RANK signaling increases after anti-HER2 therapy contributing to the emergence of resistance in HER2-positive breast cancer

RANK and RANKL expression was assessed in tumor samples from three different cohorts of patients with HER2-positive breast cancer.

RNA samples of the PAMELA trial from tumors at baseline (n = 151) were previously analyzed [22]. Here, the nCounter platform (NanoString Technologies, Seattle, WA, USA) analyzed RNA of 101 residual tumors from surgical samples of the PAMELA trial. A minimum of 100 ng of total RNA was used to measure the expression of 550 genes, including RANK and RANKL, and 5 housekeeping genes (ACTB, MRPL19, PSMC4, RPLP0, and SF3A1). Expression counts were then normalized using the the nSolver 4.0 software and custom scripts in R 3.4.3. For each sample, we calculated the PAM50 signature scores (basal-like, HER2-E, luminal A and B, normal-like) and the risk of recurrence score [23]. Intrinsic molecular subtypes were identified using the research-based PAM50 predictor as previously described [22, 24].

Immunohistochemistry (IHC) in TMAs was performed using anti-human mouse monoclonal RANK (N-1H8 Amgen) and human RANKL (M366 Amgen) as described in [9]. RANK or RANKL staining was scored on a scale of 0 to 3 for intensity (0 = no staining, 1 = weak, 2 = moderate, 3 = intense) and for the percentage of positively stained tumor cells (0–100) as previously reported [25]. The result of multiplying staining intensity by positive cell percentage is the H-score value, ranging from 0 to 300. TMA cores were scored for RANK and RANKL with the assistance of the breast cancer pathologists from the Bellvitge Hospital, if tumor cells represented > 15% of the total TMA core area. Patients were stratified according to RANK or RANKL H-scores as being protein-positive (H-score ≥ 1) or protein-negative (H-score = 0). Breast tumors from patient-derived xenografts were used as positive and negative controls. Experimental data from our laboratory in breast cancer cells and patients’ samples [26] confirmed that cells in which RANK protein expression is not detected by IHC/western blot may still respond to RANKL stimulation or denosumab inhibition [11, 26, 27]. This is probably due to the “fragility” of the RANK epitope and the limited sensitivity of the current tools to detect RANK protein expression. Thus, even with an H-score ≥ 1, we are likely underestimating samples with a functional RANK signaling pathway.

Statistical analyses were performed with the support of IDIBELL and Nottingham University Statistical Assessment Services. The ER/PR/HER2 status, grade, and tumor stage were known for each case included in the TMAs. Associations between IHC scores and clinicopathological parameters were evaluated using Pearson’s chi-squared test.

The cell lines BT474 parental (BT474) and BT474 with lapatinib resistance (BTLR) were described in [28]. SKBR3 parental (SKBR3) and SKBR3 lapatinib resistant (SKLR) lines were described in [29]. The cell line HCC1954 was obtained from ATCC (CRL-2338). BT474 cells were grown in DMEM + GlutaMAX (Gibco) supplemented with 2 mM l-glutamine (HyClone), 1× penicillin-streptomycin solution (P/S, Gibco), and 7.5% fetal bovine serum (FBS, Gibco). SKBR3 cells were grown in McCoy’s 5A + GlutaMAX supplemented with 2 mM l-glutamine, 1 mM sodium pyruvate (HyClone), 1× P/S, and 5% FBS. HCC1954 cells were grown in RPMI medium 1640 + GlutaMAX supplemented with 2 mM l-glutamine, 1× P/S, and 5% FBS. The cells were grown at 37 °C in 5% CO₂ humidified incubators. For RANKL treatments, cells were incubated in the presence of 100–300 ng/ml of RANKL. Cell lines were routinely tested for mycoplasma contamination.

