Safety and Tolerability
All patients experienced adverse events (Table 2. The nature and frequency of adverse events were similar in the two phase I cohorts. The majority of adverse events were mild to moderate in severity (grade 1–2). The profile of adverse events of the combination was dominated by the ones associated with carotuximab[23, 28–32]. Adverse events typically associated with aromatase inhibitors (hot flashes, joint stiffness) and metabolic adverse events typically associated with everolimus (dyslipidemias) arose after cycle 1 and were comparatively less frequent.
Table 2
Adverse events classified by time of occurrence (cycle 1 vs. cycle 2–6) in decreasing order of frequency.
| Cycle 1 (MTD period) n = 15 | Cycles 2 through 6 n = 12a |
| Number of patients (percent) |
Adverse event | Any grade | Grade ≥ 3 | Any grade | Grade ≥ 3 |
Headache | 10 (67) | 2 (13) | 11 (92) | 0 |
Fatigue | 7 (47) | 0 | 12 (100) | 0 |
Facial flushing and swelling | 7 (47) | 0 | 6 (50) | 0 |
Gingival hemorrhage | 6 (40) | 0 | 11 (92) | 0 |
Epistaxis | 5 (33) | 0 | 12 (100) | 0 |
Nausea and vomiting | 4 (27) | 0 | 1 (8) | 0 |
Nasal congestion | 3 (20) | 0 | 8 (67) | 0 |
Rash | 3 (20) | 0 | 9 (75) | 0 |
Mucositis, oral ulcers, oral soreness | 3 (20) | 0 | 5 (42) | 0 |
Telangiectasias | 2 (13) | 0 | 3 (25) | 0 |
Gingival tenderness | 1 (7) | 0 | 3 (25) | 0 |
Dysgeusia | 1 (7) | 0 | 1 (8) | 0 |
Insomnia | 1 (7) | 0 | 1 (8) | 0 |
Tingling, numbness | 1 (7) | 0 | 2 (17) | 0 |
Diarrhea | 1 (7) | 0 | 2 (17) | 0 |
Depression | 1 (7) | 0 | 0 | 0 |
Shaking chills and sweats | 1 (7) | 0 | 1 (8) | 0 |
Constipation | 0 | 0 | 1 (8) | 0 |
Weight loss | 0 | 0 | 1 (8) | 0 |
Nail discoloration | 0 | 0 | 1 (8) | 0 |
Dyspnea | 0 | 0 | 2 (17) | 0 |
Hot flashes | 0 | 0 | 1 (8) | 0 |
Joint stiffness | 0 | 0 | 4 (33) | 0 |
Edema | 0 | 0 | 1 (8) | 0 |
Lab abnormalities |
Hyperglycemia | 1 (7) | 0 | 1 (8) | 0 |
Anemia | 0 | 0 | 4 (33) | 0 |
Hypercholesterolemia | 0 | 0 | 2 (17) | 0 |
Hypertriglyceridemia | 0 | 0 | 2 (17) | 0 |
aThree patients were taken off study after completion of cycle 1 due to intercurrent illness, noncompliance, progressive disease (each, n = 1) |
Supplementary table 1. Pharmacokinetic analyses of everolimus and letrozole. Units of measurement in parentheses. For all parameters the geometric mean and geometric coefficient of variation (in parentheses) are provided except Tmax (median (range)) and half-life (arithmetic mean (standard deviation)). |
At least two signs of mucocutaneous telangiectasia (epistaxis, gingival bleeding, and telangiectasia), that typify hereditary hemorrhagic telangiectasia (HHT) were seen in 40% (6/15) of patients during cycle 1 and 92% (11/12) of patients who continued on the investigational therapy beyond cycle 1. HHT1 or Osler-Rendu-Weber syndrome 1 is a genetic disorder characterized by a deficiency in endoglin. Headache was the most frequent adverse event during cycle 1 and constituted the only dose limiting toxicity (DLT). As previously reported[32], headache was not associated with hypertension or focal neurologic signs. In the patient in whom imaging was performed, no radiographic abnormalities were seen. “First-dose” headache has been mitigated by splitting the first dose of carotuximab in 2 separate infusions[32], a strategy that was adopted in our trial. Although everolimus has also been associated with headache (11% when used in combination with letrozole [12]), headache was temporally associated with the infusions of carotuximab.
