Blood measure of neuronal death is exponentially higher with age, especially in females, and halted in Alzheimer’s disease by GM-CSF treatment

Summary

Aging increases the risk of neurodegeneration, cognitive decline, and Alzheimer’s disease (AD). We report that plasma concentrations of ubiquitin C-terminal hydrolase-L1 (UCH-L1) and neurofilament light (NfL) become exponentially higher from ages 2 to 85 in cross-sectional samples, serving as neuronal death/damage biomarkers across the lifespan. UCH-L1 concentrations rise faster in females, who exhibit increased AD risk. Glial fibrillary acidic protein (GFAP) concentrations increase exponentially after age 40, especially in females. Age-adjusted UCH-L1, NfL, and GFAP plasma concentrations are greatly elevated in mildly cognitively impaired participants. Treatment of human AD trial participants with granulocyte-macrophage colony-stimulating factor (GM-CSF/sargramostim) apparently halts neuronal cell death: UCH-L1 biomarker concentrations are reduced to those of 5-year-old healthy controls. GM-CSF treatment also reduces neuronal apoptosis and astrogliosis in a rat model of AD. An exponential increase in neurodegeneration with age, accelerated by astrogliosis/inflammation, may underlie the contribution of aging to cognitive decline and AD and can be halted by GM-CSF/sargramostim treatment.

Graphical abstract

Highlights

• •Plasma measures of neuronal death/damage (UCH-L1, NfL) rise exponentially from age 2
• •A plasma measure of brain inflammation (GFAP) rises exponentially from age 40
• •Women exhibit faster exponential rises in plasma UCH-L1 and GFAP than men
• •Promising Alzheimer drug GM-CSF greatly reduces neuronal death in patients with AD, AD rats

Sillau, Coughlan, Ahmed, and colleagues (University of Colorado Anschutz) found that cross-sectional neuronal death/damage (UCH-L1/NfL) plasma measures rise exponentially from age 2. Brain inflammation (plasma GFAP) rises exponentially from age 40, especially in females. A 3-week Alzheimer’s trial with cytokine GM-CSF improves cognition (MMSE) and neuropathology (plasma Aβ and Tau) and reduces neuronal death (plasma UCH-L1).

Introduction

Increasing age is the greatest risk factor for “natural” age-associated cognitive decline (AACD) and, especially in females, for developing Alzheimer’s disease (AD), but the mechanisms underlying these connections are unknown.^{1,2,3,4,5,6,7,8,9,10} Neuronal loss (apoptosis) and brain atrophy accompany AD and are associated with cognitive deficits.^{11,12,13,14,15} Neuronal apoptosis also increases with age in mice and humans, although gray matter atrophy is less clearly associated with normal aging.^{15,16,17,18,19} Increased inflammation is also correlated with aging and AD pathogenesis (termed “inflammaging”), but whether inflammation causes or is a response to neurodegeneration, or both, is also unknown.^{1,8,20,21,22,23} For example, early studies of the inflammatory disease rheumatoid arthritis (RA) found that patients exhibited a reduced risk of developing AD, which was attributed to their use of non-steroidal anti-inflammatory drugs (NSAIDs).²⁴ However, clinical trials showed no NSAID benefit to participants with either AD or mild cognitive impairment (MCI), and studies of other inflammatory diseases, such as periodontitis, detected an increase in AD risk.^21,24,25 Evidently, the role of inflammation in aging and neurodegenerative disease is complex.

The identification of blood biomarkers of brain damage is essential for advancing the development of therapies for AD and AACD.^{22,26,27,28,29} Here, we first studied whether ubiquitin C-terminal hydrolase-L1 (UCH-L1) and neurofilament light (NfL) are potential biomarkers of neurodegeneration during aging, building on numerous reports that their plasma concentrations are well correlated with neuronal damage in neurodegenerative disease, including AD, traumatic brain injury (TBI), and white matter disorders.^{30,31,32,33,34,35,36,37,38,39,40,41,42} In Creutzfeldt-Jacob disease, plasma UCH-L1 and NfL concentrations are greatly increased, and higher levels of UCH-L1 predict faster decline.⁴³ In Parkinson’s disease (PD), plasma UCH-L1 and NfL levels are increased and correlate with cognitive decline.^44,45

Although UCH-L1 was initially discovered and named as a ubiquitin C-terminal hydrolase, it is likely that this is not its primary function, as cells and animals lacking UCH-L1 are not defective in the ubiquitin-proteosome system.^30,46 Indeed, UCH-L1 is primarily a brain protein, making up 1%–5% of total protein in neurons, with some expression in endocrine tissue. Indeed, only brain damage and disease result in increased blood UCH-L1 concentrations, while diabetes results in no change in blood UCH-L1, nor does hypertension, cerebrovascular disease, dyslipidemia, or depression.^30,36 Suggestions for the neuronal functions of UCH-L1 include the regulation of energy metabolism and mitochondrial fusion, antioxidant activity, synaptic activity, and tau phosphorylation.^31,46,47,48 Indeed, in the human AD brain and in AD mouse model brains, UCH-L1 protein levels are reduced due to neuronal loss/damage.^31,49

NfL is a neuronal protein and an essential component of axons. Traumatic or disease-associated damage to axon tracks leads to the release of NfL into the cerebrospinal fluid (CSF) and plasma, where it can be detected as a biomarker of axonal damage in numerous neurodegenerative diseases, TBI, and aging.^30,39,40,41 Glial fibrillary acidic protein (GFAP) is upregulated in activated astrocytes, and its increased concentration in CSF or plasma is a marker/proxy of reactive gliosis/inflammation in aging, neurodegenerative disease, and TBI.^22,35,38,50

Here, we assessed cross-sectional plasma concentrations of UCH-L1 and NfL across the lifespan and found that these measures of neuronal loss and damage become exponentially higher starting from childhood. Plasma concentrations of GFAP become exponentially higher starting at age 40. Finally, we studied the effect on plasma concentrations of UCH-L1 when treating patients with AD with granulocyte-macrophage colony-stimulating factor (GM-CSF)/CSF2/sargramostim, a long-approved drug for stimulating immune stem cells, which we previously found to improve a measure of cognition and plasma biomarkers of neurodegeneration in a double-blind, randomized, placebo-controlled, phase 2 clinical trial.⁵¹ GM-CSF treatment reduces plasma UCH-L1 concentrations in participants with AD to the very low levels normally observed in early childhood. GM-CSF treatment also reduces the high levels of neuronal apoptosis and astrogliosis in the hippocampi of aged TgF344-AD rats, a model of AD.

Results

Plasma concentrations of UCH-L1 are exponentially higher with advancing age from early childhood

