Research ArticleCancer

Triple-negative breast cancers with amplification of JAK2 at the 9p24 locus demonstrate JAK2-specific dependence

See allHide authors and affiliations

Science Translational Medicine  13 Apr 2016:
Vol. 8, Issue 334, pp. 334ra53
DOI: 10.1126/scitranslmed.aad3001

Playing with JAKs

Janus kinase (JAK) proteins are well known to be involved in cancer progression, and drugs such as ruxolitinib target these proteins, specifically JAK1 and JAK2. Balko et al. demonstrated frequent amplification of JAK2 in triple-negative breast cancer, a particularly aggressive and deadly form of the disease and showed that it was associated with decreased survival. The authors observed that JAK2 inhibition was effective in treating this type of breast cancer in mouse models. They also found that inhibiting JAK1 along with JAK2 in this context rendered the treatment ineffective, explaining why ruxolitinib does not work in triple-negative breast cancer and suggesting that specific JAK2 inhibitors may be a better approach.


Amplifications at 9p24 have been identified in breast cancer and other malignancies, but the genes within this locus causally associated with oncogenicity or tumor progression remain unclear. Targeted next-generation sequencing of postchemotherapy triple-negative breast cancers (TNBCs) identified a group of 9p24-amplified tumors, which contained focal amplification of the Janus kinase 2 (JAK2) gene. These patients had markedly inferior recurrence-free and overall survival compared to patients with TNBC without JAK2 amplification. Detection of JAK2/9p24 amplifications was more common in chemotherapy-treated TNBCs than in untreated TNBCs or basal-like cancers, or in other breast cancer subtypes. Similar rates of JAK2 amplification were confirmed in patient-derived TNBC xenografts. In patients for whom longitudinal specimens were available, JAK2 amplification was selected for during neoadjuvant chemotherapy and eventual metastatic spread, suggesting a role in tumorigenicity and chemoresistance, phenotypes often attributed to a cancer stem cell–like cell population. In TNBC cell lines with JAK2 copy gains or amplification, specific inhibition of JAK2 signaling reduced mammosphere formation and cooperated with chemotherapy in reducing tumor growth in vivo. In these cells, inhibition of JAK1–signal transducer and activator of transcription 3 (STAT3) signaling had little effect or, in some cases, counteracted JAK2-specific inhibition. Collectively, these results suggest that JAK2-specific inhibitors are more efficacious than dual JAK1/2 inhibitors against JAK2-amplified TNBCs. Furthermore, JAK2 amplification is a potential biomarker for JAK2 dependence, which, in turn, can be used to select patients for clinical trials with JAK2 inhibitors.


Triple-negative breast cancer (TNBC) is the more lethal subtype of this disease, most of which exhibit basal-like patterns of gene expression. Currently, TNBC lacks clinically approved molecularly targeted therapies and is primarily treated with traditional chemotherapy and surgery. Although neoadjuvant (presurgery) chemotherapy (NAC) can be effective in eradicating TNBCs in the breast in about 30% of patients, many tumors do not respond or respond partially, eventually recurring as distant metastases. We and others have identified potential molecular mechanisms of drug resistance and poor outcome in TNBC (13). Some of these alterations are potentially actionable or targetable with molecularly directed therapies in clinical development.

We previously detected actionable alterations by targeted next-generation sequencing (tNGS) in the residual cancer of a cohort of patients who did not achieve a pathological complete response to NAC (2). Of the alterations identified in this study, amplifications or gains at the 9p24.1 locus, including Janus kinase 2 (JAK2), occurred at higher rates than in previous reports of untreated TNBC, suggesting a causal association with chemotherapy resistance. The JAK/signal transducer and activator of transcription (STAT) pathway is known to be mutated in other tumor types as well as myeloproliferative disease and has been shown to have a central role in driving normal and cancer stem cell growth (4, 5).

The mammalian JAK-STAT signaling pathway comprises four JAK domain–containing proteins [JAK1, JAK2, JAK3, and tyrosine kinase 2 (TYK2)] and seven STAT proteins (STAT1, STAT2, STAT3, STAT4, STAT5A, STAT5B, and STAT6) (6). Deregulation of this pathway has been implicated in the promotion of oncogenic phenotypes, including tumorigenesis, invasion, metastasis, proliferation, survival, angiogenesis, antiapoptosis, and immune evasion (7, 8). In breast cancer, the JAK-STAT pathway has been shown to be altered by the following mechanisms: (i) down-regulation of phosphotyrosine-specific phosphatases (911); (ii) increase in the amount of the JAK/STAT-activating ligand interleukin-6 (IL-6) (1214); (iii) activation of other upstream oncogenic pathways, such as ErbB1, c-Src (15), or phosphatidylinositol 3-kinase (PI3K)/mammalian target of rapamycin (mTOR) (16); and (iv) down-regulation of negative regulators of STAT such as suppressor of cytokine signaling 3 (SOCS3) (17).

Here, we demonstrate that a fraction of TNBCs harbor amplifications at the 9p24.1 locus, which includes JAK2. The copy number of the 9p24 amplicon increased during selection by chemotherapy and the development of metastasis, suggesting a role in tumor progression and therapeutic resistance. We found that JAK2 drives a JAK1/STAT3-independent signaling program that can be overcome using JAK2-specific inhibitors in combination with chemotherapy to reduce tumor-initiating stem cell–like cells and to suppress tumor growth and progression.