To ectopically express RANK, the RANK gene (TNFRSF11A) was cloned into the lentiviral vector pSD-69 (kindly provided by S. Duss and M. Bentires-Alj) under the control of hPGK promoter. As a control (ctrl), we used an empty pSD-69 plasmid generated by removing the BamHI-SalI fragment containing CcdB and CmR genes. Knockdown of RANK endogenous expression was achieved by shRNA lentiviral delivery using pGIPZ vectors containing shRNAs against human RANK (RHS4531, Dharmacon), and shRNAs sequences #3 (TATCTTCTTCATTCCAGCT) and #4 (ATTCTTCCTTGAACTTCCC) were selected based on their ability to silence RANK expression. As a control, we used pGIPZ expressing a verified non-targeting sequence (RHS4346 Dharmacon). Lentiviruses were prepared in HEK293T cells transfected with psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259) by the calcium phosphate method. Virus-containing supernatants were centrifuged at 1500 rpm for 5 min and filtered with 0.45-μm filters (Millipore). The medium from 1-cm² production cells was used to infect 2-cm² recipient cells at roughly 33% confluence, adding fresh medium (1:1) and 8 μg/ml polybrene (Millipore). Approximately 90% infection efficiency was verified 3 days after transduction by detection of GFP expressed from pGIPZ plasmids. Transduced cells were selected with 1.5 μg/ml puromycin (Sigma), starting 3 days after infection, and subsequently maintained with 1 μg/ml puromycin in the growth media.

To determine cell proliferation, 1000–4000 cells per well in 100 μl were seeded in 96-well plates. After 24 h, 100 μl of medium with or without the indicated concentrations of lapatinib (0–16 μM) was added, and cells were incubated for 4 days. The relative number of viable cells each day was determined by adding 50 μl of diluted CCK-8 reagent according to the manufacturer’s protocol (Sigma).

Cells were seeded at approximately 33% confluence in 6-well plates. The following day, they were washed and incubated in a medium without FBS. The next day, the medium was changed to 1.8 ml medium with or without 1 μM lapatinib followed by a 2-h incubation. Subsequently, 0.2 ml of medium with or without 300 ng/ml of RANKL (RANKL-LZ Amgen) was added to the wells. Ten minutes later, the extracts for immunoblots were prepared with modified RIPA buffer (50 mM Tris pH 7.4, 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate) containing 1× PhosSTOP and complete protease inhibitor cocktail (Roche), and protein concentrations determined with DC protein assay reagents (BIO-RAD). Fifteen micrograms of protein were resolved by SDS-PAGE and blotted into Immobilon-P 0.45 μm membranes (Millipore). Antibodies against the following proteins were used for probing: RANK (R&D Systems AF683), p-HER2 (#2249), HER2 (#2165), p-EGFR (#3777), EGFR (#4267), p-ERK1/2 (#9101), ERK1/2 (#9102), p-AKT (#4051), AKT (#9272), p-p65 (#3033), p65 (#8242), p-IκB (#9246), IκB (#9242) (from Cell Signaling), β-actin (sc-47778), and tubulin (Abcam ab21058).

Upon transiently transfecting HEK293 cells with affinity-tagged versions of full-length RANK (RANK-V5 in pLenti6/V5-DEST, Invitrogen), full-length HER2 (FLAG-HER2 [30]), an amino (742-NTF) [30], or carboxy-terminal fragment of HER2 (611-CTF) [31], cells were washed twice with ice-cold PBS and proteins were extracted with 20 mM Tris-HCl pH 7.4, 137 mM NaCl, 2 mM EDTA, 10% glycerol, 1% NP-40 supplemented with 50 μg/ml leupeptin, 50 μg/ml aprotinin, 0.5 mM sodium orthovanadate, and 1 mM phenylmethylsulfonyl fluoride. Equal amounts of extracts were incubated for 3 h with immunoglobulin G (Abcam ab171870), FLAG (Sigma F3165), HA (Abcam ab9110), V5 (Thermo Scientific #R961-25), HER2 (32H2 in house antibody described in [32]), or trastuzumab (Hoffmann-La Roche) antibodies. Then, protein A agarose beads (Calbiochem IP02) were added for 2 h. Immunoprecipitates were washed thoroughly with lysis buffer and boiled in reducing SDS loading buffer to be analyzed by Western blot.

