Granulocyte Colony–Stimulating Factor in Patients With Acute Ischemic Stroke
Results of the AX200 for Ischemic Stroke Trial
Background and Purpose—Granulocyte colony–stimulating factor (G-CSF; AX200; Filgrastim) is a stroke drug candidate with excellent preclinical evidence for efficacy. A previous phase IIa dose–escalation study suggested potential efficacy in humans. The present large phase IIb trial was powered to detect clinical efficacy in acute ischemic stroke patients.
Methods—G-CSF (135 µg/kg body weight intravenous over 72 hours) was tested against placebo in 328 patients in a multinational, multicenter, randomized, and placebo-controlled trial (NCT00927836; www.clinicaltrial.gov). Main inclusion criteria were ≤9-hour time window after stroke onset, infarct localization in the middle cerebral artery territory, baseline National Institutes of Health Stroke Scale score range of 6 to 22, and baseline diffusion-weighted imaging lesion size ≥15 mL. Primary and secondary end points were the modified Rankin scale score and the National Institutes of Health Stroke Scale score at day 90, respectively. Data were analyzed using a prespecified model that adjusted for age, National Institutes of Health Stroke Scale score at baseline, and initial infarct volume (diffusion-weighted imaging).
Results—G-CSF treatment failed to meet the primary and secondary end points of the trial. For additional end points such as mortality, Barthel index, or infarct size at day 30, G-CSF did not show efficacy either. There was, however, a trend for reduced infarct growth in the G-CSF group. G-CSF showed the expected peripheral pharmacokinetic and pharmacodynamic profiles, with a strong increase in leukocytes and monocytes. In parallel, the cytokine profile showed a significant decrease of interleukin-1.
Conclusions—G-CSF, a novel and promising drug candidate with a comprehensive preclinical and clinical package, did not provide any significant benefit with respect to either clinical outcome or imaging biomarkers.
Granulocyte colony–stimulating factor (G-CSF) is a growth factor already in clinical use for the treatment of chemotherapy-associated neutropenia.1
During the past 9 years, a vast body of evidence has demonstrated that G-CSF is a potent neuronal growth factor with multimodal antiapoptotic, arteriogenic, and neurogenic properties.2–4 The large amount of animal data from different laboratories and their meta-analyses indicated G-CSF as one of the most promising drug candidates for stroke treatment.5–7 In particular, G-CSF had been shown to improve long-term functional recovery after stroke even if treatment was delayed for days.8 A phase IIa clinical trial in 43 acute ischemic stroke patients demonstrated feasibility and safety of intravenous G-CSF therapy in acute stroke9 and allowed identification of an optimal dose of 135 µg/kg body weight over 72 hours for this trial. The aim of the AX200 for Ischemic Stroke (AXIS) 2 trial was to show clinical efficacy of this dose for the treatment of acute ischemic stroke.
Materials and Methods
Study Type and Randomization
AXIS 2 was a European, multicenter, placebo-controlled, randomized, and double-blind trial. The study was performed according to the International Conference of Harmonization Good Clinical Practice and was approved by the respective regulatory authorities (first approval received from the lead ethics committee of Linz, Austria on March 16, 2009; first approval for a whole country was received for Austria on March 31, 2009), and by the local ethics committees of all study sites. Informed consent of patients was required before they were entered into the study. Design and content of the consent form were according to country regulations and approved by the lead and local ethics committees. Patients with acute ischemic stroke were randomly assigned to either placebo or verum medication by an interactive Web response system.
Drug Dosing and Application
A cumulative dose of 135 µg/kg body weight recombinant human G-CSF (Filgrastim; AX200; SYGNIS, Germany) over 72 hours was compared with identical-looking placebo (saline). One-third of the total cumulative dose was administered as an intravenous bolus (30-minute infusion), whereas the remainder of the dose was administered as continuous intravenous infusion at a steady rate over 72 hours. For infusion, the study drug was diluted with 5% dextrose.