JAK2/9p24 amplifications are present in post-NAC TNBC

We previously reported an integrated molecular analysis (immunohistochemistry, gene expression, and tNGS) of a cohort of 74 post-NAC TNBCs (2). Patient demographics are reported in Table 1. From these data, we identified seven patients (10%) with amplifications in the JAK2 locus (9p24), a frequency higher than that in publically available patient cohorts of primary untreated TNBC or basal-like breast cancer (BLBC). All patients with JAK2 amplifications (JAK2AMP) detected by tNGS were confirmed by fluorescence in situ hybridization (FISH) with a 100% concordance rate (Fig. 1, A and B). All of these amplifications included the JAK2 gene, although most (88%) also included the immune checkpoint ligand PD-L1 (CD274) (fig. S1A). Frequently, multiple genes along an amplicon can contribute to oncogenic progression (1820). Supporting this idea, quantitative immunofluorescence for PD-L1 (programmed death ligand 1) revealed that putative JAK2AMP tumors also tended to have high PD-L1 expression, suggesting that CD274 may also be an important gene within this amplicon (fig. S1B).

Table 1. Clinical characteristics of cohort.

n = 111; median age, 48 years.

View this table:
Fig. 1. JAK2AMP is associated with a decreased response to NAC and poor patient survival.

(A) JAK2AMP breast tumor samples, detected by NGS, were validated by FISH and are represented as the average JAK2/centromere 9 (CEN9) ratio per cell (at least 30 cells counted for each case). Red bars show JAK2AMP cases and blue bars represent nonamplified controls. (B) Representative FISH images of JAK2/9p24 (red) and CEN9 (green). Four cases are depicted, with the upper left demonstrating a patient tumor with normal JAK2 (two copies). The remaining three cases showed JAK2 gains/amplification. Scale bars, 20 μm. (C and D) Kaplan-Meier curves assessing recurrence-free survival (RFS) and overall survival (OS) in patients with JAK2AMP and JAK2NML cancers [progression-free survival: hazard ratio (HR), 0.19; 95% confidence interval (CI), 0.05 to 0.73; overall survival: HR, 0.1; 95% CI, 0.02 to 0.44]. (E) IL6 mRNA in JAK2AMP (n = 4) and JAK2NML (n = 32) tumors quantified by NanoString analysis. Bars represent mean IL-6 mRNA expression. (F) RNA in situ analysis of JAK2AMP tumors for IL6 (red) and JAK2 (green). Scale bars, 20 μm. (G) Quantification of JAK2/IL6 RNAscope in three high-power fields across three JAK2AMP tumors, showing that single-positive cells are more common than dual-positive cells. (H) Individual data points for each of three JAK2AMP tumors analyzed by RNA in situ hybridization. Each point shows the number of single- or double-positive cells observed in each high-power field (n = 3 fields per sample).

Patients harboring JAK2/9p24 amplifications had poor Miller-Payne scores (21) in response to NAC, suggesting that these were highly chemotherapy-resistant tumors (57% category I in JAK2AMP versus 16% category I in JAK2NML) (Table 2). Consistent with this finding, JAK2AMP patients also experienced significantly inferior recurrence-free and overall survival (Fig. 1, C and D; P = 0.015 and P = 0.002, respectively), with all patients with JAK2-amplified tumors being deceased within <20 months after surgery. Although amplifications at 9p24 have been previously noted (10, 22), the genes within this amplicon that are associated with tumor aggressiveness are unclear. The JAK/STAT pathway is deregulated in a number of tumor types, with recent reports suggesting functional significance of JAK/STAT signaling in inflammatory and TNBC (4). Gene expression of IL6, an inflammatory cytokine involved in cell proliferation (23, 24) and wound healing (25) that can be secreted both from the tumor and the immune milieu to activate JAK/STAT signaling, was also significantly higher (P = 0.009) in JAK2AMP cancers, suggesting an autocrine or paracrine loop within these tumors (Fig. 1E). RNA in situ hybridization analysis of a JAK2AMP tumor suggested evidence of paracrine signaling, because individual cells tended to express either IL6 or JAK2 but not both (Fig. 1, F to H).

Table 2. Clinical comparison of patients by JAK2 status.

Characteristics of patients with complete tNGS data. OS, overall survival; RFS, recurrence-free survival.

View this table:

JAK2/9p24 amplifications are enriched by chemotherapy treatment

In examining patient data from primary untreated BLBC from The Cancer Genome Atlas (TCGA), we found that JAK2AMP occurred at a higher rate within our population of chemotherapy-treated TNBC (10% versus 2%; P = 0.08; Fig. 2A). Among amplifications and gains, JAK2 alterations were more frequent in untreated BLBC compared to untreated luminal breast cancers (Fig. 2B). JAK2 copy number was predicted JAK2 mRNA expression in both tumors and breast cancer cell lines, suggesting that copy number alterations can affect gene expression (fig. S2, A and B). Collectively, these data suggested that JAK2 gains or amplifications are predominant in TNBC and BLBC and are selected for by chemotherapy. Supporting this notion, we examined JAK2 copy number in two patients for whom serial specimens were available. JAK2 copy number was normal in the diagnostic biopsies from both patients but increased markedly in the post-NAC biopsy and was further elevated in DNA extracted from subsequent metastatic recurrences (Fig. 2C).

Fig. 2. Chemotherapy enriches the JAK2AMP tumor cell population.

(A) Comparison of JAK2 amplification rates in 68 TNBCs treated with NAC versus TCGA (primary basal-like). (B) Amplifications in JAK2 occur primarily in BLBC. Data were obtained through the cBio Web site for TCGA data access. Luminal, n = 324; basal, n = 81. (C) Enrichment of JAK2 gene copy number in longitudinal samples (pretherapy biopsy, post-NAC surgical specimen, and metastatic biopsy in two patients with TNBC). (D) JAK2/CEN9 copy number ratio across a series of 22 PDXs. PDX tumors (all annotations are truncated for readability and actual identifiers are BCM-XXXX, except for MC-1) were stained as a tissue microarray (TMA), with one to three independent cores (biological replicates) per PDX model. Bars represent means + SEM of the JAK2/CEN9 ratios for all PDXs. At least 30 cells were counted and averaged for each core. Samples in red are JAK2-amplified. Samples in patterned pink are matched PDXs from the same TNBC, established before (2147) and after (2277) chemotherapy. (E) JAK2 FISH copy number counts (relative to centromere 9 signal) at the single-cell level in matched untreated (2147) and postchemotherapy (2277) PDXs. Clinically amplified PDX (4013) and FISH from an untreated patient (primary JAK2AMP) demonstrating high level of amplification are plotted for comparison. The population of JAK-amplified cells was enriched in the PDX model established after chemotherapy treatment. P value represents result of a two-tailed t test.

