with increasing Gleason grade and hormonal treatment. This was accompanied by a parallel increase in staining intensity of AR positive (AR+) cells under ADT. Staining AR with N‐ and C‐terminal antibodies yielded similar results. The combination of %AR− cancer cells, %AR− fibroblasts, and AR intensity score led to the definition of an AR index, which was predictive of biochemical recurrence in the RP cohort and further stratified patients of intermediate risk. Lastly, androgen receptor variant 7 (ARV7)+ cells and AR− cells expressing neuroendocrine and stem markers were interspersed among a majority of AR+ cells in ADT cases. Altogether, the comprehensive quantification of AR expression in the prostate reveals concomitant changes in tumor cell subtypes and fibroblasts, emphasizing the significance of AR− cells with disease progression and palliative ADT.

Introduction

Prostate cancer (PCa) is the most common noncutaneous cancer in North American men and among the leading causes of cancer mortality. Androgen deprivation therapy (ADT) is administered upon recurrence following potentially curative treatments and in patients with advanced disease at diagnosis. Despite meaningful advances with the use of ADT and second‐generation androgen receptor inhibitors (ARIs), this signaling axis remains the main therapeutic target for the systemic control of disease progression. Unfortunately, these therapies inevitably fail, and patients die from metastatic castration‐resistant PCa (mCRPC). Several AR‐dependent mechanisms contribute to ARI resistance, such as mutations in the AR gene and genomic amplification, the expression of constitutively active splice variants (ARVs) not responsive to androgens (e.g. ARV7), and androgen‐independent activation of AR downstream of growth factor‐ and cytokine‐signaling pathways [1, 2, 3, 4, 5, 6, 7, 8].

Previous reports on AR protein expression in PCa have been conflicting. Most studies show nuclear localization in the normal and malignant prostate, with negligible cytoplasmic staining. The authors have reported increases [9, 10, 11] or decreases [12, 13, 14, 15] in expression in PCa compared with normal glands. While, in some instances, high AR expression correlates with unfavorable clinicopathological parameters and predicts disease progression or PCa‐specific death [14, 16, 17], the opposite [10, 18] or no prognostic significance [9, 11, 15] has also been demonstrated. Moreover, reports on percentage of AR positive/negative (%AR+/−) cells in PCa are contradictory, with various groups showing %AR− cells decrease [11, 19] or increase [12, 20] compared with normal luminal cells. The %AR− cells may be predictive [20, 21] or not [12, 22] of earlier biochemical recurrence (BCR) and disease aggressiveness [23]. These discrepancies may be explained by differences in cohort size, quantitative versus semi‐quantitative methods assessing AR expression, and epitopes recognized by AR antibodies.

Reports on AR expression in fibroblasts have shown that fibroblastic AR regulates the production of several growth factors fulfilling multiple functions in the prostate, including the maintenance of homeostasis by controlling epithelial cell proliferation and differentiation [24, 25, 26, 27]. The altered stromal microenvironment also allows for malignant cell growth and invasion [28, 29]. Indeed, loss of fibroblastic AR occurs concomitantly with glandular breakdown, the inability of cancer cells to differentiate or further dedifferentiation [30, 31, 32].

The present investigation aimed to perform a quantitative analysis of AR expression in all cell subtypes of the prostate in men with benign prostatic hyperplasia (BPH) or PCa and in bone metastases. It led to the definition of an AR index (using %AR− cancer cells, AR staining intensity, and %AR− fibroblasts) which was applied to study a radical prostatectomy (RP) cohort. Moreover, we compared the detection of full‐length AR and ARV7 on consecutive sections. Finally, we assessed the presence of AR− neuroendocrine and stem cells through specific markers. Such a comprehensive cell‐by‐cell approach with multiple markers is worth considering to better understand progression based on AR expression throughout disease states.

Materials and methods

Inclusion of patients

The main cohort comprised 332 consenting patients who underwent RP between 1993 and 2008 at the McGill University Health Center. Their clinicopathological characteristics are summarized in Table 1. BCR following RP was defined as two consecutive rises in blood prostate‐specific antigen (PSA) >0.2 ng/ml. Patients were excluded from survival analyses (n = 35) if they had received neoadjuvant hormone therapy (NHT), if no prostate PSA data were available at follow‐up, or if they had failed surgery (PSA detectable post‐surgery). The final RP cohort consisted of 297 cases.

Table 1.

Clinical characteristics of the RP cohort

Parameter	Total n
Age (years)*	332	61 (43–73)
Preoperative PSA †	326	8.66 ± 8.28
Gleason grade	332	n
Group 1 (GS 3 + 3)		72 (22%)
Group 2 (GS 3 + 4)		154 (46%)
Group 3 (GS 4 + 3)		75 (26%)
Group 4 (GS ≥ 8)		30 (9%)
T stage	332	n
T2		180 (54%)
T2+		40 (12%)
T3a		90 (27%)
T3b		22 (7%)
Positive surgical margin	332	94 (28%)
Extraprostatic extension	332	109 (33%)
Seminal vesicle invasion	332	22 (7%)
Follow‐up (months)*	297	95 (1–222)
Biochemical recurrence	297	81 (27%)
Distant metastases	321	18 (6%)

^*Median (minimum–maximum).