Study Design: Inclusion and Exclusion Criteria
The main inclusion criteria were initial National Institutes of Health Stroke Scale (NIHSS) score from 6 to 22, age 18 to 85 years, a time window of ≤9 hours after onset of stroke symptoms, and stroke localization in the middle cerebral artery territory. We chose a time window of 9 hours on the basis of a consensus among steering board members considering the presumed mode of action of G-CSF (both neuroprotective and neuroregenerative properties). Subgroups on the basis of time of inclusion and regression analyses with the factor inclusion time had been predefined to detect possible time window–dependent effects of G-CSF treatment. MRI was obligate before inclusion. A minimum diffusion-weighted imaging (DWI) lesion size of 15 mL was targeted, which was implemented by the following rule: on the slice showing the largest extension of the infarct, the largest diameter of the lesion should be ≥3 cm and should be visible on ≥3 consecutive slices (slice thickness 5 mm). If this rule could not be applied (eg, because of an irregular shape or partially separate volumes on the slices), then the patient could be included in the clinical study if the investigator provided written justification that the patient fulfilled this inclusion criterion. Treatment with recombinant tissue plasminogen activator (rt-PA) was allowed whenever patients fulfilled all of these criteria after having received thrombolysis. Main exclusion criteria were signs of a very severe stroke on imaging (carotid T occlusion, more than two-thirds of the media territory, signs of midline shift), hemorrhagic, and lacunar strokes.
The initial protocol was amended after 21 patients already had been included to improve recruitment. According to the amendment, the NIHSS entry limit was lowered from 8 to 6, and patients with previous rt-PA treatment were allowed to participate as described.
Study Logistics and Conduction
The study was conducted at 78 European centers, with 51 centers actively recruiting. Participating countries were Austria, Belgium, Czech Republic, Germany, Poland, Slovakia, Spain, and Sweden. Details of participating centers and investigators are given in Table I in the online-only Data Supplement. Recruitment started in August 2009, and the total study duration was 24 months. Physicians who examined patients for outcome parameters were trained and certified for modified Rankin scale (mRS) and NIHSS evaluation. The raters were not allowed to have access to patients’ laboratory records to ensure blinding. NIHSS score was obtained at baseline (day 1) and on each of the next 3 days (days 2–4); 90 days after inclusion, the NIHSS, Barthel index, and mRS scores were determined. Routine laboratory tests, differential blood counts, cytokines, and G-CSF levels were obtained at baseline and then daily for the first 4 days. All analyses were performed at a central laboratory (Clearstone Central Laboratories, Baillet, France). An independent data safety monitoring board continuously monitored serious adverse events. In addition, the data safety monitoring board held meetings after 25%, 50%, and 75% of patients had been recruited for review of adverse events, safety laboratory values, and concomitant medication.
MRI data were obtained at baseline (fluid attenuation inversion recovery [FLAIR], T2, T2*, DWI, time-of-flight magnetic resonance angiography, perfusion-weighted imaging [PWI]) and at day 30 (FLAIR, T2, T2*). MRI readings were performed using a proprietary semiautomatic reading system (BioClinica, Lyon, France) by 3 independent and blinded readers. DWI and FLAIR volumes were semiautomatically delineated using segmentation and editing tools. PWI volumes were semiautomatically derived from relative mean transit time PWI maps by including all pixels with a mean transit time >2 seconds compared with a healthy contralateral region. Mismatch was defined by both direct subtraction of total PWI and DWI volumes and by voxel-based difference.
The primary end point was a difference of the mRS score at day 90 between G-CSF and placebo-treated patients; the secondary end point was a difference in NIHSS score at day 90 between G-CSF and placebo-treated patients. Both main analyses for the primary (mRS) and secondary (NIHSS) end points as well as power calculations for both variables were based on a linear model. We chose this analysis type for these ordinal-type variables because of (1) the ease of understanding and interpreting a mean point difference on these scales in contrast to odds ratios or other measures, (2) the robustness of t-based statistics for ordinal scales,10,11 and (3) the generally increased power of this approach. Analysis of the VISTA database has shown largely equivalent results for linear and ordinal logistic regression analyses of the mRS.12 In-depth discussions of the permissibility of this approach are available.13–15 To guard against any misinterpretation of results, we also have predefined more conservative ordinal logistic regression analyses for the full and dichotomized mRS scales (0–1/2–6 and 0–2/3–6 cuts).