Plasma concentrations of UCH-L1, a measure of neuronal cell loss, were assessed cross-sectionally in 317 healthy control participants between ages 2 and 85 from three observational studies (see STAR Methods; Tables S5 and S6). The assessments were determined using the very sensitive Quanterix SIMOA platform, which was also used previously to assess plasma biomarkers in the sargramostim/GM-CSF AD trial (see discussion in STAR Methods; Table S7).⁵¹ As shown in Figure 1A, plasma UCH-L1 concentrations are exponentially higher across the entire age spectrum, from an estimated 6.22 pg/mL at age 2 to approximately 15.56 pg/mL at age 85 (estimated change per year = 1.110%, p = 5.504 × 10⁻⁸). Graphical inspection of the data and comparison to a spline fit (Figures S1–S3) indicate that a log-linear relationship of UCH-L1 concentrations with age is an excellent fit (Pearson correlation estimate [replicates log averaged = 0.30]), which is equivalent to an exponential relationship on the original scale (Figures 1A and 1B) (estimated change per year = 1.110%, 95% confidence interval [CI]: [0.716%, 1.505%], p = 5.504 × 10⁻⁸). Interestingly, most of the age-associated increase in plasma concentrations of UCH-L1 occurs in females (Figure 1C; estimated female change per year = 1.448% [95% CI: (0.942%, 1.957%); p = 3.635 × 10⁻⁸]; estimated male change per year = 0.582% [95% CI: (−0.036%, 1.203%), p = 0.0650]; and estimate of the gender ratio of ratios = 0.99146, [95% CI: (0.98362, 0.99936), p = 0.0342]). The log plots in Figure 1D and the spline analyses (Figure S1) support the conclusion that the plasma concentration of UCH-L1 increases exponentially with age in males and females. Although the rate of exponential increase in males is slower, there is no evidence against the functional form (log-linear) we chose for regression. Our exponential curves are a linear regression fit on the logarithmic transform of UCH-L1. For the age range of 40–83 years, the area under the curve (AUC) average for the expected UCH-L1 values in males was statistically significantly less than the expected values in females (ratio estimate = 0.792, 95% CI: [0.628, 0.997], p = 0.0472). Because the age effects in the model are linear in the log plot, the AUC average is equivalent to comparing the expected value for males to that for females at the midpoint of the range, 61.5 years (see also Figure S1; Tables S7 and S8).

Figure 1.

Plasma UCH-L1 concentrations are higher with advancing age in healthy control participants, especially in females

Concentrations of UCH-L1 in plasma from 314 healthy control participants, spanning ages 2–85 from the Crnic Institute Human Trisome Project (n = 103), the CUACC Bio-AD longitudinal observational study (n = 69), or the MS healthy controls biomarker study (n = 145) (3 participants lacked usable UCH-L1 data) were compared to the age of the participant. The assessment of the plasma concentration of each biological sample was replicated 2–3 times, and all such replicates were used in the analysis; see STAR Methods.

(A) Association between absolute UCH-L1 concentrations and age and the point-wise standard errors.

(B and C) UCH-L1 log plot with point-wise standard errors. Plasma UCH-L1 levels are exponentially higher with age across the lifespan (estimated change per year = 1.110%, p = 5.504 × 10⁻⁸). The effect of gender on the curve is shown in (C), which indicates that most of the promotion effect of age on plasma UCH-L1 concentrations is driven by females (estimated female change per year = 1.448% [p = 3.635 × 10⁻⁸] compared to estimated male change per year = 0.582% [p = 0.0650]; gender difference p = 0.0342).

(D) The effect of gender on the UCH-L1 log plot with point-wise standard errors. For the age range of 40–83 years, the area under the curve (AUC) average for the expected UCH-L1 values in males was statistically significantly less than the expected values in females (p value = 0.0472).

See also Figure S1 and Tables S7 and S8.

Plasma concentrations of NfL are exponentially higher with advancing age from early childhood

Plasma concentrations of NfL are commonly used to assess neuron/axon damage and the risk of future cognitive decline in AD, MCI, and other neurodegenerative diseases and have been found to increase with age.^{28,39,41,52,53} Therefore, we examined the effects of age and gender on plasma concentrations of NfL in the 317 healthy control participants. As shown in Figures 2A and 2B and Table S7, plasma concentrations of NfL were exponentially higher with increasing age in healthy control participants (p < 2.220 × 10⁻¹⁶), with the slope of the log-transformed curve being greater than that observed for UCH-L1 with increasing age, such that the estimated change per year = 2.469% per year (95% CI: [2.225%, 2.714%], p < 2.220 × 10⁻¹⁶). Of note, the 95% CIs for the NfL and UCH-L1 rates of increase are non-overlapping, suggesting a statistically significant difference (alpha = 0.05) between the two. The exponential curve of NfL plasma concentrations with age showed a trend of being steeper for males than for females (ratio of ratios = 1.00487, 95% CI: [1.00000, 1.00976], p = 0.0502), with an estimated difference per year of 2.775% (95% CI: [2.388%, 3.164%], p < 2.220 × 10⁻¹⁶) for males compared to an estimated difference per year of 2.277% (95% CI: [1.965%, 2.590%], p < 2.220 × 10⁻¹⁶) for females (Figure 2C). Spline fits show that the log-transformed data are linear (Figure S1).

Figure 2.

Plasma concentrations of NfL are exponentially higher with advancing age in healthy control participants

(A and B) Concentrations of NfL in plasma samples from 317 healthy control participants spanning ages 2–85 were compared to the age and gender of the donor. Plasma concentrations of NfL are exponentially higher with age across the entire lifespan. The assessment of each biological sample was replicated 2–3 times, and all such replicates were used in the analysis; see STAR Methods. The associations between absolute NfL concentrations and age and the point-wise standard errors are shown in (A), and the log plot is shown in (B) (estimated change = 2.469% per year [p < 2.220 × 10⁻¹⁶]).

(C) Comparison of males and females: estimated change per year was 2.775% (p < 2.220 × 10⁻¹⁶) in males compared to 2.277% per year (p < 2.220 × 10⁻¹⁶) in females. The exponential rate of change determined cross-sectionally was marginally statistically non-significantly greater in males (2.775% per year) than in females (2.277% per year) (p = 0.0502). When comparing the log-linear fit to splines for the association of NfL with age and gender in healthy controls (replicates averaged), the graph suggests that the rate of increase accelerates somewhat with older age, but a log-linear relationship is still an excellent approximation for all healthy controls (Figure S1; Tables S7 and S9) and when males and females are examined separately (Figure S1).

Plasma concentrations of GFAP are exponentially higher from age 40

The log-linear relationship between plasma concentration and age found for UCH-L1 and NfL was not appropriate for describing the age-associated changes in plasma GFAP in the 317 healthy control participants. Specifically, spline modeling of the plasma levels of GFAP (replicates averaged) passes the deviance test for comparing it to the null hypothesis of linear (p < 2.2 × 10⁻¹⁶). GFAP levels are lower across the age range from ages 2 to 25, stay constant, and rise exponentially from approximately age 40 (Figure 3A; Table S8). Relatively high levels of plasma GFAP in children have been reported previously.⁴¹ The U-shaped spline plot is apparent for both genders (Figures 3B and 3C). To quantify the age slopes, we approximated the relationships between log GFAP and age with a piece-wise linear model, with a knot at age 30 (Figure 3D). The estimated change per year for females less than 30 years old = −3.156% (95% CI: [−4.474%, −1.820%], p < 5.908 × 10⁻⁶), the estimated change per year for females greater than or equal to 30 years old = 3.110% (95% CI: [2.630%, 3.592%], p < 2.220 × 10⁻¹⁶), the estimated change per year for males less than 30 years old = −3.368% (95% CI: [−4.748%, −1.967%], p = 4.107 × 10⁻⁶), and the estimated change per year for males greater than or equal to 30 years old = 2.899% (95% CI: [2.332%, 3.469%], p = 2.220 × 10⁻¹⁶). Within gender, the pre- and post-age 30 age slopes differed significantly (p = 1.571 × 10⁻¹² and 1.798 × 10⁻¹¹ for females and males, respectively). Within age groups, the differences in gender-age slope were not significant. The GFAP AUCs were higher in females than males overall from ages 2 to 83 (ratio = 0.843, 95% CI: [0.752, 0.946], p = 0.0037) and from ages 30 to 83 (ratio = 0.819, 95% CI: [0.717, 0.935], p = 0.0033) (see also Table S8). These findings indicate that plasma markers of astrogliosis/neuroinflammation (as measured by GFAP) temporally follow and thus are likely a reaction to the earlier and ongoing age-associated neurodegeneration, with females exhibiting higher plasma markers of astrogliosis and inflammation at all ages.

Figure 3.