^† Mean (±standard deviation).

This study also included the following categories of cases: transurethral resection of the prostate (TURP) for symptomatic relief of urinary obstruction attributed to BPH (n = 10); cases having received 3 months of NHT prior to RP (n = 11); advanced TURP cases under palliative ADT (n = 26); and bone metastases (n = 13) from patients with advanced disease.

Tissue processing and immunohistochemistry

Formaldehyde‐fixed paraffin‐embedded blocks were retrieved from the Pathology Department at the MUHC. Tissue microarrays (TMAs) were built for primary tumors of de‐identified RP cases and bone metastases. Tumors were represented by 1‐mm duplicate cores. Benign cores were included for 62 hormone‐naïve RP cases. Details on the TMAs are provided in Supplementary materials and methods.

Immunohistochemistry (IHC) was performed using the following antibodies: AR N‐terminal (aa 299–315; Thermo Fisher Scientific, Waltham, MA, USA; clone 441; 1/250), AR C‐terminal (aa 900–920; Abcam, Cambridge, UK; ab227678; 1/250), ARV7 (Revmab, San Francisco, CA, USA, 31‐1109‐00; 1/500), SOX2 (SRY‐box 2; Cell Signaling Technologies, Danvers, MA, USA; 3579; 1/2,000), chromogranin A (CHGA; Dako from Agilent, Santa Clara, CA, USA; DAK‐A3; 1/300), and synaptophysin (SYP; Abcam; ab32127; 1/1,600). The IHC protocol is detailed in Supplementary materials and methods. Antibodies were first tested at several dilutions on consecutive sections. For double staining (DoubleStain IHC Kit: M&R on human tissue; Abcam), sections were stained with AR N‐terminal or CHGA (DAB) antibodies, scanned, and re‐stained with SYP or AR C‐terminal (AR/Red) antibodies.

Quantitative scoring

Staining was assessed at ×40 magnification using an inverted microscope (Olympus IX‐81; Olympus, Tokyo, Japan). Stained slides were scanned using an Aperio image scanner system (Leica Microsystems, Morrisville, NC, USA) for further visualization and image acquisition. Additional details are provided in Supplementary materials and methods. Benign foci from hormone‐naïve RP cases were noted as ‘far’ in cores with no tumor cells (>1 mm) or ‘close’ for benign areas in tumor cores (<1 mm) obtained from different blocks. Fibroblasts were counted as positive/negative. Nuclei from luminal, basal, and cancer cells were counted as positive or negative. Positive nuclei were scored by intensity (1 = weak, 2 = moderate, 3 = strong) to derive a score differing from H score (AR negative cells were not included) where i represents the staining intensity (1–3):

$$\sum _{i=1}^{3}i\times \left(\frac{\text{Number cells}\mathrm{at}\text{intensity}i}{\text{Total number of positive cells}}\right)=\text{Intensity score}.$$

Statistical analyses

AR expression was analyzed using Mann–Whitney U test or Kruskal–Wallis H test and Dunn–Bonferroni post hoc test. Survival analyses (Kaplan–Meier, Cox univariate and multivariable, and Harrel' C‐index) were used to determine clinical significance. Best cutoffs for dichotomizing AR expression were identified using Cutoff Finder [33]. Validation methods for cutoff selection and Cox analyses are described in Supplementary materials and methods. Markers were compared with clinicopathological parameters using Fisher's exact test and unpaired t‐test. The Cancer of the Prostate Risk Assessment Post‐Surgical score (CAPRA‐S) was calculated [34]. All statistical analyses were carried out in R.

Bioinformatic analyses

Gene expression microarrays were accessed through Gene Expression Omnibus for the Stanford (GSE3933), Cambridge (GSE70770), and MSKCC (GSE21032) PCa datasets. RNA‐sequencing results were authorized and accessed through dbGaP for SU2C (phs000915.v2.p2) and GDC for TCGA (phs000178.v11.p8). Details are presented in Supplementary materials and methods.

Results

Quantification of AR expression

Previous studies on AR expression by IHC have shown conflicting findings, which are summarized in supplementary material, Table S1. The authors using the AR clone 441 antibody tend to show increased %AR+ cells and AR staining intensity in PCa compared with benign prostate. Several studies showed different ranges for %AR+ luminal cells [11, 19, 35], others grouped luminal and basal cells [9, 36], and some found completely negative benign areas in PCa [37]. As such discrepancies may be due to differences in IHC protocols, AR staining was first optimized by antibody titration. Consecutive sections from tumor tissues of three PCa patients (one GS7 and two ADT) were stained at different dilutions with an AR N‐terminal antibody detecting full‐length protein and most variants. As shown for an ADT case in Figure 1A, staining intensity increased with higher antibody concentrations. Similar trends were observed for the three cases. We chose to use the lowest concentration at which the %AR− nuclei remained constant (1/250), while minimizing background and allowing for optimal AR detection at varying staining intensities for accurate quantification.

Figure 1.