To determine whether JAK2AMP was also present in experimental models of breast cancer, we performed FISH for JAK2 in sections from 22 patient-derived TNBC xenografts (PDXs). The characteristics of these xenografts have been previously described (26). Using a JAK2/centromere 9 ratio ≥2 as evidence of gene amplification, 2 of 22 (9%) xenografts were amplified at the JAK2 locus. This frequency was similar to that observed in our JAK2AMP patient cohort (all chemotherapy-resistant tumors), supporting the concept that JAK2/9p24-amplified tumors are highly aggressive and thus have improved cross-species engraftment rates (Fig. 2D). Two PDXs in this cohort were derived from the same patient before and after chemotherapy. Careful inspection of JAK2 copies in these PDXs at the single-cell level demonstrated a significant enrichment of tumor cells with JAK2 copy gains in the posttreatment PDX (2277) compared to the pretreatment PDX (2147; P = 0.004; Fig. 2E).

JAK2 mediates STAT6 but not STAT3 activation in breast cancer cells

To understand the importance of JAK2/9p24 gains and amplifications at the cellular level, we queried the Cancer Cell Line Encyclopedia (CCLE) for breast cancer cell lines with alterations in JAK2 copy number. As observed in the TCGA data, JAK2 copy gains and amplifications were more common in TNBC cell lines (Fig. 3A). We chose several of these cell lines, representing deep deletion (MDA-231, HCC1143), copy gain (HCC-70, HCC-38), and amplification (MDA-436, HCC1954). We also used SUM159PT cells, which were not included in the CCLE but have been previously characterized as reliant on JAK/STAT signaling (4). FISH analysis failed to demonstrate amplification of the 9p24 locus in SUM159PT but confirmed gain/amplification in HCC-70, HCC-38, and MDA-436 (Fig. 3B). JAK2 and baseline p-STAT3 (a known downstream effector) expression as measured by immunoblot generally correlated with JAK2AMP (Fig. 3C).

Fig. 3. JAK2 drives a STAT3-independent program in JAK2AMP TNBC cell lines.

(A) Analysis of CCLE data via the cBio portal to identify cell lines with JAK2 copy alterations. Solid red boxes, amplified; solid pink, gained; light blue, shallow deletion; dark blue, deep deletion; outlined pink, mRNA overexpression (ESR1 only). (B) FISH for JAK2 (red) and CEN9 (green) for breast cancer cell lines SUM159PT (JAK2NORMAL), HCC-70 (JAK2GAIN), HCC-38 (JAK2GAIN), and MDA-436 (JAK2AMP). Scale bars, 20 μm. (C) Cells were treated with 10% fetal bovine serum (FBS) ± 1 μM ruxolitinib for 24 hours and analyzed by immunoblot with the indicated antibodies. Cell line names in pink are JAK2GAIN and cell line names in red are JAK2AMP. (D) The indicated TNBC cell lines were treated with increasing doses of the JAK2-specific inhibitor BSK805 or ruxolitinib for 24 hours and analyzed by immunoblot. (E) TNBC cell lines were transfected with siControl (siCON), siJAK1, or siJAK2 or treated with 1 μM ruxolitinib. Two different sequences for each of JAK1 and JAK2 were used to confirm specificity. Cells were harvested 72 hours after siRNA transfection or 24 hours after ruxolitinib treatment and analyzed by immunoblot with the indicated antibodies. (F) The indicated cell lines were treated with serum-free or 10% serum–containing medium for 16 hours and then treated with 1 μM of BSK805 or ruxolitinib for 1 hour; cells were next stimulated with oncostatin M (OSM; 50 ng/ml) for 30 min, followed by cell harvest, preparation of lysates, and immunoblot analysis.

Treatment with ruxolitinib, a JAK1/2 inhibitor with a median inhibitory concentration (IC50) of 3.3 and 2.8 nM for JAK1 and JAK2, respectively (27), down-regulated both STAT1 and STAT3 phosphorylation at tyrosine (Y) 701 and 705 (the putative JAK phosphorylation sites), respectively, in all cell lines tested, but it had little effect on alternative STAT family members and JAK effectors STAT5 and STAT6 (Fig. 3C). To explore whether the effects of ruxolitinib on STAT3 phosphorylation were specific to JAK1 or JAK2 inhibition, we tested increasing concentrations of ruxolitinib against SUM159PT, HCC-38, and MDA-436 cells and compared the results to those of the JAK2 inhibitor NVP-BSK805, which has >20-fold selectivity over JAK1 and JAK3 (IC50: JAK2, 0.5 nM; JAK1, 31.6 nM; JAK3, 18.7 nM; and TYK2, 10.8 nM) (28). Treatment with ruxolitinib inhibited STAT3 phosphorylation in a dose-dependent fashion, whereas STAT3 phosphorylation was resistant to BSK805 at all doses (Fig. 3D). To confirm these results, we determined how specific small interfering RNA (siRNA) knockdown of JAK1 (siJAK1), JAK2 (siJAK2), or both JAK1 and JAK2 (siJAK1/2) or treatment with ruxolitinib would alter STAT3 phosphorylation in four TNBC cell lines (Fig. 3E). Down-regulation of JAK1 with siRNA (siJAK1) reduced Y705 p-STAT3, whereas down-regulation of JAK2 did not affect STAT3 phosphorylation. These data suggest that ruxolitinib treatment reduces STAT1/STAT3 signaling through inhibition of JAK1 but not JAK2 (Fig. 3E), thus arguing against the notion that JAK2 is required for STAT3 phosphorylation in breast cancer. Because basal activation of STATs may be regulated by substantially different mechanisms than cytokine-induced STAT activation, we pretreated HCC-38 and MDA-436 cells with BSK805 or ruxolitinib under serum-starved or serum-stimulated conditions and then activated STATs with recombinant oncostatin M. Consistent with our previous results, JAK2-selective inhibition did not affect STAT1 or STAT3 phosphorylation but did reduce or abrogate STAT5 phosphorylation. JAK1/2 inhibition with ruxolitinib completely abrogated STAT1, STAT3, and STAT5 phosphorylation (Fig. 3F).

