Safety and efficacy of sargramostim (GM-CSF) in the treatment of Alzheimer's disease

Abstract

Introduction: Inflammatory markers have long been observed in the brain, cerebrospinal fluid (CSF), and plasma of Alzheimer's disease (AD) patients, suggesting that inflammation contributes to AD and might be a therapeutic target. However, non-steroidal anti-inflammatory drug trials in AD and mild cognitive impairment (MCI) failed to show benefit. Our previous work seeking to understand why people with the inflammatory disease rheumatoid arthritis are protected from AD found that short-term treatment of transgenic AD mice with the pro-inflammatory cytokine granulocyte-macrophage colony-stimulating factor (GM-CSF) led to an increase in activated microglia, a 50% reduction in amyloid load, an increase in synaptic area, and improvement in spatial memory to normal. These results called into question the consensus view that inflammation is solely detrimental in AD. Here, we tested our hypothesis that modulation of the innate immune system might similarly be used to treat AD in humans by investigating the ability of GM-CSF/sargramostim to safely ameliorate AD symptoms/pathology.

Methods: A randomized, double-blind, placebo-controlled trial was conducted in mild-to-moderate AD participants (NCT01409915). Treatments (20 participants/group) occurred 5 days/week for 3 weeks plus two follow-up (FU) visits (FU1 at 45 days and FU2 at 90 days) with neurological, neuropsychological, blood biomarker, and imaging assessments.

Results: Sargramostim treatment expectedly changed innate immune system markers, with no drug-related serious adverse events or amyloid-related imaging abnormalities. At end of treatment (EOT), the Mini-Mental State Examination score of the sargramostim group increased compared to baseline (P = .0074) and compared to placebo (P = .0370); the treatment effect persisted at FU1 (P = .0272). Plasma markers of amyloid beta (Aβ40 [decreased in AD]) increased 10% (P = .0105); plasma markers of neurodegeneration (total tau and UCH-L1) decreased 24% (P = .0174) and 42% (P = .0019), respectively, after sargramostim treatment compared to placebo.

Discussion: The innate immune system is a viable target for therapeutic intervention in AD. An extended treatment trial testing the long-term safety and efficacy of GM-CSF/sargramostim in AD is warranted.

Keywords: Alzheimer's disease; Mini‐Mental State Examination; Pittsburgh compound B positron emission tomography; activities of daily living; amyloid; cytokine; glial fibrillary acidic protein; granulocytemacrophage colony stimulating factor; granulocyte‐macrophage colony‐stimulating factor; innate immune system; interleukin‐6; lymphocyte; monocyte; neuroinflammation; neutrophil; sargramostim; tau; tumor necrosis factor‐alpha; ubiquitin C‐terminal hydrolase L1.

FIGURE 1

Increases in markers of the innate immune system in sargramostim‐treated participants compared to placebo‐treated participants. Complete blood counts (CBC) with differential were used to determine the effects of granulocyte‐macrophage colony‐stimulating factor (GM‐CSF)/sargramostim treatment on cells of the innate immune system. Absolute numbers of monocytes (A), lymphocytes (B), and neutrophils (C) were all statistically significantly increased during the treatment phase (15 injections over 3 weeks) in the sargramostim group compared to the placebo group (P = .0005, P = .0512, and P < .0001, respectively) at the end of treatment (EOT) (see Tables S2‐4). The shorter half‐life of neutrophils is revealed by the fact that when a weekend intervened after an injection of GM‐CSF/sargramostim, the absolute neutrophil counts dropped, but then increased again during active treatment. D, The Meso‐Scale Discovery method was used to determine changes in plasma inflammatory cytokines with sargramostim treatment. At EOT compared to baseline, sargramostim treatment leads to statistically significant increases in interleukin (IL)‐2 (P = .0022), IL‐6 (P = .0154), IL‐10 (P = .0003), and tumor necrosis factor alpha (TNF‐α; P < .0001), and a decrease in IL‐8 (P = .0052). Shown are natural log transformed values. There were no significant changes from baseline in any inflammatory markers at either follow‐up 1 or follow‐up 2 visits (data not shown). E, The ratio of plasma albumin to globulin is often used to assess inflammation and its acute phase response. As expected, sargramostim stimulated a drop in the albumin/globulin ratio compared to placebo (P = .0029 at EOT). Graphs plotted as means +/– standard error of the mean.