AR N‐terminal staining optimization protocol and representative images of different cell types. (A) Stacked bar graph of the percentage of cancer cells at different staining intensities (0–3+) in the cytoplasm and nucleus for increasing concentrations of AR N‐terminal antibody. (B–H) Representative images of AR N‐terminal staining: (B) BPH (×20, insert at ×40), showing AR+ luminal cells and AR+ and AR− basal cells and fibroblasts; (C) cancer case (GS 3 + 4) from the RP cohort showing a majority of AR+ cancer cells with few AR− cancer cells, along with AR− fibroblasts (×20, insert at ×40); (D) cancer case (GS 4 + 3) from the RP cohort showing a higher proportion of AR− cancer cells and fibroblasts (×40); (E) cancer cells in an advanced ADT case with a high proportion of AR− cells interspersed among AR+ cells (×40); (F) bone metastasis showing AR+ and AR− cancer cells (×40); (G) AR− inflammatory cells in a benign area of a case from the RP cohort (×20, insert at ×40); and (H) AR+ and AR− endothelial cells in a blood vessel of a case from the RP cohort (×40, insert at ×80).

These conditions were next applied to sections of all abovementioned tissues. Results are summarized in supplementary material, Table S2, with representative images presented in Figure 1B–H.

AR negative cancer cells and fibroblasts increase with grade and hormonal therapies

The %AR+/− cells in the epithelium were first assessed in BPH and benign foci from prostate of the RP cohort. Approximately 50% of basal and >99% of luminal cells were AR+, with no difference between true BPH and benign foci from cancer (Figure 2A,B).

Figure 2.

Quantification of AR staining in the normal and malignant prostate. AR N‐terminal antibody staining quantified as (A and B) %AR− epithelial/cancer cells, (C) %AR− fibroblasts, and (D–G) AR intensity score (calculated as explained in the Materials and methods section). Benign areas from cancer cases are separated into far versus close to cancer foci. Cancer patients are referred to by cases from the RP cohort (hormone‐naïve/untreated) further separated by GS; hormone‐treated prior to RP and referred to as NHT and palliative ADT, and bone metastases at advanced stage. (A–C, E, and G) Box plots for %AR− (A) basal cells, (B) luminal/cancer cells, and (C) fibroblasts, as well as (G) AR intensity scores for luminal/cancer cells are shown for the different categories of patients. The number of fibroblasts did not change between the different categories of samples studied. All pairwise comparisons are given in these figures for overarching categories (i.e. benign versus hormone‐naïve versus hormone‐treated versus bone metastases). For p values, * represents significant increase compared to benign, while x represents significant increase compared with hormone‐naïve cancer. Significance levels are represented by */x for p < 0.05; **/xx for p < 0.01; ***/xxx for p < 0.001. Stacked bar graphs of the percentage of cells at different staining intensities (0–3+) in (D) basal and (F) luminal/cancer cells in the different categories of patients. (H) Box plots of AR gene expression z‐scores in transcriptomic datasets (TCGA, Cambridge, Stanford, MSKCC, and SU2C). Overexpression was defined as +1.96 standard deviations over the mean expression in benign samples from the same cohort (dashed line), as indicated in the Materials and methods section. For the SU2C dataset, gene expression was compared to benign samples from the TCGA dataset. Denoted above each graph is the percentage of cases overexpressing AR and the Fisher's exact test results comparing the proportion of cases overexpressing AR versus benign cases (*p < 0.05; ***p < 0.001).

The %AR− cancer cells in this cohort were higher compared with benign luminal cells (p = 2.00E−16), increasing gradually with Gleason score (GS) up to 6.4% (p ≤ 0.003 for 3 + 3/3 + 4 versus ≥8; Figure 2B). Interestingly, hormone‐treated cases showed further increase up to 8.1% (p = 8.1E−10 versus treatment‐naïve). Finally, the %AR− cancer cells in bone metastases resembled that of hormone‐treated rather than hormone‐naïve patients (p = 1.20E−03 versus hormone‐naïve; Figure 2B). Notably, 3 of 13 bone metastases consisted of >30% AR− cancer cells.

In the prostatic stroma, inflammatory cells were negative for AR (Figure 1G). Approximately 5% of endothelial cells appeared to be AR+ (Figure 1H), with heterogeneous staining in the same vessel, showing no clinical relevance (data not shown). However, there were significant changes in fibroblastic AR (Figure 2C). The %AR− fibroblasts increased from BPH to benign far from cancer (p = 5.16E−03). This was higher in cancer compared with benign (p = 1.40E−06), increasing gradually with GS up to 84% (p ≤ 2.94E−03 for 3 + 3/3 + 4 versus ≥8). Loss of fibroblastic AR was prominent upon hormonal treatment (91%, p = 9.70E−10 versus RP cohort). Interestingly, only 6 of 23 ADT cases had >5% AR+ fibroblasts.

Overall, the %AR− epithelial cells gradually increased from benign luminal to cancer cells, increasing with GS, and highest in primary tumors from hormone‐treated cases and bone metastases. AR+ fibroblasts are predominant in benign tissue, decreasing in cancer, and often lost upon hormonal treatment.