The trial was powered to detect a point difference on the mRS of ≤0.45 with 80% probability on the basis of clinical expert opinions that a 0.5-point difference on the mRS would be a clinically meaningful difference in stroke outcome. Analyses of end points were adjusted for age, NIHSS at baseline, and initial DWI volume.12 This statistical analysis model increased power of effect detection by including DWI volume compared with previously used models and resulted in a targeted sample size of 328 patients. Additional exploratory analyses were prespecified for Barthel index, infarct growth, adverse events, mortality, cytokines, and hematology.
Data Handling and Statistics
Missing values for clinical outcome scales were imputed by last observation carried forward when necessary. Before unblinding, a blind data review was conducted to define handling of protocol violations and composition of the study sets (safety set, full analysis set according to International Conference of Harmonization guideline E9 [statistical principles for clinical trials], as well as per protocol set). The primary efficacy variable mRS at score day 90 was modeled using parametric least squares with the treatment group as the main independent variable. Age, NIHSS at baseline, log10-transformed diffusion volume deficit at baseline, and the treatment with rt-PA were included in the model as covariates. Additionally, the interaction between treatment and rt-PA was included. Additional analyses were performed with logistic regression with the mRS as the ordinal scale. Statistical analyses were performed using SAS 9.1 and JMP 9.0.1 (SAS Institute, Heidelberg, Germany). P<0.05 (2-sided if applicable) was considered significant.
Study Sets and Baseline Characteristics
A total of 328 patients were randomized into the study (Figure 1). The safety set comprised 324 patients because 4 patients did not receive any study medication and were excluded from the analysis. The full analysis set, after the intention-to-treat principle, contained 323 patients because 1 patient had a hemorrhagic instead of an ischemic infarct as determined by imaging. The per protocol set contained 272 patients; 51 patients were excluded from the per protocol set, of whom 10 patients had solely lacunar infarcts, 17 patients had intracranial hemorrhage in addition to ischemic lesion at baseline, and 9 patients had ischemic infarcts outside the middle cerebral artery territory (6 anterior cerebral artery, 3 posterior cerebral artery; these deviations were identified by central MRI evaluation). Furthermore, there were 6 patients with no primary end point available (eg, lost to follow-up), 4 patients who had received <50% of study medication, 3 patients with erroneous randomization or treatment not according to randomization, 1 patient with no MRI at baseline, and 1 patient with stroke onset >10 hours. Our goal to include only patients with initial DWI volumes of >15 mL was not reached because only 208 of 272 patients (76%) fulfilled this criterion by DWI image analysis. The rule described previously, therefore, proved largely insufficient as a means of estimating DWI volume before randomization.
Demographic and stroke-specific baseline characteristics were similarly distributed between both arms, with slightly larger DWI lesions and higher NIHSS scores at baseline for the active arm (Table 1). Patients in the active arm had a greater mean PWI/DWI mismatch.
During the first 4 days, G-CSF levels followed the profile known from the AXIS trial9 (Figure 2A). Peak levels at day 1 returned to normal values at day 4. The corresponding hematologic dynamics were also in the expected range, with a similar increase in white blood cells in the G-CSF group mainly caused by neutrophils (Figure 2B). Monocytes also were elevated as expected (Figure 2C). As noted in the previous AXIS trial,6 platelet counts decreased slightly in the G-CSF arm relative to placebo (P<0.05) but remained within the normal range (Figure 2D).
Concerning the serum levels of various cytokines, such as interferon-γ, IL-1, IL-2, IL-6, IL-8, IL-10, IL-12, and tumor necrosis factor-α, G-CSF treatment significantly lowered IL-1 levels, whereas the other cytokines remained stable (Figure 3).