FIGURE 2

Improvement in Mini‐Mental State Examination (MMSE) scores in sargramostim‐treated participants compared to placebo‐treated participants. A, Mixed model analysis of MMSE data from the 20 sargramostim‐treated participants and 20 placebo‐treated participants shows a statistically significant improvement in the sargramostim group at the end of treatment (EOT; A) compared to baseline (1.45, 95% confidence interval [CI]: [0.44, 2.46], P = .0074) and compared to the placebo group (1.80, 95% CI: [0.12, 3.49], P = .0370), and at the first follow‐up visit (FU1; B) at 45 days after the EOT, compared to the placebo‐treated group (1.75, 95% CI: [0.21, 3.29], P = .0272). Shown are the results +/– standard error of the mean [SEM], setting the baselines at 0. See Table S6 for full statistical analyses. B, The improvement in total MMSE score occurred in: (a) 14 of 20 (70%) sargramostim‐treated participants, and 7 of 20 (35%) placebo‐treated participants (P = .0267) at the EOT, (b) 12 of 20 (60%) sargramostim‐treated participants and 4 of 20 (20%) placebo‐treated participants (P = .0098) at the 45‐day follow‐up visit (FU1), and (c) 11 of 20 (55%) sargramostim‐treated participants and 5 of 20 (25%) placebo‐treated participants (P = .0528; non‐significant trend) at the 90‐day follow‐up visit (FU2). (* P < .05; ** P < .01) Combining the results to identify an overall treatment effect shows that 16 of 20 (80%) sargramostim‐treated participants can be considered “responders” in that they showed a higher MMSE score compared to baseline at either the EOT or at the 45‐day follow‐up visit, compared to only 7 of 20 (35%) placebo‐treated participants (P = .0040). Difference in proportion by treatment group was tested with Chi‐square/Fisher's exact association test (see Table S6 for full statistical analyses). C, To assess the effect of sargramostim treatment on the participants throughout the study, the 20 sargramostim‐treated participants and 20 placebo‐treated participants were each divided into 10,000 random subsamples of 10 participants each, and for each simulated data set, a model was run, generating estimates for the means and contrasts. For > 98% of the simulated data sets, the estimate for change from baseline to the EOT within the sargramostim group, and the estimates for the treatment effects on change from baseline to EOT and change from baseline to FU1 were greater than zero. The mean MMSE scores were calculated and graphed as a delta distribution. The distribution was approximately Gaussian and the mean and median delta was 1.8, as was the delta of the entire 20 sargramostim‐treated participants and 20 placebo‐treated participants analyzed in (A) and (B), showing that the statistically significant difference between the sargramostim‐treated participants and the placebo‐treated participants at the EOT reflected all parts of the trial. Because we reported interim study results at symposia in 2018, we tested for a potential subsequent bias and found no significant difference between the mean total MMSE scores of participants enrolled during the first and second halves of the trial. Graphs plotted as means +/– SEM. (* P < .05; ** P < .01)