Cancer cell AR staining intensity increases upon hormonal treatment

We next analyzed nuclear AR staining by intensity scores, calculated from epithelial cells at 1+ to 3+, as described in the Materials and methods section. Expression patterns for basal cells showed a trend toward more cells with intense staining in cancer cases but not significant in intensity scores (Figure 2D,E). Luminal cells showed mainly 1+ staining, with more cells at intense staining in benign foci close to cancer, resulting in higher intensity scores compared to benign far (p = 0.006; Figure 2F,G).

Cancer cells in the RP cohort resembled benign luminal cells close to cancer in terms of staining patterns and intensity scores, irrespective of GS (Figure 2F,G). More cancer cells were stained at 3+ in hormone‐treated cases, resulting in higher intensity scores compared with hormone‐naïve cases (p = 4.6E−06). Bone metastases were distinct, with a high proportion of cells at 2+ and an intensity score higher than hormone‐naïve primary cases, but lower than treated cases (p ≤ 0.015).

These findings on nuclear AR protein levels were next compared with AR gene expression in five PCa transcriptomic datasets (Figure 2H). Whereas transcript levels were comparable to benign cases in most RP cases, they were significantly increased in a few cases of GS ≥ 8, TURP from advanced cases (p = 0.071), and in metastases, particularly in mCRPC metastases of the SU2C cohort. Taken together, there is an increased %AR− cancer cells, accompanied by an increased staining intensity of AR+ cancer cells and a gradual loss of fibroblastic AR during PCa progression and hormonal treatments. The elevated AR protein levels in cancer cells are in line with overexpressed AR transcripts in advanced disease.

Full‐length AR and AR variants are detected in cancer cells

ARVs appear to contribute to PCa progression, especially upon hormone treatment [7]. Several variants lack the C‐terminal domain [38]. We thus questioned whether AR C‐terminal antibodies detecting full‐length AR display expression patterns similar to N‐terminal antibodies recognizing full‐length AR and most ARVs, except AR45 [38]. Following titration experiments (Figure 3A), the 1/250 dilution of the C‐terminal antibody was chosen to stain consecutive cancer sections from a subset of cases. Representative images in Figure 3B–E show nuclear AR in normal and malignant cells and some positive fibroblasts. Cytoplasmic staining was seen but relatively low and constant across cases, therefore the focus was on nuclear AR (supplementary material, Table S2).

Figure 3.

AR staining with N‐ and C‐terminal and ARV7 antibodies. (A) Stacked bar graph of the percentage of cancer cells at different staining intensities (0–3+) in the cytoplasm and nucleus for increasing concentrations of AR C‐terminal antibody. (B–E) Representative images of AR C‐terminal staining (×40) in: (B) BPH; (C) cancer from case of the RP cohort; (D) palliative ADT; (E) bone metastasis. The color legend is the same for (B–E). (F–I) Correlation between AR N‐ and C‐terminal for: (F) %AR− fibroblasts; (G) %AR− basal cells; (H) %AR− luminal/cancer cells; (I) AR intensity score; (J) Box plots of ARV7 gene expression (fragments per kilobase of exon per million mapped fragments/FPKM) from the TCGA and SU2C PCa cohorts. (K) Images of two cancer foci from an ADT case showing staining with AR N‐terminal, C‐terminal, and ARV7 antibodies on consecutive slides (×20).

Analyses of concordance between N‐ and C‐terminal staining showed similar results for fibroblasts and basal cells for all samples tested (Figure 3F,G). The %AR− luminal and cancer cells were also similar for most cases (Figure 3H), except for one bone metastasis which contained a higher %AR− C‐terminal compared with N‐terminal (45% versus 31%). This suggests that ~15% of these cancer cells express high ARVs levels with lesser full‐length AR. Intensity scores were also concordant for most cases (Figure 3I). No noteworthy differences were observed between categories of cases.

To get insights into expression of ARVs, we opted to investigate the most well‐studied transcript ARV7. In transcriptomic datasets, we found that ARV7 is rarely and minimally detected in RP cases, higher in GS ≥ 8, and common in mCRPC metastases (Figure 3J). We next compared AR and ARV7 expression for three ADT cases with >10% AR− cells. An example of ARV7+ cancer cells is shown in a tumor area filled with AR+ cells, highly stained with C‐terminal and moderately with N‐terminal antibodies (Figure 3K). Another area of the same tissue showing morphologically different AR+ cancer cells was negative for ARV7. Thus, cancer cells expressing ARV7 may represent cell subsets. Altogether, these findings demonstrate similar patterns of AR+ cells detected with AR N‐ and C‐terminal antibodies, which are not suitable to study ARV7. Cancer foci are highly heterogeneous, with AR− cells being scattered among AR+ cells, including subsets expressing ARV7.

Clinical relevance of AR expression

To ascertain the clinical significance of AR expression in cancer, survival analyses were performed in the RP cohort. Optimal cutoffs were determined for dichotomization of variables and were further validated using resampling methods (supplementary material, Figure S1). Kaplan–Meier analyses showed that high %AR− cancer cells (>5.5%) and high %AR− fibroblasts (>87%) were each predictive of earlier BCR (p = 0.035; p = 0.00028; Figure 4A,B). The %AR− cancer cells and fibroblasts correlated with T stage/extraprostatic extension and GS, respectively (supplementary material, Table S3). High AR intensity score showed a trend for worse outcomes but was not significant (Figure 4C).