Other blood chemistry parameters monitored, such as electrolytes (Na, K, Ca), activated partial thromboplastin time, international normalized ratio, creatinine, or creatine kinase, are shown in Table II in the online-only Data Supplement. An unexpected finding in the G-CSF–treated group was a lowering of cholesterol levels of both high-density and low-density lipoproteins (G-CSF total cholesterol day 4 [mean±SD]: 3.4±0.8 mmol/L from baseline 5.0±1.1 mmol/L; placebo day 4: 4.4±1.0 mmol/L from baseline 4.9±1.3 mmol/L; P<0.05). A lowering of potassium levels in the G-CSF treatment arm also was remarkable (G-CSF on day 4: 3.8±0.4 mmol/L from baseline 4.3±0.5 mmol/L; placebo day 4: 4.1±0.5 mmol/L from baseline 4.3±0.5 mmol/L).
Vital Signs and ECG
Unexpectedly, hemodynamic parameters in the G-CSF treatment group were affected. Heart rate was significantly increased as compared with placebo (day 2; G-CSF arm: 82.6±1.3 bpm versus placebo arm 76.0±1.3 bpm; P<0.05) and mean arterial blood pressure (MAP) was lowered (day 2; G-CSF 95.1±1.1 versus placebo 98.3±1.1 mm Hg; P<0.05). Estimation of the product of total peripheral resistance times stroke volume from MAP and heart rate also showed a significant decrease during G-CSF treatment. There was also a small yet significant impact on pulse pressure (systolic and diastolic blood pressure; not shown).
Also, unexpectedly, in the treatment group, we observed a slight but significant overall increase of body temperature relative to placebo by 0.2°C for the mean temperature over 4 days (from 36.8±0.04°C versus 37.0±0.04°C [mean±SEM]) and an increase of 0.22°C for the maximum temperatures achieved during a 4-day observation period (37.19±0.05°C versus 37.41±0.05°C [mean±SEM]). Interestingly, blood pressure was directly related to the presence of G-CSF. Blood pressure normalized to placebo values at day 4 after the end of G-CSF infusions, whereas both heart rate and body temperature remained altered 1 day after the end of G-CSF therapy.
Abnormal findings in ECG were equally distributed between treatment groups, with only 1 new clinically relevant ECG finding reported for a patient in the placebo group.
The total incidence of adverse events was similar in the G-CSF and placebo groups (ie, 98% versus 95% of patients). The majority of adverse events in both treatment groups were judged to be mild or moderate. The most frequently reported adverse events were infections (in 56% versus 53% of patients receiving G-CSF or placebo, respectively), constipation (in 22% versus 24%; G-CSF versus placebo), hypokalemia (28% versus 15%; G-CSF versus placebo), and fever (32% versus 20%; G-CSF versus placebo). Fever was significantly more often diagnosed in G-CSF–treated patients (P=0.0163; Fisher exact test), as was hypokalemia (P=0.0068; Fisher exact test). The highest imbalance of fever was reached on day 2 (9% versus 4%; G-CSF versus placebo).
Serious adverse events occurred at a similar incidence in both treatment groups (39% of patients in both groups; Table 2). Nervous system disorders were the most frequently reported serious adverse events in both groups (22% versus 15%; G-CSF versus placebo), followed by infections (8% and 9%; G-CSF versus placebo) and cardiac disorders (both 6%). The majority of serious adverse events in both groups were judged to be not related or unlikely related to the study medication.
In total, 4 suspected unexpected serious adverse reactions were reported to the competent authorities and relevant ethic committees in accordance with regulatory requirements. One serious adverse event in the placebo group (pyrexia) and 3 serious adverse events in the G-CSF group (brain edema and 2 events of ischemic stroke) were judged to be likely related to the study medication by the investigators. In the end, none of these serious adverse events was judged as definitely related to study medication.
In total, 65 patients died within the observation period of 90 days: 35 patients (21.7%) in the G-CSF group, and 30 patients (18.4%) in the placebo group (P=0.4; Fisher exact test). Most frequent causes of death included bronchopneumonia, cerebrovascular accident, and brain edema. All serious adverse events leading to death were not related or unlikely related to study medication, with the exception of 1 event of brain edema in the G-CSF group judged to be likely related to this treatment by the investigator.