Cells were seeded at approximately 33% confluence in 6-well plates. The next day, the medium was changed to medium with or without 100 ng/ml RANKL followed by an additional 24 h incubation period. To analyze mRNA expression levels, total RNA was purified with Maxwell RSC simplyRNA Tissue kit (AS1340 Promega). For each sample, cDNA was retrotranscribed from 1 μg of RNA using 200 U SuperScript II plus random hexamer oligos following the manufacturer’s protocol (Invitrogen); cDNA from 20 ng RNA for each sample was analyzed by SYBR green real-time PCR (Applied Biosystems) with 10 μM primers using a LightCycler® 480 thermocycler (Roche). Analyses were performed in triplicates using the LightCycler® 480 software (Roche). Peptidylprolyl isomerase A, PP1A, was used as the reference gene. The primer sequences used in the analyses are as follows: PP1A (fw ATGGTCAACCCCACCGTT, rev TCTGCTGTCTTTGGGACCTTG), RANK (fw GCAGGTGGCTTTGCAGAT, rev 5’GCATTTAGAAACATGTACTTTCCTG), BIRC3 (fw GGTAACAGTGATGATGTCAAATG, rev TAACTGGCTTGAACTTGACG), ICAM1 (fw AACTGACACCTTTGTTAGCCACCTC, rev CCCAGTGAAATGCAAACAGGAC), TNFα (fw AAGCTGTAGCCCATGTTGT, rev TGAGGTAACAGGCCCTCTGAT), and IL8 (fw CTGCGCCAACACAGAAATTA, rev CATCTGGCAACCCTACAACA).

The expression of RANK and RANKL in HER2-positive breast cancer patients was analyzed by immunohistochemistry (IHC) in two independent sets of tissue microarrays (TMAs): a collection of HER2-positive tumor samples from treatment-naive patients (n = 197) and a cohort with tumors resistant to neoadjuvant trastuzumab-based chemotherapy (n = 43) from patients with residual invasive disease at surgery.

In the first collection, the integrity of the tissue allowed scoring of 67 and 72 patients for RANK and RANKL expression, respectively. Considering positive those with H-score ≥ 1 for the tumor cells, RANK expression was found in 14/67 (20.9%) cases and transmembrane RANKL staining in just 2/72 (2.8%) of the samples (Fig. 1a). In the anti-HER2-resistant tumor samples, we could score 22 patients for RANK and 21 for RANKL (Fig. 1a). In these, 9/22 (40.9%) were positive for RANK and 2/21 (9.5%) for transmembrane RANKL in tumor cells (Fig. 1a). Representative pictures of RANK and RANKL positive samples are shown in Fig. 1b, and H-scores for the whole tumor core from all samples (excluding those with integrity issues) and controls are presented in Fig. S1A and B. Pictures of the whole TMA core area in both collections are shown in Fig. S2.

Next, we evaluated the clinico-pathological factors associated with RANK expression in treatment-naive HER2-positive tumors (Fig. 1c). RANK expression was significantly associated with tumors from younger patients (less than 50 years old; p = 0.034) and tumors with a higher Ki67 proliferation index (p = 0.02). A trend of increased frequency of RANK expression was found in ER/PR negative tumors (p = 0.170 and p = 0.090, respectively), and higher histological grade (p = 0.138) (Fig. 1c). Similar patterns were observed in tumors resistant to anti-HER2 treatment (Fig. 1c). In both series, the limited number of samples prevented additional statistically significant associations, but general patterns coincided with those reported in previous studies of RANK/RANKL expression in human breast cancer samples [11, 12, 33]. Importantly, the frequency of RANK/RANKL-positive samples was higher in anti-HER2-resistant compared to treatment-naive HER2-positive tumors.