Figure 4.

AR is predictive of biochemical recurrence. (A–E) Kaplan–Meier curves with risk tables for BCR in the RP cohort. (A) %AR− cancer cells ≤5.5% versus >5.5%. (B) %AR− fibroblasts ≤87% versus >87%. (C) AR intensity score <175 versus ≥175, as described in the Materials and methods section. (D) Combination of markers: low %AR− cancer cells and low %AR− fibroblasts and low AR intensity score (dark blue); only one marker high (light blue); two markers high (pink); all markers high (red). These were combined into an AR index, where having at least two of these markers high (pink and red curves) represents a high AR index. (E) Combination of AR index and CAPRA‐S risk score as: (dark blue) low risk; (light blue) intermediate (int) risk and low AR index; (green) intermediate risk and high AR index; (yellow) high risk and low AR index; or (red) high risk and high AR index. (F) Harrel C‐index for (light green) individual measures of AR expression: %AR− cancer cells, AR intensity, %AR− fibroblasts; (dark green) AR index; (blue) CAPRA‐S risk score, based on combination of clinical parameters: preoperative PSA, age, Gleason grade, T stage, surgical margins; and (navy blue) AR index in combination with CAPRA‐S risk score or clinical parameters. (G) Stacked bar graph of AR expression patterns in all categories of cases. Expression patterns were described as high versus low %AR− cancer cells, %AR− fibroblasts, and AR intensity, using the optimal cutoffs described above for the RP cohort.

Interestingly, these three measures of AR expression were not correlated (supplementary material, Figure S2). Accordingly, they were combined to reflect changes in AR expression with progression and hormonal treatments. Cases showing at least two of these unfavorable features (high %AR− cancer cells, high %AR− fibroblasts, and high AR intensity score) had shorter time to BCR in Kaplan–Meier analyses (p < 0.0001; Figure 4D). Thus, we defined an ‘AR index’, considered ‘high’ for cases with at least two of these unfavorable features, and otherwise considered ‘low’. A high AR index correlated with high Gleason grade, T stage, and extraprostatic extension (Table 2). The AR index was significant in univariable Cox analyses and remained significant in multivariable analyses when combined with standard clinicopathological parameters (preoperative PSA, age, Gleason grade, T stage, and surgical margins) or CAPRA‐S risk groups as covariates (Table 3). Finally, we found that the AR index further stratifies patients in the intermediate and high CAPRA‐S risk groups, but not the low‐risk group (Figure 4E, Table 3 and supplementary material, Figure S3). The results of these Cox analyses were validated using a bootstrap method (Table 3). In addition, by comparing the C‐index for each measure of AR expression, we found that the CAPRA‐S score and standard clinical parameters performed better when combined with the AR index (Figure 4F).

Table 2.

Correlation of AR index with clinicopathological parameters

Parameter	Low AR index (n = 145)	High AR index (n = 28)	p
Age (years) mean*	60.5 (±4.95)	61.4 (±3.96)	0.3225
Preoperative PSA mean*	7.56 (±4.95)	9.12 (±4.90)	0.1302
Gleason grade			0.0023
Group 1 (GS 3 + 3)	32 (22%)	3 (11%)
Group 2 (GS 3 + 4)	70 (48%)	8 (29%)
Group 3 (GS 4 + 3)	35 (35%)	10 (36%)
Group 4 (GS ≥ 8)	7 (5%)	7 (53%)
T stage			0.0350
T2	98 (68%)	13 (46%)
T3a	40 (28%)	11 (39%)
T3b	6 (4%)	4 (14%)
Positive surgical margin	33 (23%)	11 (39%)	0.0949
Extraprostatic extension	45 (31%)	15 (54%)	0.0296
Seminal vesicle invasion	6 (4%)	4 (14%)	0.0580

Categorical variables analyzed by Fisher's exact test. Significant p values are shown in bold italics.

^*Continuous variables analyzed by unpaired t‐test.

Table 3.

Uni‐ and multivariable Cox analyses of AR expression in the radical prostatectomy cohort