Plasma concentrations of GFAP are exponentially higher with age in healthy control participants after age 40

Plasma concentrations of GFAP in 317 healthy control participants spanning ages 2–85 were compared to the age and gender of the donor.

(A) Comparison of the log-linear fit to splines for the biomarker association with age in the healthy controls (replicates averaged). The deviance test for comparing the spline model to the null hypothesis of linear: p < 2.2 × 10⁻¹⁶, with the graph suggesting a U-shape, with GFAP concentrations accelerating after age 40. A spline plot by gender in the healthy controls (replicates averaged) also suggests a U-shape, with GFAP concentrations accelerating significantly after age 40 (Tables S7 and S10). The assessment of each biological sample was replicated 2–3 times, and all such replicates were used in the analysis; see STAR Methods.

(B–D) Deviation from linearity is evidenced in the absolute spline plot (C). To quantitate the relationships between log GFAP and age, a piece-wise linear model was constructed with a knot at age 30 (D) for both genders (estimated female change per year for less than 30 years old = −3.156% [p < 5.908 × 10⁻⁶]; estimated female change per year for greater than or equal to 30 years old = 3.110% [p < 2.220 × 10⁻¹⁶]; estimated male change per year for less than 30 years old = −3.368% [p = 4.107 × 10⁻⁶]; estimated male change per year for greater than or equal to 30 years old = 2.899% [p = 2.220 × 10⁻¹⁶]). Within gender, the pre- and post-age 30 age slopes differed significantly (p = 1.571 × 10⁻¹² and 1.798 × 10⁻¹¹ for females and males, respectively). Within age groups, the differences in gender-age slope were not significant. The GFAP AUCs were higher in females than males overall from ages 2 to 83 (p = 0.0037) and from ages 30 to 83 (p = 0.0033) (see also Table S10).

Correlation analyses of plasma measures of age-associated brain degeneration

The plasma measures of neurodegeneration (UCH-L1 and NfL) and astrogliosis (GFAP) for all healthy control participants are highly correlated with each other (UCH-L1:NfL, UCH-L1:GFAP, and NfL:GFAP, p < 10⁻¹¹) (Figure S2). This finding provides some evidence that UCH-L1 is indeed a proxy marker of neuronal death/damage, like the established marker NfL, and that brain aging across the lifespan reflects potentially parallel increases in neurodegeneration, both cellular and axonal, and astrogliosis/inflammation.

Comparisons of plasma measures of age-associated brain neurodegeneration and inflammation in MCI due to AD, mild-to-moderate AD, and healthy controls

AD dementia and its precursor, MCI due to AD, are both strongly associated with age and are accompanied by neurodegeneration and astrogliosis in the brain.²² Having established full age curves for plasma markers of neurodegeneration (UCH-L1 and NfL) and astrogliosis/inflammation (GFAP), we compared these to the age-associated concentrations of NfL, GFAP, and UCH-L1 in plasma samples from the 32 participants with MCI due to AD from the Bio-AD study²² and in plasma samples from the 36 participants with mild-to-moderate AD from our sargramostim/GM-CSF AD trial at baseline (before any placebo or GM-CSF intervention)⁵¹ (Figure S4). A diagnosis of MCI or mild-to-moderate AD was associated with higher overall levels of both NfL (Figure S4A) and GFAP (Figure S4B) compared to age-matched healthy control (HC) participants at 73.7 years, which is the mean age for the MCI participants (MCI/HC estimate age of 73.7: NfL, ratio estimate = 1.497, p = 5.923 × 10⁻⁶ and GFAP, ratio estimate = 1.798, p = 7.816 × 10⁻⁷; AD/HC estimated age of 73.7: NfL, ratio estimate = 1.878, p = 6.728 × 10⁻⁷ and GFAP, ratio estimate = 1.391, p = 0.0018). Plasma concentrations of NfL were marginally non-significantly higher in participants with mild-to-moderate AD than in participants with MCI at 73.7 years, while GFAP concentrations were significantly lower (AD/MCI estimate age of 73.7: NfL, ratio estimate = 1.254 and p = 0.0680 and GFAP, ratio estimate = 0.774 and p = 0.0464), as expected from previous studies.^22,39 Interestingly, the plasma concentration of NfL showed age-associated higher levels in the participants with mild-to-moderate AD (2.714% per year, p = 0.0122) (Figure S4A), whereas GFAP plasma concentrations did not show age-associated higher levels in the participants with mild-to-moderate AD (−0.030% per year, p = 0.9761) (Figure S4B). Plasma concentrations of both NfL (Figure S4A) and GFAP (Figure S4B) showed age-associated higher levels for MCI (NfL: 4.138% per year, p = 0.0017; GFAP: 3.466% per year, p = 0.0398). There were no statistically significant differences among the age slopes for healthy controls, patients with AD, and patients with MCI for NfL. The age effect in GFAP for healthy controls greater than 30 years old was statistically significantly different than that for patients with mild-to-moderate AD (p = 0.0056).

UCH-L1 concentrations in plasma from the participants with MCI due to AD are higher overall than in healthy control participants at the mean age for the participants with MCI, at 73.7 years (ratio estimate = 1.559, p = 0.0001), whereas the plasma UCH-L1 concentrations of the participants with mild-to-moderate AD at the baseline visit (before any placebo or GM-CSF intervention) are approximately the same as those of the healthy control participants (ratio estimate = 1.042, p = 0.7072) (Figure S4C). UCH-L1 concentrations in MCI were higher than in mild-to-moderate AD at 73.7 years (ratio estimate = 1.497, p = 0.0007). UCH-L1 plasma concentrations increased for participants with mild-to-moderate AD by an estimated 2.420% per year of age (p = 0.0054), while the estimated age-associated increase for MCI is marginally statistically non-significant (estimate = 2.499% per year, p = 0.0941), possibly because it is underpowered with the modest sample size. There were no statistically significant differences among the age slopes for healthy controls, patients with mild-to-moderate AD, and patients with MCI for UCH-L1 (interaction test p = 0.2040).

An exploratory receiver operating characteristic (ROC) analysis was carried out (Table S11) to determine the discriminatory capacity of a model combining UCH-L1, NfL, and GFAP and age, compared to age alone. The AUCs for the full model were 0.8121 (MCI:HC), 0.8792 (MCI:AD), and 0.8622 (AD:HC), which are all higher than the AUC of age alone (p < 0.001).

Treatment of AD trial participants with sargramostim/GM-CSF lowers the UCH-L1 measure of neuronal loss to that of healthy controls many decades younger

In addition to identifying potential mechanisms of brain aging, age-associated biomarkers may be used to assess the efficacy of interventions that may slow or halt the aging process. We have previously discovered GM-CSF/sargramostim as a potential treatment for AD, whose likely mechanisms of action may include targeting the aging process in the brain. Specifically, based on early studies showing that patients with RA had a reduced risk of developing AD, we hypothesized that this protection might be due to a physiological reaction against RA’s associated inflammation, with the beneficial side effect of reducing the risk of AD, which exhibits brain inflammation. We tested our hypothesis and found that treatment of a mouse model of AD with GM-CSF, an immune-system-stimulating/modulating cytokine that stimulates the proliferation of phagocytes in both the bone marrow and the brain and is upregulated in the plasma of patients with RA, reduced brain amyloid levels by half, increased brain synaptophysin levels, and restored memory to normal after a few weeks of subcutaneous administration.⁵⁴ GM-CSF treatment also improves the impaired cognition and reduced neuronal function in aged wild-type (WT) mice, indicating that the beneficial effects of GM-CSF on brain function are general and not exclusively derived from targeting AD pathology.^54,55 These findings in AD and aged mice have been confirmed by others.^56,57

Building on this foundation, we completed a phase 2, double-blind, randomized, placebo-controlled trial of human recombinant GM-CSF (sargramostim) (250 μg/m²/day subcutaneous injection, 5 days/week for 3 weeks) in participants with mild-to-moderate AD.⁵¹ Treatment with sargramostim led to improved scores in the mini-mental state examination (MMSE) by almost two points (compared to baseline and to placebo) and moved the concentrations of AD-associated plasma biomarkers—Aβ40 and total tau—toward normal.⁵¹ Interestingly, the largest change in a measure of AD neuropathology at the end of treatment was in the plasma UCH-L1 concentrations, which had decreased in the sargramostim-treated group by 40% compared to baseline (p = 0.0017) and by 42% compared to placebo (p = 0.0019).⁵¹ Now, we can compare these results to the age curves for UCH-L1 in the healthy control participants shown in Figure 1 to determine how effective sargramostim/GM-CSF treatment was in reducing this measure of neuronal loss.