STAT3 is the most studied downstream effector of STAT signaling in breast cancer (8). To determine the transcriptional effect of STAT3 inhibition in these cell lines, we performed microarray analysis in HCC1143 (low JAK2), HCC-70 (JAK2-gained), HCC-38 (JAK2-gained), and SUM159PT (JAK2-overexpressing) cells after 4 or 24 hours of JAK inhibition with ruxolitinib. STAT3 phosphorylation was completely inhibited in all of the cell lines (fig. S3A). Surprisingly, at a false discovery rate (FDR) of 5%, no genes were significantly altered in HCC1143, HCC-70, and HCC-38, and only three genes were altered in SUM159PT (fig. S3, B and C). These results suggest that STAT3 does not have a substantial transcriptional function in these cell lines in vitro. Alternatively, STAT3 may only function within a minor population (for example, a cancer stem cell–like population), which may dilute the transcriptional effects observed. Consistent with the microarray results, ruxolitinib treatment (<10 μM) did not inhibit cell proliferation or viability in vitro, either alone (fig. S4A) or in combination with the chemotherapy drugs doxorubicin and docetaxel (fig. S4B).

JAK2 signaling affects mammosphere potential

Because JAK/STAT signaling plays a role in stem cell functionality and self-renewal (4, 24), we determined whether specific siRNA knockdown of JAK1 (siJAK1), JAK2 (siJAK2), or both JAK1 and JAK2 (siJAK1/2) or treatment with ruxolitinib would affect mammosphere formation, a marker of the tumor-initiating or cancer stem cell-like population. We found that in both HCC-38 (JAK2GAIN) and MDA-436 (JAK2AMP), siJAK1 enhanced mammosphere formation, but siJAK2 reduced mammosphere-forming ability (fig. S5). The effect of JAK1 knockdown with siRNA in HCC-38 cells was abrogated by JAK2 knockdown when both siRNAs (siJAK1/2) were used in combination. Consistent with this result, ruxolitinib (JAK1/2 inhibitor)–treated HCC-38 cells also exhibited increased mammosphere formation compared to untreated cells, although the same result was not observed in MDA-436 cells. In these cells, siJAK2 and siJAK1/2, but not siJAK1, inhibited mammospheres to the same degree, suggesting a dominant JAK2-specific effect (fig. S5). Once again, the effect of JAK1/2 knockdown on mammosphere formation in MDA-436 cells was phenocopied by JAK1/2 inhibition with ruxolitinib (fig. S5B).

We next used JAK2-targeted short hairpin RNAs (shRNA) in a doxycycline-inducible vector. To select for cells with mammosphere-forming capacity, cells were treated with paclitaxel, a chemotherapeutic drug known to spare cancer stem cell–like with tumor-initiating capacity (29, 30). In both HCC-38 and MDA-436 cells, doxycycline-induced inhibition of JAK2 abrogated the enrichment of mammosphere-forming, drug-resistant cells spared by paclitaxel (Fig. 4, A to D). Inducible JAK2 knockdown was confirmed by immunoblot (Fig. 4E). Pharmacological inhibition with the JAK2-specific inhibitor BSK805 produced similar results in both cell lines, further suggesting that the inhibition of mammospheres is a JAK2-specific effect (Fig. 4F and fig. S6). A continued effect of JAK2 inhibition or inducible shJAK2 on mammosphere potential was observed when HCC-38 mammospheres were dissociated and replated in a secondary sphere formation assay (fig. S6, C and D). To directly assess tumorigenic potential, we performed a limiting dilution of MDA-436 or HCC-38 cells (1 × 106, 1 × 105, or 1 × 104 cells) injected orthotopically into nude mice, with or without pretreatment with BSK805 or vehicle control, or using doxycycline-inducible shRNAs [nontargeting shRNA control (shNTC) or shJAK2, #49] pretreated with doxycycline (Fig. 4G). shJAK2 induction significantly reduced the tumorigenic population of both cell lines (P = 2.49 × 10−7 and P = 6.23 × 10−8 in MDA-436 and HCC-38 cells, respectively), whereas BSK805 was effective in reducing the tumor-initiating population in MDA-436 cells (P = 0.00132) but not in HCC-38 cells (P = 0.68). However, the HCC-38 model often forms small dormant nodules that inconsistently develop into malignant tumors, complicating the analysis of the results with this cell line.

Fig. 4. JAK2 knockdown abrogates tumorsphere expansion after chemotherapy.