	Univariable		Multivariable		Validation
Variable	HR (95% CI)	p	HR (95% CI)	p	Mean HR (95% CI)
Clinical parameters and AR index
AR index	4.24 (2.36–7.62)	1.41E−06	2.87 (1.49–5.48)	0.0015	3.04 (1.44–6.43)
Age*	1.02 (0.97–1.08)	0.346	1.00 (0.94–1.05)	0.938	1.00 (0.94–1.06)
Preoperative PSA*	1.06 (1.03–1.09)	3.23E−05	1.03 (0.98–1.08)	0.2267	1.03 (0.96–1.11)
Gleason
Group 1–2 (ref)	–	–	–	–	–
Group 3	3.84 (2.08–7.09)	1.67E−05	2.81 (1.45–5.46)	0.0022	3.01 (1.29–7.04)
Group 4	8.05 (3.66–17.70)	2.08E−07	3.16 (1.17–8.51)	0.0229	3.52 (0.79–15.63)
T stage
pT2 (ref)	–	–	–	–	–
pT2+	1.73 (0.64–4.66)	0.2795	1.17 (0.33–4.14)	0.8127	1.07 (0.08–14.62)
pT3a	2.52 (1.34–4.73)	0.0042	1.36 (0.63–2.89)	0.4241	1.38 (0.55–3.49)
pT3b	12.41 (5.10–30.21)	2.82E−08	2.88 (0.78–10.57)	0.1118	3.00 (0.52–17.50)
Surgical margin	2.08 (1.18–3.67)	0.0112	1.41 (0.64–3.10)	0.3906	1.47 (0.61–3.56)
CAPRA‐S risk score and AR index
AR index	4.24 (2.36–7.62)	1.41E−06	3.74 (2.05–6.82)	1.74E−05	3.90 (2.19–6.95)
CAPRA‐S
Low risk (ref)	–	–	–	–	–
Intermediate risk	3.01 (1.46–6.23)	0.0029	2.92 (1.41–6.04)	0.0038	3.08 (1.36–7.01)
High risk	8.03 (3.58–18.02)	4.28E−07	7.15 (3.14–16.27)	2.79E−06	7.69 (2.99–19.81)
CAPRA‐S stratified by AR index
Low risk (ref)	–	–			–
Intermediate risk + low index	2.26 (1.04–4.90)	0.0388			2.31 (1.02–5.22)
Intermediate risk + high index	9.97 (4.03–24.68)	6.55E−07			11.04 (3.89–31.34)
High risk + low index	4.72 (1.79–12.42)	0.0017			4.90 (1.57–15.26)
High risk + high index	29.06 (10.60–79.64)	5.79E−11			32.77 (13.35–80.44)

Cox analyses of AR index adjusted for standard clinical parameters or CAPRA‐S risk score. AR index was also used to further stratify CAPRA‐S risk groups. Validation results are presented as the mean HR and 95% CI of 300 multivariable analyses. Significant values are shown in bold.

HR, hazard ratio.

^*Age and PSA were analyzed as continuous variables.

To further understand biological relevance, we analyzed the AR expression patterns of the additional categories of cases included in this study (Figure 4G). All benign samples had a low AR index, while 82% of NHT and 96% of ADT cases had a high AR index. Notably, only 4% for untreated RP cases scored high for all three unfavorable parameters, compared with 58% of hormone‐treated cases.

Altogether, a high %AR− cancer cells, high %AR− fibroblasts, and high nuclear AR expression in AR+ cancer cells are associated with clinical parameters, predict recurrence, and can further stratify patients with intermediate CAPRA‐S scores.

Neuroendocrine and stem cells are scattered among AR expressing cancer cells

As AR− cancer cells are clinically significant and increases upon hormone treatment, we investigated whether they display neuroendocrine or stem cell features. Consecutive sections from ADT cases with >10% AR− cells were double‐stained for AR/CHGA, AR/SYP, or for SOX2 alone. Representative images are shown in Figure 5. Various expression patterns were observed, often in adjacent areas of the same tissue. In Figure 5A, a few CHGA+, SOX2+, and SYP+ cells were scattered in a vast area of predominantly AR+ cells. Heterogeneity was noticed within cells of this subtype. Another area with a higher %AR− cells concurrently displayed several CHGA+ and SOX2+ cells (Figure 5B) but no SYP+ cells in the same field (data not shown).

Figure 5.

Neuroendocrine and stem cells scattered among AR+ luminal‐like cancer cells. Images of tumor foci on consecutive slides of advanced ADT cases, either dually stained with AR N‐terminal (brown) + SYP (red) or AR C‐terminal (red) + CHGA (brown), and for SOX2 (brown). (A) Tumor area with high AR cell staining with a few neuroendocrine/stem cells (×10; inserts ×40) within the same field. AR N‐ and C‐terminal antibodies stained most cancer cells. (Top left) An AR N‐terminal negative cell (blue box, magnified in insert, ×40), co‐stained and positive for SYP (top right). (Bottom left) Two CHGA positive cells (arrow and one magnified in insert, ×40) and (bottom right) two SOX2 positive cells (arrow and one magnified in insert, ×40). Overall, these five AR− cells are each positive for only one of the neuroendocrine/stem markers tested. (B) Another tumor area of the same case showing more CHGA stained neuroendocrine cells and SOX2 stained stem cells among AR+ cells (×20). SYP was negative in this tumor foci (not shown). (C) Sections of a second ADT case were similarly stained (×20; ×inserts 40). Images of two adjacent areas are compared. Area 1 (top) contains AR− cells, and a high proportion of SYP+ and CHGA+ (neuroendocrine) and SOX2+ (stem) cells among AR+ cells. In contrast, the adjacent Area 2 (bottom) shows only AR+ cells and no SYP, CGHA, or SOX2 staining, despite the presence of a few AR− cells (arrows in left panel). (D) In the same case, another area showed layers of AR+ cells, separated by layers of highly stained CHGA+, SYP+, and SOX2+ cells (×20; inserts ×40).