FIGURE 3

Changes in Alzheimer's Disease Assessment Scale Cognitive subscale (ADAS‐Cog13) score and its memory domain subscale in sargramostim‐treated participants compared to placebo‐treated participants. A, There were no significant changes in ADAS‐Cog13 scores between sargramostim‐treated participants and placebo‐treated participants during the treatment phase (through the end of treatment [EOT] ≈day 19). Note that the scores shown on the Y‐axis have been inverted so as to mirror Figure 1 and show improved cognition in the upward direction. However, at follow‐up 1 (FU1), 45 days after the EOT, the ADAS‐Cog13 showed a statistically significant increase (worsening) in the sargramostim group compared to baseline (4.46, 95% confidence interval [CI]: [2.11, 6.82], P = .0009), and compared to the placebo group (4.33, 95% CI: [0.90, 7.76], P = .0147) and then returned to the level of the placebo group at the 90‐day follow‐up visit. The treatment effect was stronger when baseline ADAS was adjusted for (5.54, 95% CI: [2.31, 8.78], P = .0013), and there was a statistically significant effect of baseline on expected change score at the 45‐day follow‐up (–0.173 per scale unit, 95% CI: [–0.298, –0.049], P = .0077). B, To compare to the memory‐predominant Mini‐Mental State Examination (MMSE) measure, the memory domain subscale of ADAS‐Cog13 (ADAS delayed word recall + ADAS word recognition + ADAS orientation + ADAS word recall avg) was analyzed. At the EOT, which is when the MMSE showed a statistically significant improvement (Figure 2), the ADAS‐Cog13 Memory Subscale showed a statistically trending improvement from baseline in the sargramostim group compared to placebo (–1.84, 95% CI: [–3.82, 0.11], P = .0632). Unlike the MMSE, however, the ADAS‐COG13 memory subscale showed no statistically significant improvement within the sargramostim group between baseline and the EOT. The treatment effect was attributable to the worsening of the placebo group. Nor was there a treatment effect for the ADAS‐Cog13 Memory Subscore at the 45‐day follow up. Although there was no statistically significant baseline effect on change scores for the EOT, thus justifying the uncorrected cell means models, ¹⁸ the statistical trend for the treatment effect of the ADAS‐Cog13 memory subscale at the EOT disappeared when baseline was adjusted for. Shown are the results +/– standard error of the mean, with the baselines set to 0. See Tables S7 and S8 for full statistical analyses.

FIGURE 4

Sargramostim treatment reduces biomarkers of neuropathology compared to placebo‐treated participants. Plasma biomarkers of Alzheimer's disease amyloid and tau pathology and neurodegeneration were assessed using the SIMOA platform (Quanterix). A, Amyloid beta (Aβ)40 data from N3PA plates are plotted (+/– standard error of the mean [SEM]) and show a statistically significant increase in the sargramostim group at the end of treatment (A), compared to baseline (P = .0127) and compared to the change in placebo from baseline (P = .0105). B, Assessed using N4PB plates, total tau levels decreased 17% compared to baseline (P = .0327) and 24% (P =.0174) compared to the change from baseline in the placebo group (A). C, At the end of treatment, another measure of neurodegeneration, plasma ubiquitin C‐terminal hydrolase L1 (UCH‐L1), decreased 40% compared to baseline (P = .0017) and 42% compared to the change from baseline in the placebo group (P = .0019). Shown are the results +/– SEM, with the baselines set to 0. See Tables S12‐14 for full statistical analyses

FIGURE 5

Correlations among and between changes in behavior and biomarkers of Alzheimer's disease. A, In addition to the effect of sargramostim treatment on both Mini‐Mental State Examination (MMSE) and immune system cells shown in Figure 1, we assessed the correlations between these measures (Table S21). The Pearson correlation between change in MMSE and change in absolute neutrophil counts was statistically significant at the end of treatment (EOT; 0.409; P = .0098). B, The correlation between change in MMSE and change in absolute lymphocyte counts was also statistically significant at the EOT (0.353; P = .0276). These results show that the improvement in cognition measured by MMSE is correlated with (and possibly caused by) the increase in innate immune system stimulation and its downstream effects. C, There was a very strong correlation between changes in plasma glial fibrillary acidic protein, a measure of astrocyte activation, and plasma neurofilament light, a measure of neuronal damage, from baseline at all time points for all participants (Pearson coefficient = 0.752, 0.693, 0.663 at the EOT, follow‐up 1 [FU1], follow‐up 2 [FU2]; all P < .0001; Panel C shows the correlation for the EOT), indicating a likely mechanistic link between neuronal damage and brain glial inflammation in AD. D, E, Within the sargramostim‐treated group, changes in MMSE were positively correlated with changes in Activities of Daily Living at the EOT (Pearson coefficient = 0.476; P = .034, Panel D) and at FU1 (0.656; P = .017, Panel E), the time points that showed a statistically significant treatment effect of sargramostim on MMSE.