The plasma concentrations of UCH-L1 in 18 participants with mild-to-moderate AD at baseline and at the end of treatment with sargramostim/GM-CSF are plotted together with the data from healthy control participants in Figure 4A and show that the absolute values of plasma UCH-L1 are greatly reduced after sargramostim/GM-CSF treatment (ratio estimate = 0.497, p = 0.0008). Indeed, sargramostim/GM-CSF treatment reduces the concentrations of UCH-L1 in the plasma of trial participants to an average level far below those of similarly aged healthy control participants (ratio estimate = 0.502, p = 0.0019) and equivalent to those found in healthy control participants six decades younger (Figure 4A). For example, a 67.8-year-old participant with mild-to-moderate AD treated with sargramostim/GM-CSF, corresponding to the mean age for this group, would have had a plasma UCH-L1 concentration of 6.47 pg/mL (geometric mean, 95% CI: [4.45, 9.39], not baseline calibrated—sargramostim/GM-CSF baseline only), which is equivalent to that expected of a 5.4-year-old healthy control participant, based on our cross-sectional data in Figure 1. The spaghetti plot of each participant from baseline to the end of treatment is shown in Figure 4B. The lack of any effect of placebo treatment is shown in Figures 4C and 4D.

Figure 4.

Treatment of participants with mild-to-moderate AD with sargramostim/GM-CSF reduces plasma UCH-L1 measure of neuronal loss to levels equivalent to those of healthy controls many decades younger

(A) Association plots and point-wise standard errors are shown for the 18 participants with mild-to-moderate AD from the phase 2, double-blind, randomized, placebo-controlled trial with recombinant human GM-CSF (the sargramostim/GM-CSF AD trial) at baseline before they were treated with sargramostim/GM-CSF (AD untreated—GM-CSF baseline, n = 18) or after they were treated with sargramostim/GM-CSF (AD GM-CSF treatment, n = 18), together with the correlation curve for healthy control participants from Figure 1. The assessment of each biological sample was replicated 2–3 times, and all such replicates were used in the analysis; see STAR Methods. The comparison shows that treatment of participants with mild-to-moderate AD with sargramostim/GM-CSF leads to absolute UCH-L1 values that are greatly reduced compared to their starting (baseline) levels (ratio estimate = 0.497; p = 0.0008). Furthermore, sargramostim/GM-CSF treatment reduces plasma UCH-L1 concentrations to levels far below those expected of similarly aged healthy control participants (ratio estimate = 0.502, p = 0.0019 at 67.8 years, the mean age for the sargramostim/GM-CSF-treated patients with mild-to-moderate AD). The average plasma UCH-L1 level of the participants with AD equals the level expected of a 5.4-year-old healthy control participant (horizontal dotted line). The reduced plasma UCH-L1 levels associated with GM-CSF treatment of participants with AD are statistically significantly lower than those of all healthy control participants above approximately age 37.8 (vertical dot-dash line) but are statistically indistinguishable (p > 0.05) from healthy control participants younger than age 37.8.

(B) A spaghetti plot of plasma UCH-L1 measures before and after GM-CSF treatment of participants with AD.

(C) Plot of 18 baseline and placebo-treated participants with AD together with the correlation curve for healthy control participants from Figure 1.

(D) Spaghetti plot of plasma UCH-L1 measures before and after placebo treatment of participants with AD.

GM-CSF treatment reduces neuronal cell death/apoptosis in the CA1, CA3, and dentate gyrus/hilus regions of an AD rat model

To investigate the mechanism by which GM-CSF treatment reduces plasma concentrations of UCH-L1, a measure of neuronal death, we examined aged (18- to 20-month-old) TgF344-AD rats, a model of AD that shows the complete brain pathology of human AD (amyloid and tau deposition and neuronal loss).⁵⁸ TgF344-AD rats were treated with recombinant rat GM-CSF or saline placebo for 5 weeks, and their brains and those of age-matched WT control F344 rats were assessed by immunohistochemical staining for caspase-3, a marker of apoptosis that is increased in humans and animal models with AD and during aging.^50,59 As shown in Figure 5, aged TgF344-AD rats (treated with placebo/saline) exhibited large numbers of caspase-3-positive cells in the CA1, CA3, and dentate gyrus/hilus regions of the hippocampus compared to WT F344 rats. GM-CSF treatment of aged TgF344-AD rats significantly reduced the elevated number of caspase-3-positive cells compared to placebo-treated TgF344-AD rats, almost reaching the low number observed in the untreated WT F344 rats. Notably, the vast majority of caspase-3-positive cells in the TgF344-AD rats were neurons (90%–95%) based on co-staining for the MAP2 neuronal marker (see STAR Methods).

Figure 5.

Treatment with GM-CSF reduces neuronal cell death in the CA1, CA3, and dentate gyrus/hilus in a rat model of AD

Aged male TgF344-AD rats (18–20 months), treated with GM-CSF or placebo (saline) injection for 5 weeks, were assessed for neuronal cell death by immunohistochemical staining for caspase-3 (red) together with DAPI staining (blue), followed by blinded counting of caspase-3-positive cells. Arrowheads in the insets indicate caspase-3-positive cells. The numbers of caspase-3-positive cells were determined in the CA1 (A–C), CA3 (D–F), and dentate gyrus/hilus (G–I) regions of the hippocampus in age-matched F344 male wild-type (WT) rats (A, D, and G), TgF344-AD rats injected with saline for 5 weeks (B, E, and H), and TgF344-AD rats treated with GM-CSF for 5 weeks (C, F, and I). Quantitative analyses showed a significantly higher number of caspase-3-positive cells in all three hippocampal regions of TgF344-AD rats injected with saline compared to age-matched WT rats, which decreased significantly with GM-CSF treatment (p values are as indicated: J–L). For each bar, data are represented as mean ± SEM (p values are as indicated; see STAR Methods). Scale bar: 100 mm (20× magnification). All experiments were performed 2–6 times (technical replicates) with similar results (biological replicates: n = 5 for WT rats, n = 7 for Tg344-AD rats injected with saline, and n = 7 for Tg344-AD rats treated with GM-CSF).

GM-CSF treatment reverses astrogliosis in the CA1 and dentate gyrus/hilus of an AD rat model

In addition to being associated with normal human aging, MCI, and AD, as shown above, higher plasma concentrations of GFAP are also associated with lower measures of some cognitive attributes and with defects in white matter microstructure.²² Furthermore, astrogliosis is increased in the brain of the Dp16 mouse model of Down syndrome, which is reversed by GM-CSF treatment.⁵⁵ To determine the effect of GM-CSF treatment on astrogliosis in the AD brain, hippocampal sections of the aged male TgF344-AD rats treated with GM-CSF or placebo/saline injection for 5 weeks were assessed for patterns of activated astrocytes by immunohistochemical staining for GFAP together with DAPI co-staining. Quantitative analyses of the percentage of area of GFAP-positive astrocytes showed a significant increase in the CA1, CA3, and dentate gyrus/hilus regions of saline-treated TgF344-AD rats compared to age-matched WT rats. The percentage of area of GFAP staining was significantly reduced in the CA1 and dentate gyrus/hilus regions, but was not statistically significantly reduced in the CA3 region, in GM-CSF-treated TgF344-AD rats compared to saline-treated TgF344-AD rats (Figure 6).