In another ADT case, adjacent areas in the same piece of prostate tissue displayed strikingly different expression profiles regarding these markers (Figure 5C). Area 1 contains several SYP+, CHGA+, and SOX2+ cells interspersed between AR+ cells. In the adjacent field (Area 2), no cells are positive for the tested neuroendocrine or stem cell markers. In a third area of this tissue (Figure 5D), almost half AR+ cancer cells formed layers, intermingled with AR− layers of cells uniformly expressing neuroendocrine or stem cell markers. This implies that the three cell subtypes co‐exist within the same tumor foci. Collectively, these findings support that the increase in %AR− cells in hormone‐treated cases is associated with an increase of neuroendocrine and stem cells scattered among luminal‐like AR+ cells.

Discussion

IHC studies of AR in the last 30 years have led to contradictory results, some of which are reconciled by the present cell‐by‐cell quantification of the protein in benign and malignant prostates of patients at different stages of disease and under different hormonal environments. Our findings on nuclear AR levels revealed significant changes in expression during PCa initiation and progression, affected by hormonal treatment. To our knowledge, no such detailed analysis of AR in all cell types has previously been performed. A complete quantitative approach is necessary to understand the relevance of AR, given its expression in various cell types and the use of ADT/ARIs throughout disease progression. The importance of staining intensity and the proportion of AR+/− cancer cells would have been missed if results were analyzed solely by H‐score. Furthermore, quantification was only possible at high magnification since AR+ cells remain prominent even in advanced cases, with variable patterns of AR− cancer cells scattered among them. The low and constant cytoplasmic AR staining was not linked to progression, as previously reported [10]. This does not rule out the importance of cytoplasmic post‐translational AR modifications by phosphorylation or interactions with other proteins [39].

Optimization of staining protocols was of utmost importance to ensure reliability of expression levels. This minimized background staining, reaching a plateau on the %AR− cells and providing a range of scoring intensities. Indeed, the %AR+/− cells vary drastically in the literature. This may be partly attributed to experimental conditions. Three studies reported %AR+/− cells similar to ours [11, 23, 40], one of which found that %AR+ cells decrease with grade [23]. A decrease in AR+ cells with progression was reported, but overall %AR+ cells were low [12, 14, 20, 32]. Reports on AR in basal cells range from AR−, <1% AR+, to mainly AR+ [12, 23, 32, 41]. In this study, we detected low levels of AR in most luminal cells in BPH and near half basal cells and fibroblasts (illustrated in Figure 6). AR− cells in the basal layer are composed of few ‘true’ stem cells, quiescent basal/reserve cells, and few neuroendocrine cells [42, 43, 44, 45]. AR in basal cells is likely necessary to regenerate the epithelium through a rapid androgenic response leading to their maturation and differentiation. Fibroblastic AR is consistent with androgens regulating their production of growth factors fulfilling functions such as epithelial homeostasis [25]. The situation was different in benign glands of cancer cases, where luminal cells close to cancer exhibited higher AR staining intensity. It is possible that ‘normal’ cells undergo molecular changes not yet detectable phenotypically, as reported in spatially resolved copy number alterations [46]. This may also be due to factors released by tumor or stromal cells [28], which explains why benign cells farther from cancer are less affected. Although stromal AR is important for carcinogenesis [47], it is unclear how or why fibroblasts are progressively AR−. It was reported that AR expression in fibroblasts inhibits their proliferation, which may allow selection for AR− fibroblasts during progression [48].

Figure 6.

Schematic representation of cells differentially expressing AR in the normal and malignant prostate with changes observed upon progression and hormonal treatments. AR expression is denoted by the color of the nuclei: white = 0+, light grey = 1+, dark grey = 2+, black = 3+. Cell subtypes are denoted by the color of the cytoplasm. (Top) The benign prostatic epithelium is composed of luminal cells (blue), rare neuroendocrine (NE) cells (purple), and basal reserve cells (red), including rare AR− stem cells. Fibroblasts (green) are presented in the stroma. Luminal cells AR+, ~50% of basal cells are AR+, while neuroendocrine and stem cells are AR−. Both AR+ and AR− fibroblasts are present in the normal prostate. (Middle) Upon tumorigenesis, the proportion of AR− cells increases and a higher proportion of fibroblasts are AR−. AR+ cancer cells are prominent. (Bottom) With ADT, there is a loss of luminal‐like cells that are AR+ and depend on androgens for survival, explaining the observed decrease of blood PSA in patients. Meanwhile, other luminal‐like AR+ cells (dark blue) adapt to progressively become less sensitive to androgens and eventually androgen‐independent while maintaining AR expression, along with emergence of ARV expression. AR− neuroendocrine and stem‐like cells are selected for, while fibroblastic AR is gradually lost. These changes lead to tumor foci composed of varying proportions of cancer cells of the luminal, neuroendocrine, and stem subtypes, as illustrated in ADT cases in Figure 5.

In cancer cells of RP cases, we found no change in AR intensity, but a gradual increase in AR− cancer cells with GS. A decrease in AR+ fibroblasts was also observed with progression, in agreement with previous reports [30, 31, 32]. This may play a role in the decrease of AR+ cancer cells, resulting from the decrease in fibroblastic AR‐mediated secretion of paracrine factors controlling PCa cell growth, and leading to reduced luminal differentiation rather than de‐differentiation [28, 49], affecting the balance of AR+/AR− cancer cells. The lack of correlation between AR intensity, %AR− cancer cells, and %AR− fibroblasts suggests that these are independent, although all contribute to progression.