Although the adverse events were mostly grade 1 or 2, they all tended to persist, accumulate, and worsen with subsequent cycles of therapy. These adverse events were most notably fatigue which all patients beyond cycle 1 experienced; normocytic normochromic hypoproliferative anemia suspected to result from carotuximab’s effect on the proerythroblasts[23, 33]; epistaxis which would be triggered by progressively milder stimuli or even occur spontaneously. Although gingival hemorrhage is a distinctive, “on-target” adverse event associated with carotuximab, the higher frequency with which gingival hemorrhage was seen in this trial (92% vs. 31–50% [28, 29, 32]) alongside with gingival tenderness and mucositis may reflect a synergistic effect between everolimus and carotuximab. Similarly, a synergistic effect between carotuximab and everolimus may underpin the universal occurrence of fatigue (reported with both everolimus and carotuximab), and the high frequency of skin rash (typically associated with everolimus, reported in 20% of patients [12] vs. 75% in this study).
Toxicities associated with VEGF inhibition (hypertension, proteinuria, thrombosis) and serious infusion reactions were not seen. The dose of everolimus was reduced in 4 patients due to rash (n = 2, dose reduction within the first cycle) and fatigue (n = 1, dose reduction in cycle 4); in one patient the reason for dose reduction on cycle 6 day 15 was not reported. Carotuximab was dose reduced in 1 patient on cycle 4 day 15 due to gingival hemorrhage. Eleven patients (73%) completed protocol therapy. Reasons for discontinuation included (all, n = 1) intercurrent illness, noncompliance, progressive disease (all taken off study on cycle 2); and development of anasarca/generalized edema (taken off study on cycle 4; association with the investigational therapy could not be determined).
Pharmacokinetics.
Letrozole was rapidly absorbed from the gastrointestinal tract (maximum concentration (Cmax) reached within 2 hours). With repeated daily administration, Cmax and the area under the concentration-time curve (AUC) increased by 5.5 and 2.4-fold, respectively, consistent with accumulation[34] (supplementary table 1). The measured concentrations of letrozole (which reaches steady state in 2–6 weeks[35]) on cycle 2 day 1 were consistently higher and more variable as compared with the predicted concentrations indicating non-linear pharmacokinetics (Fig. 1B).
PK analyses of everolimus were consistent with significantly lower maximum and trough concentrations (Ctrough) from those previously reported (Cmax: 5 mg, 2.63 ng/mL; 10 mg, 5.38 ng/mL vs. 32 and 61 ng/mL, respectively[36]; Ctrough: 5 mg, 0.354 (estimated) ng/mL; 10 mg, 0.323 (estimated) ng/mL vs. 5.4 and 13.2 ng/mL, respectively[36], Fig. 1C and supplementary table 1). Similarly, the AUC’s at steady state were calculated to be at 1/10 of the AUC’s previously reported [36, 37]. The half-life of everolimus was similar in the 2 dose levels (5 mg: 13.4 hours, 10 mg: 11.8 hours), and significantly lower from the 30 hours previously reported[36]. There was a good concordance between the estimated and actual steady-state concentrations (supplementary Fig. 1A). No PK interactions between letrozole and everolimus are known to occur [37] indicating that carotuximab accelerated significantly the extravasation of everolimus (estimated volume of distribution exceeded 4000 and 5000 liters for 5 and 10 mg of everolimus, respectively) as well as its clearance to the point that predose everolimus concentrations dropped below the lower limit of quantitation.
Serum concentrations of carotuximab known to saturate CD105 receptors (200 ng/mL) were rapidly achieved and consistently maintained (supplementary Fig. 1B). With dose escalation from 3 to 15 mg/kg, AUC and half-life increased supraproportionally while clearance decreased by a similar magnitude probably due to target saturation (supplementary table 2). PK simulations using a two-compartment model suggest accumulation of carotuximab in the extravascular volume of distribution (Fig. 1D).
Pharmacodynamics and Antitumor Activity.
All patients underwent definitive surgery with curative intent. One patient with a basal-like intrinsic molecular subtype (see below) progressed clinically and radiographically and she transitioned to preoperative chemotherapy. Downstaging from stage 2/3 to 0/1 was achieved in 4/17 tumors (23.5%; 95% CI, 6.81–49.9), stage did not change in 8/17 (47.1%; 95% CI, 23–72.2), and 4/17 tumors were upstaged (23.5%; 95% CI, 6.81–49.9) (Fig. 2A). Of the 4 tumors that were upstaged, 3 were invasive lobular and 1 invasive mammary carcinomas. Given the diffuse growth pattern and lack of desmoplastic reaction in invasive lobular carcinoma, this finding may reflect underestimation of the extent of the original disease rather than interval tumor growth[10]. Representative radiographic responses are shown in Fig. 2B and supplementary Fig. 2.