Figure 6.

Treatment with GM-CSF reduces astrogliosis in a rat model of AD

Brain sections of the aged male TgF344-AD rats treated with GM-CSF or placebo (saline) injection for 5 weeks were also assessed for patterns of GFAP-positive astrocytes by immunohistochemical staining for GFAP together with DAPI staining, followed by blinded analyses of the expression patterns of the percentage of area stained for GFAP-positive astrocytes: the CA1 (A–C), CA3 (D–F), and dentate gyrus/hilus (G–I) regions of the hippocampus in age-matched F344 male wild-type (WT) rats (A, D, and G), TgF344-AD rats injected with saline for 5 weeks (B, E, and H), and TgF344-AD rats treated with GM-CSF for 5 weeks (C, F, and I). Quantitative analyses showed a significantly increased GFAP % area staining in the CA1, CA3, and dentate gyrus/hilus regions of saline-treated TgF344-AD rats compared to age-matched WT rats (p values are as indicated: J–L). In contrast, the GFAP % area staining was significantly reduced in the CA1 and dentate gyrus/hilus regions, but not statistically significantly in the CA3 region, in GM-CSF-treated TgF344-AD rats compared to saline-treated TgF344-AD rats (p values as indicated: J–L). For each bar, data are represented as the mean ± SEM for separate groups of mice. Scale bar: 200 mm (20× magnification). All experiments were repeated 2–6 times (technical replicates) with similar results (n = 5–8 rats/group; biological replicates).

See STAR Methods.

Discussion

The identification of biomarkers of brain aging is an area of active investigation. For example, blood biomarkers of aging in multiple organs, including the brain, have been reported recently, focusing on function rather than cell loss/damage, and a study in marmosets showed that serum UCH-L1, NfL, and GFAP concentrations are higher in late age compared to adult age.^26,60 However, the limited age range of the individuals in these previous studies did not yield insights into when brain aging begins, which we now report. Although a recent multi-omics aging study showed that many molecules rise exponentially from age 44,⁶¹ except for NfL,⁴¹ no previous studies^22,26,27,62 have shown an exponential age-associated elevation in plasma markers of brain damage across the life span.

Although postmortem analyses provide information about accumulated damage, the plasma concentrations of the biomarkers that we evaluated, UCH-L1, NfL, and GFAP, may be measures of current active damage in the brain, possibly of neuronal cell loss, axon damage, and astrogliosis, respectively. This hypothesis is supported by multiple studies of acute brain injury, which show that the half-lives of these proteins in the blood are short (∼24 h for NfL and GFAP but only 8.5 h for UCH-L1) and that brain neuronal loss and gliosis after TBI correlate with the plasma levels of UCH-L1 and GFAP.^33,63,64 Thus, the increases we observed cross-sectionally over many decades are unlikely to be due to accumulation and may instead serve as measures of processes currently underway in the brain that are enhanced with increased age. Interestingly, the numerical density of normal diploid neurons remained stable in a cross-sectional autopsy study of brains from cognitively normal individuals and individuals across various stages of the MCI-AD process. In contrast, the proportion of abnormal neurons destined for imminent apoptosis, as assessed by single-cell chromosome aneuploidy, is low in cognitively normal individuals, highest (∼30%) in preclinical and mild AD, and declines greatly in the late stages of AD when accumulated cell loss is highest⁶⁵ (for a discussion, see Potter et al.⁶⁶). Indeed, some 95% of the final neuron loss in AD brains can be calculated to be due to the loss of aneuploid neurons.⁶⁵ Similarly, although plasma levels of UCH-L1, NfL, and GFAP are increased in patients with early PD, most neuronal loss occurs before clinical motor or cognitive symptoms appear, and plasma UCH-L1 levels in participants with more advanced PD, especially in those with dementia, are indeed lower than in participants with early PD or even normal controls because, although the accumulated loss is high, the rate of loss has decreased.^44,45

An exponential increase in any product of a biological process implies the existence of at least one positive feedback loop that accelerates the process. The exponential increases with age of the cross-sectionally assessed plasma markers of the rates of brain degeneration, first of neuron loss (UCH-L1) and axon damage (NfL) and then of astrogliosis/neuroinflammation (GFAP), and their strong correlations with each other imply the existence of such a positive feedback loop in the process of brain aging. Although speculative, neuron loss and axon damage evident via plasma biomarkers from early childhood could induce gliosis/inflammation to phagocytose the resulting debris and initiate the inflammaging cascade, resulting in a vicious cycle of more neuronal damage and death and more inflammation.

Interestingly, the positive feedback loop involving neuronal loss/damage and inflammation in brain aging that the data imply includes the same components as an already-established positive feedback loop in the development of AD. Specifically, gliosis/neuroinflammation in AD increases the expression of cytokines that lead through several steps to increased production of Aβ peptides and their apoE-dependent polymerization into neurotoxic oligomers and filaments. The Aβ oligomers, in turn, further increase neuroinflammation through further activation of microglia and astrocytes.^{20,21,67,68,69} Positive feedback loops in pathogenic pathways are ideal targets for therapeutic intervention, as illustrated by repurposing GM-CSF for treatment of AD.

In sum, our findings allow several conclusions to be drawn.

• (1)The exponential increases in the concentrations of two plasma biomarkers of current neuronal damage, UCH-L1 and NfL, with increasing age in healthy control participants, indicate that normal brain aging is a lifelong process that becomes behaviorally apparent only later, as the accumulated neuronal damage overcomes neurogenesis, functional redundancy, and resiliency in some individuals but not all.⁷⁰
• (2)The higher average UCH-L1 concentrations in plasma in participants with MCI compared to healthy control participants provide potential support for an additional “N” measure for the AD amyloid/tau/neurodegeneration (A/T/N) assessment tool, which should be validated in future studies. Based on cross-sectional data, the finding that the elevations in the concentrations of the plasma biomarkers of neuronal damage, UCH-L1 and NfL, in healthy control participants occur earlier than the increase in astrogliosis (GFAP), which accelerates significantly after age 40, potentially indicates that astrogliosis and its attendant inflammation (inflammaging) are likely to be both a response to and an accelerator of the age-related neuronal damage in a positive feedback loop rather than an initiating cause. The higher age-associated plasma concentrations of UCH-L1 and GFAP in females should be further explored, as greater rates of neuronal cell loss and astrogliosis may explain or contribute to their increased risk for age-associated neurodegenerative diseases such as AD.
• (3)The standard errors of the mean for the concentration versus age curves for all three biomarkers are very tight across the lifespan, and together and separately, these biomarkers are able to identify individuals with MCI, thus potentially predicting future disease. However, there is also variance between individuals, which may reflect increased risk for or resilience to the development of AD, MCI, or AACD caused by genetic variants such as carrying an APOE4 allele,⁶⁸ lifestyle, environment, or other as-yet unidentified factors.¹⁰