To date, studies on the clinical relevance of AR expression in PCa were often inconclusive and contradictory. The present findings on AR expression led to an AR index with significant predictive power, regrouping the staining intensity in AR+ cells, the prevalence of AR− cells, and fibroblastic AR loss. Kaplan–Meier, multivariate Cox analysis, and Harrel's C‐index demonstrated a strong association of the AR index with earlier BCR and allowed further stratification of CAPRA‐S intermediate‐risk cases. Further studies are required to verify if its predictive value may be improved by adding other markers, such as AR post‐translational modifications, AR interactors, or other factors such as Ki‐67, p53, PTEN, MYC, or ERG status [50].

Mechanisms of AR− cell selection and AR+ cell adaptation in cancer are promoted by ADT, where the proportion of AR− cells increases, AR+ fibroblasts are lost, and AR intensity increases. Figure 6 illustrates how ADT eliminates the fully differentiated luminal‐like AR+ cells requiring androgens for survival, leaving behind AR+ cells that progressively adapt and become androgen‐independent. This is observed clinically with PSA fluctuations under intermittent ADT until castration‐resistance [51]. These AR+ cells grow under the influence of non‐androgenic growth factors, resulting in activated pathways that may be androgen‐independent, although cells are not necessarily totally androgen‐insensitive or AR‐independent, in line with AR crosstalks in diverse signaling pathways [1, 52]. In patients, AR remains targetable in advanced disease, as seen by the temporary response to ARIs [8]. Accordingly, it would be of interest to assess whether the AR index predicts response to palliative ARIs.

Our observation of increased nuclear AR intensity under NHT compared with hormone‐naïve cases is in line with reports on higher AR and ARV7 [53, 54]. This was substantiated in palliative ADT cases, pointing to enhanced expression of AR and possibly its variants, and also demonstrated for AR in bone metastases. This is supported by our analysis of AR and ARV7 transcripts in most advanced PCa compared with benign and primary tumors of several datasets. Of note are reports on approximately 30% of CRPC cases showing AR amplification [55] and high AR expression in bone metastases being associated with poor prognosis [56]. Preliminary IHC results on ARV7 in ADT cases show its nuclear expression in subsets of cancer cells within tumor foci comprised of more AR+ cells (N‐ and C‐terminal), implying that differences in staining with the N‐ and C‐terminal antibodies used cannot account for ARV7 expression. Variant‐specific antibodies, RNA in situ hybridization, and spatial profiling would be most informative in this regard.

Our findings on elevated %AR− cancer cells after ADT and in metastases are of particular interest considering the noncurative ADT and ARI therapies. These cells display neuroendocrine and stem‐like features, according to CHGA, SYP, and SOX2 expression. Their distribution in tumor foci supports a nonclonal growth of AR− cells, as they are interspersed among AR+ cells. It is unclear if AR+/− cancer cells result from a lack of differentiation, trans‐differentiation, or dedifferentiation [57]. Neuroendocrine cell heterogeneity was noted, suggesting that multiple markers are necessary to study this subpopulation [58]. As cell patterns were shown to differ between adjacent foci, choosing tumor areas to build TMAs or dissecting cancer cells for further analysis may have substantial consequences. Nonetheless, our observation of different cell subtype patterns within a same patient and among patients is aligned with our recent report highlighting these cell subtypes in PCa transcriptomic datasets and the traceability and clinical relevance of PCa cell‐subtype genes in liquid biopsies of mCRPC patients [59].

In conclusion, comprehensive cell‐by‐cell quantification of AR expression in the prostate, PCa, and bone metastases revealed concomitant changes in epithelial cells and fibroblasts throughout cancer initiation, progression, and treatment pressures. We define for the first time an AR index which was associated with recurrence and further stratified intermediate‐risk patients. Altogether, our findings confirmed the heterogeneity of AR expression in line with the clinical heterogeneity of this disease, further highlighting difficulties encountered with currently available treatments.

Author contributions statement

SC designs the project and involved in all steps. SD, TB, ES, TA and FZZ optimized methods, carried out experiments and analyzed staining data. LH updated clinical data. FB and AA reviewed and confirmed the histopathology of sections and clinical data analyses, respectively. All authors were involved in the writing/revising the manuscript and approved the final version.

Ethics approval

This project was approved by the Research Ethics Board of McGill University Health Centre (BDM‐10‐115).

Supporting information

Supplementary materials and methods

Figure S1. Statistical validation of cutoff points for AR expression

Figure S2. Independence of AR expression features in the RP cohort

Figure S3. The AR index further stratifies intermediate and high CAPRA‐S risk groups

Table S1. Summary of reported findings on AR expression by IHC in the prostate and PCa

Table S2. Summary of the cell‐by‐cell quantification results for AR N‐terminal and C‐terminal nuclear staining intensity in different categories of patients

Table S3. Correlation of %AR− cells, %AR− fibroblasts, and AR intensity score in cancer cells with various clinicopathological parameters

Data availability statement

Data and analyses can be made available upon request to the corresponding author.