Our previous results suggested that RANK and RANKL expression may increase upon acquisition of anti-HER2 treatment resistance (Fig. 1a). To determine the possible changes in RANK and RANKL expression induced by dual HER2 blockade, gene expression profiling was performed in paired surgical tumor samples obtained before and following treatment with lapatinib and trastuzumab (and endocrine therapy if the tumor was hormone receptor-positive) from the PAMELA phase II clinical trial [22]. At baseline, the expression of RANK was significantly associated with the PAM50 intrinsic subtypes (Fig. S3A; p < 0.001); non-luminal subtypes (Basal-like and HER2-enriched) had the highest RANK expression. No significant differences in RANKL gene expression across PAM50 intrinsic subtypes were observed, although RANKL levels were slightly increased in the luminal A subtype (Fig. S3A), as previously reported [33]. Moreover, RANK gene expression was higher in hormone receptor-negative tumor samples (p < 0.001) while RANKL showed the opposite trend (Fig. 2a) confirming previous findings [12, 34]. ERBB2 gene expression at baseline had a weak positive correlation (r = 0.16) with RANK and the opposite trend (r = − 0.21) with RANKL expression (Fig. S3B and C). RANK gene expression increased (Fig. 2b, denoted by red lines in the left graph and a green line in the right graph) following dual treatment with lapatinib and trastuzumab (p < 0.001), while RANKL expression did not significantly change when analyzing residual disease samples at surgery (Supplementary Table 1). Specifically, the mean RANK expression in baseline samples was − 6.22 (standard deviation (SD) = 1.22) versus − 5.58 (SD = 1.14) in the paired surgical samples. In contrast, the mean RANKL expression in the baseline samples was − 7.36 (SD = 1.28) versus − 7.44 (SD = 1.33) in the surgical samples. When populations were studied separately according to hormone receptor expression, the same findings were observed, an increase in RANK mRNA expression in both hormone receptor positive and negative HER2+ tumors after HER2 inhibition (Fig. S4A and B).

As RANK expression increased after dual lapatinib/trastuzumab treatment in HER2-positive breast cancer patients, we decided to test whether in vitro short-term treatment with both anti-HER2 drugs, alone or in combination, would influence RANK expression in three different HER2-positive breast cancer cell lines. While SKBR3 and BT474 cells are sensitive to lapatinib and trastuzumab, HCC1954 cells are less sensitive to lapatinib and resistant to trastuzumab [35].

Lapatinib treatment, alone or in combination with trastuzumab, resulted in higher RANK mRNA expression in SKBR3 when compared with non-treated cells (Fig. 3a). Lapatinib or trastuzumab treatment, as well as their combination, also increased RANK expression levels in BT474 cells. In HCC1954 cells, RANK expression increased with lapatinib alone or in combination treatment after 12 h, whereas trastuzumab alone did not alter RANK expression levels. Also, we analyzed RANK expression in SKBR3 cells, either parental (sensitive to lapatinib and trastuzumab) or resistant to trastuzumab (SKTR), to lapatinib (SKLR), or to both (SKTLR and SKLTR; derived from SKTR and SKLR, respectively) [29]. RANK gene and protein expression levels were significantly higher in lapatinib-resistant SKLR and dual lapatinib/trastuzumab-resistant SKLTR cells when compared to SKBR3 parental cells (Fig. 3b, c). Increased RANK mRNA expression was also observed in BT474 cells with acquired lapatinib resistance (LR) when compared to lapatinib-sensitive parental cells according to public datasets, platform ID: GPL570 [36]; (Fig. 3d), an increase we verified by RT-qPCR (Fig. 3e).