G-CSF had no significant effect on any of the prespecified primary and secondary outcome variables (Table 3). The mean mRS score at day 90 was 3.3 for the G-CSF treatment group and 3.1 for the placebo group (P=0.3). The secondary outcome parameter (difference in the NIHSS score at day 90 between G-CSF treatment and placebo) did not show any positive effect of treatment either (mean NIHSS scores were 8.9 for G-CSF and 8.5 for placebo; P=0.61). No significant differences were seen for other explanatory end points (Barthel index, mortality, dichotomized mRS, mRS analysis by logistic regression, and final lesion size at day 30).
To detect potential effects of previous thrombolysis with rt-PA on G-CSF treatment, a prespecified subgroup analysis was performed (Table III in the online-only Data Supplement). Patients who received thrombolysis had better mRS scores than those not treated with rt-PA, but there was no difference between the subgroups with regard to the efficacy of G-CSF (no rt-PA/G-CSF, 3.53 [95% CI, 3.13–3.94]; no rt-PA/placebo, 3.23 [95% CI, 2.82–3.63]; P=0.29; rt-PA/G-CSF, 3.05 [95% CI, 2.76–3.34]; rt-PA/placebo, 3.07 [95% CI, 2.78 – 3.37]; P=0.92). Likewise, no significant difference in the NIHSS score (day 90) with regard to G-CSF treatment could be detected in the rt-PA–treated and nontreated subgroups.
Final infarct volume (MRI FLAIR) was not significantly different between treatment groups (G-CSF: 59.0±6.9 mL, placebo: 65.9±6.6 mL; Table 3). However, there was a considerable initial disadvantage for G-CSF because of a larger perfusion deficit and a larger mismatch area at baseline. Considering the smaller total and cortical final infarct volumes, one could speculate on an efficacy signal on lesion growth. An exploratory analysis of the final infarct volume considering PWI deficit, treatment group, and rt-PA cotreatment revealed significant interactions of PWI and treatment group (P<0.01), suggesting that G-CSF may have an inhibitory effect on infarct growth in patients with large PWI deficits.
From the preclinical point of view, G-CSF was a promising candidate for the therapy of acute ischemic stroke. The drug showed clinical safety in 2 pilot trials14,15 and showed hints of possible clinical efficacy (AXIS 2a study);6 this led to a phase IIb trial testing 135 µg/kg G-CSF against placebo in patients with medium and large ischemic infarcts in the middle cerebral artery territory ≤9 hours after symptom onset. Unfortunately, the promising preclinical data could not be translated into clinical efficacy in a larger patient cohort. Treatment with G-CSF over 3 days did not show any improvement in clinical outcome after 3 months. Slight imbalances in baseline parameters were widely adjusted for in the present analysis model and cannot explain the absence of efficacy.
The reasons for the failure of G-CSF remain obscure. The overall study quality and conduct were good, with the only exception being a deviation from the intended DWI lesion size on inclusion. Analysis of pharmacokinetics and pharmacodynamics of G-CSF showed exactly the expected effects on the hematopoietic system, thus proving adequate adherence to dosing protocols.
Inclusion of rt-PA–pretreated patients could have been potentially problematic, because one could assume that the effect of G-CSF might get diluted because of the efficacy of rt-PA. This set of patients was, however, required to fulfill all inclusion criteria after rt-PA treatment as well, and separate analyses of rt-PA–treated and nontreated patients revealed no obvious difference in the effects of G-CSF on clinical outcome.
Potential other reasons for our inability to detect any treatment effect may lie in a relatively large number of centers (n=51) needed to recruit the patients in the study, and this would likely increase residual variation and not obscure differences between treatment groups. The time window (9 hours) may be considered too long for an initial proof of concept trial, but prespecified subanalyses at shorter inclusion times have not detected any signs of efficacy at earlier times. Unwanted unblinding in the study because of leukocytosis was guarded against by the separation of treating and evaluating physician and, in any case, would have worked in favor of a beneficial effect of G-CSF. Finally, in view of the absence of any major difference in outcome parameters means or medians, it is very unlikely that increasing the study population further would have yielded any result different from the one obtained.