(A) HCC-38 cells stably transduced with doxycycline-inducible shRNA targeting JAK2 (two independent sequences) or shNTC were grown in 10% FBS and treated with vehicle (control), IC50 of paclitaxel (50 nM) ± doxycycline (100 ng/ml), or doxycycline for 4 days and allowed to recover in fresh medium for 3 days; cells were then trypsinized and assessed for their ability to form mammospheres. Colony numbers are expressed relative to untreated controls. (B) On day 7, colonies were stained with MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide] and photographed. Scale bars, 200 μm. (C and D) Experiments identical to (A) and (B) were carried out in MDA-436 cells (paclitaxel IC50, 150 nM). (E) Immunoblot analysis demonstrating doxycycline-inducible knockdown of JAK2 at 72 hours in HCC-38 and MDA-436 cells. (F) MDA-436 and HCC-38 cells were grown in 10% FBS and treated with vehicle (control), paclitaxel (at the cell line IC50) ± 5 μM BSK805, or 5 μM BSK805 alone for 4 days and then allowed to recover in fresh medium for 3 days. Colonies were quantitated as described above. All experiments were replicated at least twice (n = 3). (G) Limiting dilution assay results for MDA-436 or HCC-38 cells treated with shNTC + doxycycline or shJAK2 + doxycycline, or parental cells treated with BSK805 (1 μM) or vehicle control. Cells were pretreated for 1 week before inoculation, and treatment was maintained in the mice for 1 week after orthotopic (no. 4 mammary fat pad) injection. Data are presented as the number of tumors (palpable at 30 days for MDA-436 and either palpable or confirmed microscopically after biopsy at 60 days for HCC-38) formed out of the number of mice inoculated. P value represents the χ2 assay result. Statistics were performed using the protocol at

Finally, we asked whether the addition of BSK805 to paclitaxel would improve therapeutic efficacy in vivo. In mice bearing MDA-436 or HCC-38 xenografts, the addition of BSK805 to paclitaxel markedly reduced tumor growth. Two of seven HCC-38 tumors were completely eliminated and did not recur up to 30 days after withdrawing treatment (Fig. 5, A and B). Dissociated cells from tumors harvested at the completion of treatment were evaluated in mammosphere assays. Tumor cells from xenografts treated with paclitaxel exhibited increased mammosphere formation compared to cells from untreated tumors or cells from tumors treated with paclitaxel and BSK805 (Fig. 5, C and D), suggesting that inhibition of JAK2 suppressed chemotherapy-resistant cells with tumor-initiating capacity. We also evaluated the efficacy of JAK2 inhibition in combination with paclitaxel or as a single agent against TNBC PDX-4013 (Fig. 5E). This PDX was identified as JAK2-amplified (Figs. 2D and 5F). BSK805 treatment, alone or in combination with paclitaxel, substantially reduced tumor growth over vehicle control. In addition, we treated four other PDX models of TNBC: two of these demonstrated JAK2GAIN [BCM-4272 (low gain) and BCM-2147 (high gain)] and two had normal JAK2 copy numbers (BCM-4664 and BCM-3107; all shown in Fig. 2D). The models with JAK2GAIN were intrinsically resistant to taxanes (docetaxel), similar to findings in our clinical data. Although JAK2-specific inhibition did not have substantial antitumor effects in any of the four models, JAK2 inhibition overcame intrinsic docetaxel resistance in the PDXs with JAK2GAIN (fig. S7). Collectively, these results suggest that therapeutic benefit may be obtained from the addition of JAK2-specific inhibitors to chemotherapy in breast cancers with JAK2 copy gains or gene amplification.

Fig. 5. Pharmacological JAK2 inhibition in vivo abrogates tumor-initiating potential after chemotherapy.

(A and B) Female athymic mice were injected with MDA-436 or HCC-38 cells in the no. 4 mammary fat pad. Mice bearing tumors ≥150 mm3 were randomized to treatment with vehicle, paclitaxel [20 mg/kg per day × 4 doses intraperitoneally (i.p.)], or paclitaxel (20 mg/kg per day × 4 doses i.p.) + BSK805 [100 mg/kg per day orally (p.o.)]. Paclitaxel doses are represented by arrows. Tumor volumes were measured twice weekly. Two complete responses to dual therapy were achieved in mice bearing HCC-38 tumors. Bars represent means ± SEM. Differences were analyzed by one-way analysis of variance (ANOVA) with Tukey’s contrasts. (C) Representative images of mammospheres from treated tumors in (A) and (B). Scale bars, 200 μm. (D) Quantification of mammospheres from tumors harvested at the end of treatment. Bars represent means + SEM for n = 9 measurements. (E) Severe combined immunodeficient (SCID)/beige mice implanted with PDX model PDX4013 were randomized to treatment with vehicle (intraperitoneal saline and oral gavage with oral suspension agent), paclitaxel (20 mg/kg per day × 4 doses i.p.), BSK805 (80 mg/kg per day p.o.), or paclitaxel (20 mg/kg per day × 4 doses i.p.) + BSK805 (80 mg/kg per day p.o.). Tumor volumes were measured twice weekly. (F) Image of JAK2-FISH in the PDX4013 model demonstrating gene amplification. Scale bar, 20 μm.


Here, we report a series of studies supporting the role of JAK2 in TNBC. JAK2 amplification was more frequent in TNBC treated with chemotherapy than in newly diagnosed untreated tumors. In addition, JAK2 copy number increased in a limited number of patients for whom serial biopsies were collected at different times in their TNBC progression. The rate of JAK2 amplification in a panel of serially passaged patient-derived breast xenografts was similar to that observed in residual TNBC after NAC, altogether supporting a role for JAK2 in drug resistance and tumor-initiating or cancer stem cell–like capacity.

The JAK2 gene is located in 9p23-24, which also contains other candidate oncogenes such as GASC1 (also known as JMJD2C/KDM4C), UHRF2, KIAA1432, C9orf123, and PD-L1 (CD274) (10, 31). Here, 88% of the JAK2-amplified TNBC also overexpressed PD-L1. Nodular sclerosis Hodgkin’s lymphoma also exhibits amplification of 9p24.1, which contains both JAK2 and PD-L1 loci (31, 32). In these cells, JAK2 induces transcription of the PD-1 ligand via STAT1 and is associated with increased sensitivity to JAK2 inhibitors. A similar association in TNBC remains to be explored, as does the possibility that breast tumors with an operative JAK2–PD-1 axis would be more sensitive to immune checkpoint inhibitors (33). Rui et al. (34) explored the cooperation of different genes within the 9p24 amplicon in primary mediastinal B cell lymphoma and nodular sclerosis Hodgkin’s lymphoma, showing a synergistic antitumor effect of JAK2 and JMJD2C inhibition. These data suggest that this amplicon contains multiple oncogenes that can act cooperatively or independently. Thus, to define the role of JAK2 in TNBC, we performed genetic and pharmacological inhibition of JAK2 in JAK2-amplified breast cancer cells and xenografts as well as PDXs.