In contrast to other potential treatments for AD, sargramostim/GM-CSF is also beneficial for preventing normal AACD in mouse models and for treating many neurological injuries and diseases that do not have associated AD pathology, for example, in animal models of Down syndrome, stroke, TBI, and PD^71,72,73,74; in humans with chemobrain⁷⁵; and in a preliminary clinical trial in participants with PD.⁷⁶ Its broad therapeutic applicability may be related to the ability of GM-CSF to cross the blood-brain barrier, to be both neuroprotective and anti-apoptotic, to stimulate arteriogenesis and blood flow, to promote axon preservation/regeneration and neuronal plasticity, and to induce the proliferation of neural stem cells^{74,77,78,79,80,81,82} (for more references and discussion, see Ahmed et al.²¹). In the TgF344-AD rat model of AD, GM-CSF treatment reduced both neuronal death and astrogliosis, based on caspase-3 and GFAP staining, respectively, in the brain after 5 weeks of treatment, whereas sargramostim/GM-CSF treatment of participants with AD in the phase 2 trial led to reduced plasma concentrations of UCH-L1 but not of NfL or GFAP, likely due to the very short treatment period of 3 weeks in the human trial and the short half-life of plasma UCH-L1 compared to NfL and GFAP.⁸³

The finding that 3 weeks of sargramostim/GM-CSF treatment of participants with mild-to-moderate AD in a phase 2, double-blind, randomized, placebo-controlled trial led to improved performance on a cognitive measure (almost 2 points on MMSE) and changes in plasma levels of markers of AD neuropathology (Aβ40, total tau, and UCH-L1), but no ARIA,⁵¹ together with the findings reported here, allow the following additional conclusions to be drawn.

• (4)A reduction in plasma UCH-L1 may be a sensitive biomarker for testing the efficacy of many AD treatments, including therapeutic changes in lifestyle.

Our finding that GM-CSF treatment reduced the elevated neuronal apoptosis and gliosis in the hippocampus in the TgF344-AD rat model of AD to levels that were close to those of age-matched WT control rats suggests that the ability of GM-CSF to improve cognition as measured by MMSE, to partially normalize the concentrations of amyloid and tau plasma biomarkers, and to greatly reduce the plasma concentrations of the biomarker of neurodegeneration, UCH-L1, in participants with AD⁵¹ is likely due to a reduction in the number of apoptotic neurons in the brain. Indeed, GM-CSF has been shown to suppress apoptosis, and we and others have found that apoptotic/damaged brain neurons in neurodegenerative diseases, including in AD, are often aneuploid and that aneuploidy leads to apoptosis, suggesting that GM-CSF may prevent this process.^{4,66,84,85,86,87,88,89,90,91} Furthermore, cells naturally undergo senescence with increased age, and their removal with “senolytic” molecules can reverse some features of aging, including improving cognition in animal models of AD.^67,92,93,94 Similarly, through its ability to stimulate the production and activity of innate immune phagocytes (i.e., microglia) in the brain,⁵⁴ GM-CSF treatment may enhance the removal of damaged, apoptotic, and senescent neurons, thus allowing the remaining neurons to function more effectively. In sum, our mechanistic studies in TgF344-AD rats are consistent with the following conclusion.

• (5)GM-CSF may have an anti-apoptotic and/or senolytic activity on neurons that may underlie its ability to prevent neuronal loss and reverse cognitive decline due to neurodegenerative disease or normal aging.

In participants with mild-to-moderate AD, the benefits of GM-CSF treatment in improving cognition and reducing plasma markers of neuropathology are evidently temporary, as markers of neuropathology return to almost pre-treatment levels by 45 days after sargramostim/GM-CSF treatment is halted, yet there is still a statistically significant benefit to cognition, as measured by MMSE, at 45 days post-treatment compared to the placebo.⁵¹ Treatment of healthy aged individuals with sargramostim/GM-CSF may reduce age-associated neuronal damage and reverse AACD most effectively with continuous application or may similarly confer a more long-lived benefit after treatment is halted.

Limitations of the study

• (1)Patients with MCI due to AD were identified by clinical diagnosis, not confirmed with CSF markers or positron emission tomography (PET). There were no cutoffs used. Thus, we could not categorize participants into A/T/N positivity. Although plasma AD biomarkers are increasingly used, future studies to confirm our findings would benefit from an analysis of CSF samples or from the use of mass spectrometry to analyze plasma biomarkers.
• (2)Combined data were obtained from cross-sectional/baseline assessments from different studies with different inclusion/exclusion criteria. Caution is warranted in interpreting the data as indicating longitudinal change in outcomes, as we cannot rule out cohort effects and resulting bias. Future studies should assess these biomarkers longitudinally across the lifespan and/or pool studies with harmonized recruitment/enrollment techniques to confirm the results.
• (3)High CVs in UCHL-1 measures may impact the clinical application of these findings, with cutoffs, as well as replication of results.

Resource availability

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Huntington Potter (huntington.potter@cuanschutz.edu).

Materials availability

No new materials have been generated in the project.

There are restrictions to the availability of human plasma samples due to the lack of an external centralized repository for their distribution and our need to maintain the stock. We are glad to share plasma samples with reasonable compensation by requestor for its processing and shipping.

All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

Data and code availability

• •All data are included in the manuscript and supplemental information and are available on Mendeley Data: https://data.mendeley.com/datasets/wrx5pbhr6z/1.
• •No custom code was developed for or used in any of the studies.
• •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

Acknowledgments

Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under award no. R01AG071151 (H.P.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work was supported by Department of Defense PRARP grant AZ160059 (H.P.), National Institutes of Health grant R01AI50305 (J.M.E.), the state of Colorado (H.P.), Alzheimer’s Association Part the Cloud grant PTC C-16-422172 (H.P.), the Global Down Syndrome Foundation (H.P. and J.M.E.), the Anna and John J. Sie Foundation (J.M.E.), numerous private donors (H.P.), the University of Colorado Human Immunology and Immunotherapy Initiative (HI3), and the University of Colorado Hospital.

Author contributions

Conceptualization, H.P., S.H.S., N.E., C.C., M.M.A., and H.J.C.; methodology, S.H.S., C.C., M.M.A., K.N., and B.M.B.; investigation, S.H.S., C.C., M.M.A., K.N., P.A., M.D.G., B.M.B., and J.M.E.; supervision, H.P. and J.M.E.; writing – original draft, H.P.; and writing – review & editing, H.J.C., S.H.S., C.C., M.M.A., B.M.B., H.P., and T.D.B.

Declaration of interests

H.P., T.D.B., C.C., M.M.A., and S.H.S. are inventors on several non-licensed US patents or pending applications owned by the University of South Florida or the University of Colorado and related to this research.

Declaration of generative AI and AI-assisted technologies in the writing process

No AI or AI-assisted technologies were used in the writing process..

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Chicken anti rat-MAP2	Phosphosolutions	Cat # 1100-MAPT; RRID: AB_2492141
Rabbit anti-GFAP antibody	Abcam	Cat #7260; RRID: AB_305808
Rabbit anti-Caspase-3 antibody	Cell Signaling Technology	Cat # 9662; RRID: AB_331439
Biological samples
Human plasma samples	This paper	Unpublished MS controls, published AD samples22,51 and controls from Human Trisome Project52
Chemicals, peptides, and recombinant proteins
Rat GM-CSF	Fuji film	Cat # 300-03
Critical commercial assays
SIMOA®, SR-X Analyzer	Quanterix	SIMOA®, SR-X Analyzer Neuro-4-Plex B
Experimental models: Organisms/strains
Transgenic rat model of AD	Terrence Town	TgF344-AD rat
Wild-type leukinelittermate rats	Terrence Town	F344 rat
Software and algorithms
SAS 9.4 and R	SAS® Academic Software	SAS 9.4 and R

Experimental model and study participant details

Participants: All participants signed informed consent forms approved for each study by the Colorado multi-institutional review board (COMIRB numbers below)

We analyzed healthy control plasma samples from three different studies. Healthy control participants (n = 317) who were part of the Crnic Institute Human Trisome Project (HTP, n = 103; age range: 2–61 years; 54% female), the University of Colorado Alzheimer’s and Cognition Center (CUACC) Bio-AD longitudinal observational study (n = 69; age range: 53–83; 70% female), or the multiple sclerosis (MS) healthy controls biomarker study (termed Nair) (n = 145; age range: 16–86; 64% female; 3 participants lacked usable UCH-L1 data). See demographics (including gender) and inclusion/exclusion criteria in Tables S4 and S5. Information on ancestry, race, and ethnicity were not collected unless as indicated in the demographics. The HTP is focused on studying biomarkers and clinical features of people with Down syndrome (DS) and includes typical control participants without DS, whose average biomarker values in three age groups for NfL, UCH-L1, and GFAP were published previously,⁵² and the CUACC Bio-AD study is focused on studying the effects of inflammation on the development of AD.²² The Nair MS biomarker study is investigating biomarkers associated with MS and includes healthy control participants. Together, these three healthy control cohorts span ages 2–85.