All patients had residual disease on surgical pathology; pCR was achieved in one of two sites in a patient with multifocal disease. To capture treatment effect, we also assessed pathologic response by means of Residual Cancer Burden (RCB) scoring [38] (Fig. 2C). Among the remaining tumors, moderate treatment effect (RCB-II) was seen in 14 (14/16, 87.5%; 95% CI, 61.65–98.45) while residual disease was extensive in 2 (RCB-III, 12.5%; 95% CI, 1.55–38.35). Semiquantitative assessment of the residual tumors for stromal tumor infiltrating lymphocytes (TILs) was consistent with no or minimal levels of TILs in 15/17 tumors (patient who progressed and transitioned to chemotherapy excluded; 88.2%; 95% CI, 63.6–98.5) [39, 40].
To characterize gene expression changes induced by the investigational therapy, we performed genomic analyses on a NanoString nCounter platform (Breast Cancer 360 panel) of paired macrodissected tumor samples obtained at the time of diagnosis and on cycle 2 day 1. We successfully tested and analyzed 10 paired samples; one on-treatment sample did not pass quality control.
By PAM50 Molecular Subtype Signature, at diagnosis, 6 patients had luminal B, 3 patients had luminal A, and 1 patient had basal-like intrinsic subtype (Fig. 2D, supplementary Figs. 3 and 4). Of the 6 patients with luminal B breast cancer at diagnosis, 5 converted to luminal A on cycle 2 day 1 yielding a “molecular downstaging” rate of 83.3% [41]. Excluding the patient with the basal-like intrinsic subtype who had a distinctive gene expression profile (Fig. 3A and supplementary Fig. 5A), differential gene expression analyses between paired pre- and on-treatment samples show significant downregulation in the expression of genes involved in or regulating cell division (39/53 genes significantly downregulated at a p adjusted for multiple comparisons by Bergamini-Hochberg < 0.001; enrichment score p 8.89 10− 23/False Discovery Rate 2.73 10− 22) (Fig. 3B and C). The expression of the proliferation marker Ki-67 also decreased with the investigational therapy and its changes aligned with intrinsic subtype changes or lack thereof (Fig. 3D). The second most overrepresented functional category of downregulated genes, included genes involved in DNA repair most notably BRCA1 (enrichment score p 2.08 10− 7/False Discovery Rate 3.21 10− 6). A possible explanation for the upregulation and attendant treatment-related downregulation of these genes, may rely on fact that the ER induces transcriptional stress via R-loop formation, i.e. 3-stranded nucleic acid structures comprising a DNA:RNA duplex and a displaced single-stranded DNA[42]. Persistent R-loop accumulation can compromise genomic integrity which in turn, is mitigated by the BRCA1/2 genes[43, 44].
The presence of non-luminal intrinsic subtypes in the immunophenotypically HR+/Her2- subgroup has been well recognized [45, 46]. By comparison to the luminal tumors, overexpression of genes that typify a basal-like subtype (cytokeratin 5 [47, 48], BCL11A and FOXC1 [49, 50]) or are associated with resistance to endocrine therapies (STAT1 [51], cyclin E [52]), and the lower expression of ESR1 and PGR may underpin the absence of gene expression changes and clinical response to the investigational therapy (supplementary Fig. 5A and B).
CD105 (endoglin) was predominantly expressed in the vessels (Fig. 3E). CD105 expression, as assessed by immunohistochemistry, did not change significantly with the investigational therapy. The lack of change in CD105 expression may be due to early on-treatment CD105 evaluation because decreases in signal enhancement between baseline and cycle 4 day 1 breast MRIs were observed (Fig. 2B and supplementary Fig. 2). The absence of change may also be consistent with the mechanism of action of carotuximab whereby the antibody accelerates the cleavage and shedding of endoglin by coupling it with the membrane-anchored matrix-metalloproteinase 14 without changing the expression of the endoglin itself[53].
The protocol left postoperative treatment decisions at the discretion of the treating physicians. None of 4 patients whose tumors downstaged with the investigational therapy received adjuvant chemotherapy; 4/7 (57%) and 2/3 (66%) patients whose tumors did not change stage or were upstaged, respectively, received adjuvant chemotherapy. With a median follow up of 45.6 months after surgery for curative intent (53.4 months from cycle 1 day 1), 3 patients relapsed (Fig. 3F, supplementary Fig. 6). The patient with the basal-like intrinsic subtype experienced an early recurrence despite neoadjuvant chemotherapy. Among the 5 patients whose tumors converted from luminal B to luminal A, 2 received adjuvant chemotherapy. None of the remaining 3 patients experienced a recurrence (follow up, 24.1–50.4 months; median, 40.1).