We treated JAK2AMP HCC-38, HCC-70, and MDA-436 TNBC cells and JAK2-overexpressing SUM159PT cells with ruxolitinib, an adenosine 5′-triphosphate–competitive inhibitor of JAK1 and JAK2, approved for the treatment of intermediate- or high-risk myelofibrosis (35). Treatment with ruxolitinib alone or in combination with chemotherapy at doses that eliminated p-STAT3 did not affect gene expression, cell viability, or mammosphere formation in cell lines with JAK2AMP. One caveat to this trend was the MDA-436 cell line, which did demonstrate a reduction in mammosphere formation with ruxolitinib treatment. These cells did not exhibit a reduction in mammosphere formation with JAK1 siRNA, suggesting that the JAK2-mediated effects of ruxolitinib potentially explained the reduction in MDA-436 mammosphere formation. In general, however, these results are concordant with the findings of Barbie et al. (36) where treatment with ruxolitinib had no effect in a different panel of TNBC cell lines. Because STAT3 is activated/phosphorylated by JAK1 or JAK2 (37), we performed siRNA knockdown of JAK1 and JAK2 in HCC-70, HCC-38, and MDA-436 cells. Knockdown of JAK1 inhibited p-STAT3 but not p-STAT6, whereas knockdown of JAK2 did not inhibit p-STAT3. These findings support STAT3 activation by JAK1, as previously shown by Britschgi et al. (38) in MDA-468, MDA-231, and 4T1 TNBC cells. Furthermore, our findings suggest that STAT3 may not be the primary effector of the antitumor effects of JAK2 inhibitors.

Marotta et al. (4) recently reported the role of the IL-6/JAK2/STAT3 pathway in the growth of stem cell–like human breast cancer cells. This association was preferentially active in CD44+/CD24 cells, which occur with higher frequency in BLBCs and was confirmed in a panel of TNBC cell lines. Unlike our study, this panel did not include cells with JAK2AMP. In JAK2AMP HCC-38 and MDA-436 cells, knockdown of JAK2 but not JAK1 inhibited mammosphere formation, both in the presence and absence of paclitaxel. Further, paclitaxel induced mammosphere formation in these JAK2AMP cell lines, and this effect was dampened by knockdown of JAK2 with a doxycycline-inducible shRNA. Treatment with NVP-BSK805, a small-molecule inhibitor with >20-fold selectivity for JAK2 over JAK1, did not inhibit basal p-STAT3 but did inhibit mammosphere formation. Similar results were observed in vivo, where HCC-38 and MDA-436 cells harvested from tumors treated with paclitaxel and BSK805 exhibited markedly reduced mammosphere-forming capacity compared to cells from tumors that had been treated with paclitaxel alone. Together, these results support a connection between STAT3-independent JAK2 signaling and a stem cell–like phenotype. They also suggest that targeted development of selective JAK2 inhibitors in patients with TNBC and BLBC may be useful, particularly in combination with chemotherapy.

There are limitations to this study in that it will need further research to effectively elucidate the role of JAK1 and JAK2 in TNBC and determine how signaling through these molecules is affected by JAK2AMP. First, transcriptional and signaling events affected by JAK1/2 inhibition or activation may be diluted and overlooked if they occur within small populations of cells, such as cancer stem cell–like cells. This could be one potential explanation for the observed lack of major transcriptional changes in breast cancer cells as a result of ruxolitinib treatment in our study. Second, the role of STAT3 and other STATs in tumor phenotypes (metastasis, the immune microenvironment, etc.) remains to be elucidated. A large amount of work will be required to establish which STATs are the primary effectors of the cancer stem cell–like cell phenotype in JAK2AMP cells and tumors.

Despite these limitations, these data are consistent with a process of JAK2-supported clonal evolution/selection after neoadjuvant therapy. Increased JAK2 copy number correlates with tumor progression, acquisition of drug resistance, and poor patient outcome. In concordance with previous reports, our results show that JAK2 is involved in cancer cell “stemness.” This role appears limited to breast cancers with JAK2 gene amplification and has similarities with hematologic neoplasms that have altered JAK2 signaling. On the basis of these observations, we propose that JAK2 is a therapeutic target in JAK2AMP TNBC that can be blocked with inhibitors with predominant activity against JAK2, such as NVP-BSK805. Our results raise the consideration that pan-JAK inhibitors with activity against JAK1 may be less effective against JAK2-amplified TNBC compared to JAK2-selective inhibitors, with potential additional toxicity.


Study design

This study was designed to evaluate the role of JAK2 amplifications observed through NGS efforts in TNBCs. Demonstration of JAK2 amplifications in clinical samples was used to justify cell culture and in vivo (xenograft and PDX) models demonstrating that JAK2 may be a useful target in TNBCs, particularly those with amplifications of the JAK2 locus (9p24). For mouse studies, randomization to individual treatment groups was made once tumors reached a target of 100 to 500 mm3, depending on the tumor model. All experiments were replicated at least twice for a total of three independent experiments. Studies or experiments with more than three replicates are noted in the figure legends.

Patients and tumor specimens

Surgically resected tumor samples (n = 111) were from patients with TNBC diagnosed and treated with NAC at the Instituto Nacional de Enfermedades Neoplásicas. Clinical and pathological data were retrieved from medical records under an institutionally approved protocol (INEN 10-018). Tumors were determined to be triple-negative if they were negative for ER (estrogen receptor), PR (progesterone receptor), and HER2 (human epidermal growth factor receptor 2) overexpression measured by immunohistochemistry. The results were further verified by comparison with the NGS results.