Using three community-dwelling healthy control cohorts that are diverse and heterogeneous with a wide range of ages adds confidence to the cross-sectional measures of the plasma biomarkers. If different populations had different levels and slopes of a marker with age, then pooling them would be averaging them and would cause a deviation away from linearity across the full age range, with parametric linear fit trying to smooth it out. However, there is no appreciable deviation from linearity for UCH-L1 or NfL when examined with the spline fits (Figure S1). GFAP measures across age are more complex, increasing exponentially only after age 40.

Participants assessed as having mild cognitive impairment (MCI) due to AD were part of the CUACC Bio-AD longitudinal observational study (n = 45) and were diagnosed based on an interdisciplinary consensus conference with review of cognitive testing, neurological examination, clinical dementia rating scale (CDR), and brain MRI. Participants with mild-to-moderate AD were from our published Phase II, double-blind, randomized, placebo-controlled trial⁵¹ with recombinant human GM-CSF (the “sargramostim/GM-CSF AD trial”). We include available plasma samples taken at the baseline visit prior to treatment with sargramostim/GM-CSF or placebo (n = 36) as the cohort of AD individuals, as well as available plasma samples taken at the end of three weeks of treatment with either sargramostim/GM-CSF (n = 18) or placebo (n = 18) to provide information on the effect of GM-CSF treatment on the plasma levels of UCH-L1. Some plasma samples were unavailable for measurements of NfL and GFAP concentrations.

Sex/gender influence

Gender was determined by self-report or physician report. Gender was associated with different results in the age-associated biomarker curves with females showing a higher and more rapid increase with age of UCH-L1 and GFAP.

Sample size and randomization

Samples from all available participants from the three observational (HTP controls, n = 103); the MS study controls (n = 145), the BioAD (n = 69). Participants assessed as having mild cognitive impairment (MCI) due to AD were part of the CUACC Bio-AD longitudinal observational study (n = 45), and the one interventional (GM-CSF/sargramostim) study (n = 43) were included in the analysis. For the interventional (GM-CSF/sargramostim) study, participants were assigned random ID numbers by a random number generator and assigned to either drug or placebo treatment randomly as described.⁵¹

Human subjects institutional approval and oversight

All human subjects research was approved by the Colorado Multiple Institutional Review Board (COMIRB). Specifically, we analyzed healthy control plasma samples from three different studies. Healthy control participants (n = 317) who were part of the Crnic Institute Human Trisome Project (HTP, n = 103; age range: 2–61 years; 54% female) (COMIRB #15–2170; NCT02864108; Dr. Joaquin Espinosa), the University of Colorado Alzheimer’s and Cognition Center (CUACC) Bio-AD longitudinal observational study (n = 69; age range: 53–83; 70% female) (COMIRB #15–1774; Dr. Brianne Bettcher), or the multiple sclerosis (MS) healthy controls biomarker study (termed Nair) (n = 145; age range: 16–86; 64% female; 3 participants lacked usable UCH-L1 data) (COMIRB #21–3703; Dr. Kavita Nair). The Colorado Multiple Institutional Review Board also approved the Pilot Phase 2 Trial of the Safety & Efficacy of GM-CSF (Leukine) in the Treatment of Alzheimer’s Disease (COMIRB # 12–1273; NCT01409915; Dr. Huntington Potter) from which plasma biomarker data on AD participants before and after treatment with GM-CSF/sargramostim are reported here. The biomarker data reported on the participants with mild cognitive impairment were part of the University of Colorado Alzheimer’s and Cognition Center BioAD project approved by the COMIRB (COMIRB #15–1774; Dr. Brianne Bettcher).

Animal model institutional permission and oversight

The TgF344-AD rat was developed as a model of AD by inserting transgenes that express the Swedish mutant human APP (APPsw) and mutant human presenilin 1 (PSEN1 delta E9) genes that cause familial AD.⁵⁸ As a result, and due to the fact that the rat MAPT (Tau) gene resembles the human version, the TgF344-AD rats overexpress human Aβ peptide and develop the full complement of human AD brain pathology: amyloid deposits, p-Tau-positive neurofibrillary tangles, and neuronal loss. Untreated age-matched wild-type (WT) F344 rats were used as controls. All rats were 18–20 months of age and only males were analyzed to reduce the effect of hormonal cycling.

All animal research was approved by the University of Colorado Institutional Animal Care and Use Committee (IACUC#00878).

Method details

Measurement of plasma biomarker concentrations

Concentrations of UCH-L1, GFAP, and NfL in plasma samples were assessed in healthy control participants and in participants with MCI due to AD using published methods,^22,51 specifically with the Quanterix single molecule array, or SIMOA, SR-X Analyzer system and the Neuro-4-Plex B kits. Concentrations of UCH-L1, GFAP, and NfL in the samples from mild-to-moderate AD participants in the sargramostim/GM-CSF AD trial were determined in our previously published manuscript using the same methods.⁵¹

The low pg sensitivity of the SIMOA platform is important for our studies. Plasma concentrations reported for UCH-L1 in the literature vary according to the exact population studied and the method of measurement used. For example, Papa and colleagues³⁴ used an in-house developed ELISA assay that shows ∼83 pg/mL for all controls, which includes normal and ‘trauma’ controls that are twice as high. A data table is missing in that manuscript, but the graph looks like 70 pg/mL for normal controls, which corresponds to hospitalized patients without TBI rather than to the type of healthy community dwelling control participants that we used. The very best commercial ELISA assay has a lower detection limit of ∼37 pg/mL, whereas the SIMOA platform is much more sensitive (LOD 1.9 pg/mL; LLOQ 9.36 pg/mL), which likely explains our lower measured levels.

Similarly, Mannix and colleagues⁹⁵ reported that children <18 years show UCH-L1 plasma concentrations of 150 pg/mL; they used the Alinity i TBI test from Abbott, which is designed to measure the highly elevated plasma concentrations of UCH-L1 that follows TBI and is not validated for measuring low levels in healthy controls. Again, the SIMOA platform is much more sensitive.