To confirm that the elevated RANK expression levels were accompanied by increased activation of the RANK signaling pathway, the expression of RANK downstream gene targets BIRC3, ICAM1, TNFα, and IL8, indicative of NF-κB pathway activation [37, 38], was analyzed in sensitive and lapatinib-resistant (LR) cells treated with or without RANKL. Lapatinib-resistant SKLR cells showed higher gene expression levels of RANK, BIRC3, ICAM1, TNFα, and IL8 compared with control SKBR3 cells, and their levels were further increased after pathway stimulation with RANKL, except for IL8 (Fig. 3e). In lapatinib-resistant BTLR, increased expression of RANK and its downstream targets ICAM1 and IL8 was detected, and their levels increased further upon RANKL stimulation compared to sensitive BT474 cells. In these cells, BIRC3 expression did not change whereas TNFα was barely expressed (Fig. 3e).

Taken together, RANK expression increased after dual treatment with lapatinib and trastuzumab in HER2-positive human breast cancer cell lines, mimicking the results seen in breast cancer samples from the PAMELA trial (Fig. 2b). Additionally, two HER2-positive cell lines (SKBR3 and BT474) with acquired resistance to lapatinib (SKLR and BTLR) showed increased expression of RANK and several downstream targets, when compared to their respective parental controls (Fig. 3b–e).

To verify that RANK plays a direct role in the cellular response to lapatinib, we studied the consequences of RANKL stimulation and RANK loss in control and lapatinib-resistant SKBR3 cell lines. A small increase in lapatinib tolerance was observed in SKLR but not SKBR3 cells in the presence of RANKL (Fig. S5A). RANK silencing with two specific shRNAs reduced lapatinib resistance in SKLR cells, although sensitivity was not fully restored to the levels observed in WT cells (Fig. 4a, b). These results indicate that the activation of RANK signaling contributes to lapatinib resistance; however, it is not the only mechanism responsible for the emergence of resistance in SKLR cells. This is in line with the multiple anti-HER2 resistance mechanisms reported for these cells [39‐41].

Increased IκB and p65 phosphorylation was observed in lapatinib-resistant SKLR compared to SKBR3 cells upon RANKL stimulation (Fig. 4c, Fig. S5B and C), confirming the elevated NF-κB signaling in lapatinib-resistant cells. As expected, RANKL-induced NF-κB activation was abrogated upon RANK silencing in SKLR cells. RANKL treatment did not significantly alter the phosphorylation status of AKT nor ERK in SKBR3, SKLR, and the RANK-silenced cells (Fig. 4c, S5B and C). After lapatinib treatment, HER2, AKT, and ERK1/2 protein phosphorylation levels were undetectable in all cell lines, but baseline phosphorylation of p65 and IκB was maintained (Fig. 4c, S5B and C), demonstrating that NF-κB activation is not dependent on ErbB signaling and may support the survival of HER2-positive breast cancer cells in the presence of lapatinib.

To test if RANK signaling and enhanced NF-κB activation may directly contribute to resistance, we stably transduced HER2-positive cell lines with RANK overexpressing (psD69-RANK) and empty control (psD69-empty) vectors. RANK overexpression was confirmed by increased RANK mRNA levels (Fig. 5a) and induction of NF-κB downstream targets (BIRC3, ICAM1, TNFα, and IL8) in SKBR3, BT474, and HCC1954 cells (Fig. 5b). These RANK-overexpressing cell lines showed enhanced expression of all NF-κB targets analyzed after RANKL treatment compared to the corresponding parental cells (Fig. 5b). Next, we tested whether increased activation of RANK signaling would alter the cell response to lapatinib; RANKL stimulation of control cells (empty) did not alter lapatinib sensitivity (Fig. 5c). In contrast, RANK overexpression coupled with RANKL treatment resulted in an increased resistance to lapatinib in all HER2-positive cell lines tested (Fig. 5c), and this effect was abrogated by the RANKL inhibitor denosumab as expected (Fig. S6A).