Fluorescence in situ hybridization

JAK2 copy number was quantified by comparing the ratio of JAK2 to centromere 9 probe signals. Tumor cells were examined directly using an Olympus AX70 epifluorescence microscope equipped with narrow band-pass filters. Each slide was initially scanned at low power to identify appropriate areas of tumor tissue with clearly defined nuclei. The 100× objective was then used to score signals in 40 to 50 nonoverlapping tumor cell nuclei to determine the average number of JAK2 and centromere 9 copies per cell. The risk of sampling error resulting from tissue heterogeneity was minimized by scoring at least five tumor areas for each specimen. In two cases, fewer nuclei were interpretable, and these counts were normalized to a value for 40 cells. The ratio of the JAK2 and centromere 9 copy number averages was used to determine the presence of JAK2 gene amplification. Specimens with a JAK2/centromere 9 ratio ≥ 2 were scored as positive for JAK2 gene amplification. Specimens were considered to have JAK2 copy number gain if they had an average of ≥ 3 JAK2 signals.

RNA in situ hybridization

RNA in situ hybridization was performed on 5-μm formalin-fixed paraffin-embedded sections using RNAscope reagents (Advanced Cell Diagnostics), with probes specific for human JAK2 and IL6 using the manufacturer’s recommended protocol.

Automated quantitative analysis for PD-L1

PD-L1 expression was quantified in TMAs using automated quantitative analysis as described previously (39). Automated quantitative analysis allows exact and objective measurement of fluorescence intensity within a defined tissue area, as well as within subcellular compartments. Briefly, a series of monochromatic high-resolution images were captured using an epifluorescence microscope platform, and the signal intensity of the target of interest was measured according to a previously described algorithm (40). For each TMA histospot, images were obtained for each fluorescence channel: DAPI (4′,6-diamidino-2-phenylindole; nuclei), Alexa 546 (cytokeratin), or Cy5 (target probe). To distinguish tumor from stroma, an epithelial tumor “mask” was created by dichotomizing the pan-cytokeratin signal. Target protein was quantified in the tumor (cytokeratin-positive), the stroma (absence of cytokeratin positivity), or the total tissue area (all DAPI-positive cells) (41).

Immunoblot analysis

Immunoblot analysis was performed as previously described (42) using antibodies for β-actin (#4970), calnexin (#2679), JAK1 (#3344), JAK2 (#3230), p-STAT3 (Y705; #9145), p-STAT6 (Y641; #9364), STAT3 (#12640), p-STAT5 (Y694; #9359), and p-STAT1 (Y701; #7649), all of which were purchased from Cell Signaling Technology. Immunoreactive bands were detected by enhanced chemiluminescence after incubation with horseradish peroxidase–conjugated secondary antibodies (Promega).

Chemicals and inhibitors

BSK805 was obtained from Novartis and solubilized in dimethyl sulfoxide (DMSO) for cell culture or suspended in hydroxypropyl methylcellulose and Tween 80 for studies in vivo. Ruxolitinib for cell culture was purchased from Selleckchem. Doxycycline was purchased from Sigma-Aldrich. Paclitaxel for injection was obtained from the Vanderbilt University Hospital Outpatient Pharmacy. BMS-911543 was obtained from Bristol-Myers Squibb (BMS) and suspended in an 80:20 mixture of polyethylene glycol 400 and 200 mM citrate buffer (pH 3).

Cell lines

MDA-436 and MDA-231 were obtained from the American Type Culture Collection (ATCC) and grown in Dulbecco’s modified Eagle’s medium (DMEM) + 10% FBS. HCC-38, HCC1954, HCC1143, and HCC-70 were also obtained through ATCC (ICBP50 panel) and cultured in RPMI 1640 + 10% FBS. SUM159PT cells were a gift from the laboratory of J. Pietenpol (Vanderbilt University). Cell lines were either obtained directly from ATCC or had been recently fingerprinted in a previous study (43).

siRNA knockdown

Cells in 60-mm dishes were transfected with siRNA targeting JAK1 (s7647 and s7648, Ambion) or JAK2 (s7648 and s7650, Ambion) or nonsilencing control using DharmaFECT 4 (Dharmacon) transfection reagent according to the manufacturer’s protocol.

Lentiviral vector transduction

shRNAs for JAK2 clone IDs V2LHS_61653 and V2LHS_61649 were obtained from the Open Biosystems shRNA library and cloned into pIND-CLucZ, a gift from T. Stiewe (Addgene plasmid #53224). Purified plasmid was transfected into 293FT cells along with psPAX2 and pMD2.G to generate lentivirus. Conditioned medium was applied to target cells (MDA-436 and HCC-38) in the presence of Polybrene for 2 days before puromycin selection. Expression of JAK2 in the presence or absence of doxycycline (100 ng/ml) was confirmed by real-time quantitative polymerase chain reaction (qPCR) using previously published methods (44) and the following primer sequences: forward, 5′-TCTTTCTTTGAAGCAGCAAG-3′; reverse, 5′-CCATGCCAACTGTTTAGCAA-3′.

Viability assays

Viability was ascertained by sulforhodamine B, as previously described (1, 2, 42, 44).

Mammosphere assays

Mammosphere assay conditions have been previously described (31, 44). Briefly, single-cell suspensions were seeded in six-well ultralow attachment plates (Corning) in serum-free DMEM/F12 with recombinant human epidermal growth factor (20 ng/ml; R&D Systems), hydrocortisone (Sigma-Aldrich), and 1× B27 (Invitrogen). Cell viability was assessed by MTT staining, and the number of mammospheres measuring >100 μm was determined in an automated fashion using the GelCount mammalian cell colony counter (Oxford Optronix).