Rat AD model

The 18- to 20-month-old TgF344-AD male rats were injected subcutaneously with GM-CSF (83.3 μg/kg/day; 5 days/week; n = 7) or with saline (200 μL/day; 5 days/week; n = 7) for 24 injections total over 32 days. On day 32, the rats were anesthetized with sodium pentobarbital, perfused intracardially with PBS for 5–7 min, and the brains were removed rapidly. The right hemisphere was immersed in freshly prepared 4% paraformaldehyde (PFA) in PBS for 24 h at 4°C. After fixing with PFA, 4 μm-thick paraffin-embedded hippocampal brain sections were mounted on glass slides and processed for immunohistochemistry for the apoptosis marker Caspase-3 (Rabbit anti-Caspase-3 antibody; Cell Signaling Technology; Cat# 9662; dilution 1:100) or for GFAP (Rabbit anti-GFAP antibody; Abcam, MA, USA; Cat#7260; dilution 1:200) and co-stained with DAPI. A detailed protocol for immunohistochemistry and imaging is described in.⁵⁵ We assessed total GFAP staining by ImageJ analysis and visually counted the numbers of Caspase-3-positive cells in the CA1, CA3, and dentate gyrus/hilus regions of the hippocampus (blinded as to genotype and treatment). Specifically, for counting the Caspase-3-positive cells in hippocampal subregions, we used hippocampal sections with similar septo-temporal areas to compare between groups. Therefore, when we examined the Caspase-3-positive cells in the hippocampus, all of our images were taken from similar regions for all rats. We repeated 2-6 sections/rat, counted Caspase-3-positive cells from each section, and then the mean of these repeated values was used as one value for each rat and used for statistical analysis considering with the number of technical replicates (section/rat). For determining the proportion of Caspase-3-positive cells that were neurons, co-staining for MAPT was performed (Chicken anti-MAP2 antibody; PhosphoSolutions, CO, USA; Cat# 1100-MAP2; dilution 1:500) and the coded slides were examined visually. Specifically, two random aged WT mice were selected for this analysis, shown below.

Caspase-3-positive cells that were also MAPT2-positive:

CA1 = 95.48% and 93.39%

CA3 = 92.25%

DG = 94.20% and 91.86%

Quantification and statistical analyses

Data were analyzed using mixed model regression, with unstructured error covariance on repeated measures, for the effects of biomarkers, disease status, and treatment on the logarithmic transforms of the plasma biomarkers of UCH-L1, GFAP, and NfL. Although data from healthy control participants and participants with MCI due to AD were cross-sectional, the framework of mixed model regression could still be adapted, as regression with independent data is merely a simplified version of regression with correlated data. Healthy control, MCI due to AD, mild-to-moderate AD at baseline, mild-to-moderate AD treated with placebo/saline, and mild-to-moderate AD treated with sargramostim/GM-CSF were allowed different covariance matrices. Because the biomarker data was measured using 2 or 3 replicates, all of the observations were used, instead of averaging replicates and weighting. The age effects were modeled as log-linear, with separate age slopes and intercepts for healthy control, MCI due to AD, and mild-to-moderate AD. The intercept for mild-to-moderate AD depended on the treatment x study time, but a common age slope was assumed for all mild-to-moderate AD participants, and independent of treatment effects. The log linearity assumption was checked graphically, and by comparing to spline fits.⁹⁶ It was apparent that log linearity was a reasonable fit for the UCH-L1 and NfL data, but not for the GFAP data, which was subjected to spine analysis. Piece-wise log-linear regression models were considered if deviation from pure log-linearity was judged important. Interactions with gender were also considered. Linear combinations of model parameters were estimated, along with 95% confidence intervals, back transformed, and tested. Histograms of the Coefficient of Variation among replicates found some biomarker measures obtained with the SIMOA, SR-X Analyzer system, particularly UCH-L1, for example, can have Coefficients of Variance higher than 20%, as has also been found in UCH-L1 assessments after TBI using various assays, indicating that the variance is an inherent feature of the measure and not due, for example, to unreliable outliers. For example, a histogram of the UCH-L1 coefficients of variance of the healthy controls is shown in supplemental information, Figure S3.

There are known limitations to using biomarkers with high CVs in analyses, with most studies utilizing less than 20% cutoffs at the higher end. The primary concern is that, for example, if analysis of the same sample only twice yields two markedly different values, there are understandable concerns for reliability of measurement (and clinical applicability for an individual). Variability of such measures limits their utility for assessing individuals from a small number of replicates, but the laws of large numbers still allow us to estimate the average with a medium to large sample, when the purpose is to assess an overall effect of age or intervention. For cross-sectional comparisons between large cohorts, the average or log average of all individual participant’s replicate measures is often used, as we have done here, rather than first averaging an individual’s few measurements. Measures of UCH-L1 often show high variance,^33,38,63,97 but UCH-L1 levels were FDA approved in 2018 as part of a measure of brain damage after TBI.⁹⁸ Because some of our cohorts had two replicates for each participant (healthy controls and participants with MCI due to AD from the Bio-AD study) and some (healthy controls from the HTP, sargramostim/GM-CSF AD trial participants with mild-to-moderate AD, and healthy controls from the MS study) had mostly three replicates for each participant, and some of the biomarkers, especially UCH-L1, have high variability among replicates, we decided in the modeling to use each replicate as a separate measure in the main analysis, rather than using the averages of the individual participants’ replicate measures. The variation among replicates was modeled by introducing an additional noise term into the model. A common variance for the replicate noise term was assumed across all treatments because of software option limitations. Model predicted values of the response, along with pointwise standard errors, were plotted. All data calculations are provided in Tables S6, S7, S8, S9 and S10. As a test of confidence in the results, the averages of the participants’ replicate measures for the biomarkers were also modeled, weighted by the number of replicates, and the conclusions of age-associated exponential increases in the three biomarkers, UCH-L1, NfL, and GFAP, and the effect of GM-CSF treatment on reducing UCH-L1 levels were obtained. The overall results and conclusions were the same. Univariate statistical significance was set at alpha = 0.05, two-sided, for all tests unless otherwise stated. Statistics were computed using SAS 9.4 and R 4.1.3. There are remaining limitations to using this approach in terms of both within-site and cross-site replication of findings, and we acknowledge that it is challenging to move a biomarker into clinical practice for AD with high CVs, although it has been done for UCH-L1 in TBI, and multiple replicate measurements will provide higher confidence that the average is a valid measure of the biomarker plasma concentration in an individual.

The number of Caspase-3-positive cells in the TgF344-AD rat model were analyzed separately for each region using negative binomial count models with a log link, robust standard error method, and weighting by the number of technical replicates. Score tests were used for p values. A protective omnibus test for group differences was performed, followed by pairwise contrasts. The protective omnibus method for controlling multiple testing permits the pairwise contrasts to be performed univariately, provided that the omnibus test is statistically significant, which it is for each hippocampal subregion. To avoid the potential distorting effect of assessing cell numbers by the area, in this study, we used similar septo-temporal axes of brain sections and compared counted cell numbers between groups. Because individual cells are counted in several sections across similar regions of the whole of the tissue regions of interest and not per square millimeter, atrophy is not the problem that would arise from total tissue expression measures by immunohistochemistry, which might better be adjusted by using stereology.

The statistical analyses for GFAP protein expression were performed by using a negative binomial count model with a log link, robust standard error method, and weighting by the number of technical replicates (Figure 6). Score tests were used for p values. An omnibus test for group differences was performed, followed by pairwise contrasts. The protective omnibus method for controlling multiple testing permits the pairwise contrasts to be performed univariately, provided the omnibus test is statistically significant, which is for each hippocampal subregion.

The ability of the biomarkers to discriminate between health controls, MCI, and AD was investigated fitting logistic regression models, with logarithmically transformed biomarkers and age as explanatory variables, and the group as the outcome. For each model, subsamples were used to impose a common age range for the groups. Explanatory power was assessed with receiver operating characteristic (ROC) curves, showing the tradeoff between sensitivity and specificity, and the area under the curve.

Additional resources

Pilot Phase 2 Trial of the Safety & Efficacy of GM-CSF (Leukine) in the Treatment of Alzheimer’s Disease (COMIRB # 12–1273; NCT01409915; https://www.clinicaltrials.gov/study/NCT01409915).

Crnic Institute Human Trisome Project (COMIRB #15–2170; NCT02864108; https://clinicaltrials.gov/expert-search?term=NCT02864108).

Published: December 19, 2025