The dose chosen in this trial was based on exploratory analyses of the previous AXIS trial and was the second highest dose explored in that trial. Selection of this dose level was based on the observation that no further increase in exploratory efficacy occurred beyond that. It is possible that this was not the optimal dose selected, and it is also possible that a longer duration of the treatment could have been beneficial. However, the total lack of treatment response makes it questionable whether a comparably small change in treatment dose or timing would have produced a therapeutically relevant effect.
Several unexpected effects of G-CSF on physiological parameters were identified in the trial. First, G-CSF caused a significant lowering of serum potassium concentrations and a doubling of the incidence of hypokalemia. This did not, however, result in any cardiac disorders in G-CSF–treated patients. Second, G-CSF led to a decrease in mean MAP and an increase in heart rate. The decrease in MAP is most likely caused by a direct acute effect of G-CSF on the vasculature with a lowering of peripheral resistance. Effects of G-CSF on the arteriogenesis and angiogenesis have been described in animal models, although direct relaxation of the vascular tone has not yet been reported.8 This effect is likely mediated by stimulation of the G-CSF receptor on endothelial cells. We presume that the increase in heart rate is indirectly caused by a combination of MAP lowering and temperature increase. Third, there was a significant increase in mean body temperature by 0.2°C and an increased incidence of reported fever.
Despite their small effect size, several of these systemic effects may have counteracted potential beneficial effects of the drug. These effects had not been seen before in trials using subcutaneous administration14,15 and with the standard application of the drug in neutropenia (≈10 µg/kg body weight per day subcutaneous). They are possibly related to the intravascular delivery of the protein as well as the high dose used in this trial.
The data regarding the efficacy of G-CSF in rodents are probably the most convincing of any stroke drug candidate that has reached the clinic so far. The negative outcome of this trial deepens concerns that rodent stroke models may be questionable for predicting stroke drug efficacy in humans.11
G-CSF does not improve stroke outcome after 90 days when applied as 3-day intravenous therapy, despite promising preclinical and clinical data. This result strengthens the notion of a principal problem in translating findings from the animal laboratory to clinical stroke patients whose basis is currently not understood.
AXIS2 Study Organization
Data safety monitoring board: Gerlinde Egerer (Chair), Gerd Mikus, Volker Limmroth, Meinhard Kieser.
Monitoring and data collection: FGK, Munich Germany.
Image acquisition procedures and analysis: BioClinica, Lyon (Luc Bracoud).
MRI reading group: Jochen B. Fiebach, Marc Hermier, Juliette Bouffard.
Study medication: Theorem, Bad Soden Germany (Rüdiger Weber).
Participating centers and investigators: See Table I in the online-only Data Supplement.
The authors gratefully acknowledge all patients who agreed to be recruited and screened for this study. The authors thank all participating staff of the hospitals for their invaluable help in performing the study.
Drs Schneider, Vogt, Kollmar, Laage, Schwab, and Schäbitz are inventors of patent applications claiming the use of granulocyte colony–stimulating factor for the treatment of stroke. Drs Schneider, Vogt, Kollmar, Rathgeb, and G. Charissé are or were employees of SYGNIS Bioscience. The study was funded by SYGNIS Bioscience. Drs Ringelstein, Thijs, Norrving, Chamorro, Aichner, Grond, and Saver received fees and expenses from SYGNIS for their steering board work.
Guest Editor for this article was James Grotta, MD.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.113.001531/-/DC1.
Participating centers and investigators: See Table I in the online-only Data Supplement.
- Received March 20, 2013.
- Revision received June 21, 2013.
- Accepted June 25, 2013.
- © 2013 American Heart Association, Inc.
- Rader M
- Minnerup J,
- Heidrich J,
- Wellmann J,
- Rogalewski A,
- Schneider A,
- Schäbitz WR
- Philip M,
- Benatar M,
- Fisher M,
- Savitz SI
- Schäbitz WR,
- Laage R,
- Vogt G,
- Koch W,
- Kollmar R,
- Schwab S,
- et al
- Vogt G,
- Laage R,
- Shuaib A,
- Schneider A