Xenograft studies

Female athymic mice were injected with HCC-38 or MDA-436 cells in the no. 4 mammary fat pad. After 4 weeks, mice bearing tumors ≥150 mm3 were randomized to treatment with vehicle, paclitaxel (20 mg/kg per day i.p. for 4 days), or paclitaxel + BSK805 (100 mg/kg per day p.o.). Tumor diameters were measured using calipers twice per week, and volume in cubic millimeter was calculated with the following formula: volume = (width2 × length)/2. For patient-derived xenografts, BCM-4013, BCM-2147, BCM-4272, BCM-3107, and BCM-4664 were derived by transplantation of a fresh patient breast tumor biopsy (ERPRHER2) into the cleared mammary gland fat pad of immunocompromised SCID/beige mice; tumor samples (2 × 2 mm) were serially passaged in SCID/beige mice by fat pad transplantation under general anesthesia. When tumors reached an average size of 150 to 500 mm3, mice were randomized (n = 5 per group) to treatment with vehicle, docetaxel (20 mg/kg i.p.), BMS-911543 (JAK2 inhibitor; 30 mg/kg p.o.), or docetaxel + BMS-911543. Animals received three cycles of treatment, 14 days each: docetaxel was delivered on day 1, and BMS-911543 was administered daily for 5 days, followed by 2 days off for two consecutive weeks. Tumor diameters were measured using calipers as above.

Microarray analysis

For microarray experiments, cells were treated for 4 or 24 hours in the presence of DMSO or ruxolitinib (1 μM) and harvested on ice for RNA purification. Total RNA was purified using a Maxwell 16 (Promega) instrument. The experiment was repeated for three independent biological replicates before microarray hybridization to Affymetrix Human Gene 1.0 ST arrays. Data were normalized in R using the RMA (Robust Multichip Array) algorithm, filtered on the basis of variance across all samples, and collapsed to the gene level using the genefilter package before analysis by one-way ANOVA (for each cell line). No fold change cutoffs were used; all gene expression data were filtered on the basis of statistical significance. P values were corrected for FDR using the Benjamini and Hochberg method (45). Significantly altered genes (FDR <5%) were compared among the treatment groups for each cell line using Tukey’s post hoc test.

Genomic data analysis

Genomic data from TCGA breast (46) and CCLE (47) were accessed through cBioPortal (48).

Statistical analysis

For cell proliferation assays, significant differences were determined by repeated measures ANOVA with Bonferroni’s post hoc test, except as noted below. Paired t tests were used to determine significant differences in siRNA proliferation assays, Caspase-Glo 3/7 assay, real-time qPCR assays, and immunohistochemistry scores. A P value of <0.05 was considered statistically significant. Bar graphs show means ± SD, unless otherwise stated in the figure legend.


Fig. S1. Focal and nonfocal amplifications at 9p24/JAK2 exist in ER-negative breast cancer.

Fig. S2. JAK2 copy number is associated with JAK2 gene expression.

Fig. S3. Ruxolitinib has low impact on gene expression in vitro.

Fig. S4. Ruxolitinib does not reduce cell viability or enhance chemosensitivity in vitro.

Fig. S5. JAK2 knockdown reduces mammosphere potential in HCC-38 and MDA-436 cells.

Fig. S6. Pharmacological inhibition of JAK2 reduces mammosphere formation after chemotherapy selection in vitro.

Fig. S7. JAK2 inhibition overcomes taxane resistance in PDXs with JAK2GAIN.


Acknowledgments: SUM159PT cells were a gift from the laboratory of J. Pietenpol (Vanderbilt University). The pIND-CLucZ was supplied by T. Stiewe (Addgene plasmid #53224). NVP-BSK805 was provided by Novartis. BMS-911543 was provided by BMS. Funding: This study was funded by the Department of Defense Breakthrough Award BC131494 (J.M.B., R.S.C., M.E.S., and J.M.G.), a grant from the IBC Network Foundation (J.M.B.), and S. G. Komen for the Cure Foundation grants SAC100013 (C.L.A.) and CCR14299052 (J.M.B.). Other sources of support include NIH/National Cancer Institute grant K99/R00-CA181491 (J.M.B.), Breast Cancer Specialized Program of Research Excellence (SPORE) grant P50 CA098131, the Vanderbilt-Ingram Cancer Center Support grant P30 CA68485, and BMS. Author contributions: J.M.B. and C.L.A. were responsible for study design and direction. J.M.B. and L.J.S. performed the experiments and analyzed the results. All authors contributed to data analysis, interpretation, and authorship of the final manuscript. Competing interests: C.L.A. serves in a scientific advisory role to Novartis, AstraZeneca, Genentech, Millennium, Celgene, Roche, and Pfizer. He serves on the Scientific Advisory Board of the Komen Foundation. J.M.B. receives research support from Genentech/Roche, Bristol-Myers Squibb (BMS), and Incyte Corporation, has received consulting/expert witness fees from Novartis, and is an inventor on provisional patents for various cancer immunotherapy targets and biomarkers. V.A.M. is an employee and shareholder of Foundation Medicine and R.Y. and P.J.S. were at the time this study was conducted. V.A.M. receives patent royalties for T790M, which is licensed through Memorial Sloan Kettering Cancer Center (MSKCC). J.M.G. is now an employee of Genentech but was at Vanderbilt at the time of the study.D.L.R. is a consultant, advisor and/or serves on a Scientific Advisory Board for Genoptix, Ventana, Amgen, AstraZeneca, Agendia, Biocept, BMS, Cell Signaling Technology, Cepheid, Daiichi Sankyo, GSK, InVicro/KonicaMinolta,Merck, NanoString, PerkinElmer, PAIGE, and Ultivue, holds equity in PixelGear, and receives research support from Astra Zeneca, Cepheid, Navigate/Novartis, NextCure, Lilly, Ultivue, and PerkinElmer. The other authors declare that they have no competing interests. Data and materials availability: Microarray data have been deposited at the Gene Expression Omnibus (GEO), with accession GSE70508.

Stay Connected to Science Translational Medicine

Navigate This Article