We then analyzed RANK downstream signaling in these cell lines after treatment with lapatinib and/or RANKL. p65 was strongly phosphorylated in RANK-overexpressing cell lines upon RANKL treatment and in parental HCC1954 cells fitting with the higher RANK expression levels of these cells (Fig. 5d, Fig. S6B and C). Phosphorylation of p65 was not affected by lapatinib treatment. ERK1/2 phosphorylation levels increased after RANKL treatment to a greater extent in the RANK-overexpressing cells compared to control ones (Fig. 5d, Fig. S6B and C). AKT phosphorylation increased after RANKL stimulation in all cells irrespectively of RANK levels. Interestingly, RANKL-mediated activation of ERK1/2 and AKT in SKBR3 and BT474 cells overexpressing RANK was completely abrogated in the presence of lapatinib, meaning that ErbB signaling is required for RANK/RANKL-driven activation of ERK and AKT in these cells. In HCC1954 cells, AKT phosphorylation was also abolished by lapatinib. In contrast, the increased p-ERK levels upon RANKL stimulation in HCC1954 RANK-overexpressing cells were not affected by lapatinib (Fig. 5d, Fig. S6B and C).

In summary, enhanced RANK signaling in HER2-positive cells led to higher NF-κB activation and increased lapatinib resistance.

To investigate whether RANK/RANKL activation of ERK and AKT might take place, at least partially, via direct crosstalk with ErbB receptors, we compared the phosphorylation levels of HER2 in cells with and without RANK overexpression upon RANKL stimulation. RANK overexpression led to higher levels of p-HER2 in SKBR3 and BT474, but not in HCC1954 cells, compared with the corresponding controls (Fig. 5d, Fig. S6B and C). Importantly, in all HER2-positive cell lines, concomitant RANK overexpression and stimulation with RANKL resulted in decreased HER2 phosphorylation, indicating that RANKL might impinge on the HER2/ErbB signaling pathway (Fig. 5d).

To further study the putative crosstalk between RANK and ErbB signaling, we analyzed NF-κB and ErbB signaling after stimulation with ErbB ligands in RANK-overexpressing HER2-positive cell lines and corresponding controls at different time points. A slight increase in p65 phosphorylation was observed in SKBR3 and BT474 RANK-overexpressing cells compared with control cells (Fig. S7). EGF stimulation faintly increased p-p65 levels in HER2-positive cell lines, but this was not observed after heregulin (HRG) treatment (Fig. S7). As extensively reported [42], treatment with EGF and HRG efficiently induces ERK phosphorylation in all HER2-positive cell lines (Fig. S7), but no clear differences were observed between RANK-overexpressing cells and corresponding controls. Of note, 5 min after ErbB ligand stimulation, pERK levels are higher but a decrease in HER2 phosphorylation was observed, accompanied by less pERK and pHER2 after 10 min of ErbB ligand stimulation (Fig. S7). Thus, the reduced HER2 phosphorylation observed in RANK-overexpressing cells 10 min after RANKL stimulation may reflect previous activation of HER2/ERK signaling.

Due to the change in HER2 phosphorylation upon activation of RANK signaling with RANKL, we hypothesized that the two receptors might physically interact. To enable efficient immunoprecipitation and detection, we transiently co-expressed affinity-tagged versions of the receptors in HEK293 cells, including an amino (742-NTF) [30] and a carboxy-terminal fragment of HER2 (611-CTF) [31]. As shown in Fig. 6a, RANK-V5 was detected after immunoprecipitation of HER2 or 611-CTF HER2, but not in 742-NTF or any of the control samples (IgG), indicating that RANK interacts with the carboxy-terminal region of HER2. The reverse immunoprecipitation of RANK-V5 corroborated these results (Fig. 6b). To confirm the interaction between the two receptors under endogenous expression levels in the context of breast cancer, we chose SKBR3 cells that, compared to other breast cancer cell lines, express higher levels of HER2 and intermediate/lower levels of RANK and do not express EGFR [43]. HER2 was immunoprecipitated with the antibody trastuzumab (HCP) that binds to the HER2 extracellular domain, and the presence of RANK in the immunoprecipitate was tested by Western blotting. As seen in Fig. 6c, trastuzumab precipitated endogenous RANK demonstrating that the two receptors physically interact in breast cancer cells in an EGFR-independent manner.