Intranasal insulin modulates cerebrospinal fluid markers of neuroinflammation in mild cognitive impairment and Alzheimer’s disease: a randomized trial

Intranasal insulin (INI) has shown promise as a treatment for Alzheimer’s disease (AD) in pilot clinical trials. In a recent phase 2 trial, participants with mild cognitive impairment (MCI) or AD who were treated with INI with one of two delivery devices showed improved cerebral spinal fluid (CSF) biomarker profiles and slower symptom progression compared with placebo. In the cohort which showed benefit, we measured changes in CSF markers of inflammation, immune function and vascular integrity and assessed their relationship with changes in cognition, brain volume, and CSF amyloid and tau concentrations. The insulin-treated group had increased CSF interferon-γ (p = 0.032) and eotaxin (p = 0.049), and reduced interleukin-6 (p = 0.048) over the 12 month trial compared to placebo. Trends were observed for increased CSF macrophage-derived chemokine for the placebo group (p = 0.083), and increased interleukin-2 in the insulin-treated group (p = 0.093). Insulin-treated and placebo groups showed strikingly different patterns of associations between changes in CSF immune/inflammatory/vascular markers and changes in cognition, brain volume, and amyloid and tau concentrations. In summary, INI treatment altered the typical progression of markers of inflammation and immune function seen in AD, suggesting that INI may promote a compensatory immune response associated with therapeutic benefit.

Treatment effect on CSF analytes. CSF analytes examined for the present study are listed in Table 2.
Means for all analytes are presented in Supplementary Table 1. Briefly, analyte concentrations were analyzed by repeated measures analysis of variance, with baseline concentration of the analyte, cognitive diagnosis, age, ApoE4 status, baseline MMSE, and sex as covariates as described in "Methods" section. Interactions between treatment arm and time were observed for IFN-γ, eotaxin, and IL-6 (ps = 0.032, 0.049, and 0.048). For ease of depiction, least squares mean adjusted change scores and standard deviations are presented in Fig. 2. CSF IFN-γ concentrations increased over the course of the trial in the insulin-treated group while decreasing in the placebo group. CSF eotaxin concentrations increased in the insulin-treated group to a greater degree than in the placebo group. CSF IL-6 decreased in the insulin-treated group and remained stable in the placebo group. CSF MDC and IL-2 were also found to have trending interactions (ps = 0.083 and 0.093, Fig. 2); MDC increased in the placebo group and decreased slightly in the insulin group, while IL-2 increased more in the insulin group than in the placebo group.
Next, we explored relationships between changes in CSF immune, inflammation, and vascular markers, with key study outcomes: (1) CSF markers of AD (Aβ40, Aβ42, Aβ42/40 ratio, total tau, tau-p181, Aβ42/tau ratio), MRI measurements of AD (global WMHV, deep WMHV, hippocampal and entorhinal cortex volume, www.nature.com/scientificreports/ temporal-parietal meta-region of interest), and performance on cognitive and functional tests (Alzheimer's Disease Assessment Scale-Cognition version 13/ADAS-Cog13, memory composite, Activities of Daily Living-MCI version/ADL-MCI, Clinical Dementia Rating Scale-Sum of Boxes/CDR-SB). As detailed below, the insulin and placebo groups showed distinctly different patterns of associations between immune/inflammatory/vascular markers and study outcomes. These patterns are depicted in a heatmap in Fig. 3, which includes all observed nominally significant associations (all ps < 0.05). Due to the exploratory nature of this analysis no correction was made for multiple comparisons.
Cognitive and functional tests. For the insulin group (Fig. 3F), ADAS-Cog change negatively correlated with change in MDC (r = − 0.63, p = 0.012). In the placebo group (Fig. 3G) a negative correlation was observed between changes in ADAS-Cog and IL-12/IL-23p40, as well as between CDR-SOB and MIP-1alpha (r = − 0.49, Table 2. Clinical, imaging and CSF AD variables from the parent trial, and newly analyzed CSF vascular, inflammatory and immune markers.  Vascular injury. Next, we examined relationships between changes in CSF markers of vascular injury and changes in CSF markers of AD, MRI measurements of AD, and performance on cognitive and functional tests. In both the insulin-treated and placebo groups, changes in VEGF-D correlated with Aβ42 changes (insulin r = 0.62, p = 0.013; placebo r = 0.59, p = 0.01). This was the only nominally significant correlation demonstrated by both groups.

Discussion
In this retrospective study, we compared the effects of 12 months of treatment with intranasal insulin versus placebo on CSF markers of immune function, inflammation, and vascular injury in a trial cohort for whom insulin treatment was associated with beneficial effects on clinical outcomes and AD CSF and imaging biomarker profiles. Insulin treatment increased concentrations of CSF IFN-γ and eotaxin, with a similar trend for IL-2.
In contrast, IL-6 was lowered with insulin treatment compared to placebo, and MDC tended to increase in the placebo group while remaining stable in the insulin group. In exploratory analyses, we then examined relationships between changes in CSF markers of inflammation, immune function and vascular integrity and clinical/ AD biomarker outcomes, reasoning that correlated changes with insulin treatment might reflect underlying mechanisms related to its beneficial effects, whereas correlated changes in the placebo group might reflect the natural progression of AD. Supporting this dichotomy, we observed strikingly distinct patterns of associations in the two groups. Although our results should be considered exploratory and in need of future validation, they suggest that INI may promote a beneficial compensatory immune response that moderates the progression of AD pathology and symptoms. We first examined whether insulin treatment changed levels of CSF immune, inflammation, and vascular markers during the 12 month treatment period compared with placebo. We found that INI increased IFN-γ, whereas IFN-γ concentrations decreased for the placebo group. Modulation of IFN-γ may have wide-ranging impact, given that it plays a critical role in orchestration of the innate and adaptive immune systems, in part through effects on microglial immune function and surveillance 30 . While one might assume that markers of increased inflammation would be associated with faster AD progression, recent reports have described the opposite relationship: higher levels of IFN-γ and other cytokines were associated with slower AD progression 31,32 . INI may thus be raising IFN-γ concentrations to levels associated with slowed AD pathology. Insulin has been shown to directly stimulate IFN-γ production by CD8 + T cells 33 ; these cells are found in CSF where they provide immune protection, and have also been documented in human brain in white matter, located primarily in the perivascular space 34 . These locations provide ready access to stimulation by intranasal insulin which raises CSF insulin levels and which follows perivascular channels via bulk flow, bypassing the blood brain barrier. Our findings raise the possibility that insulin-induced increases in IFN-γ concentrations in the CNS may provide neuroprotection and attenuate cognitive decline.
We also observed that INI administration increased CSF eotaxin concentrations. Eotaxin is upregulated in the presence of inflammation and increases in both CSF and plasma with age 35 ; however, its role in the brain and its relationship to AD progression are complex. Although a small study reported no differences in CSF eotaxin concentrations between normal adults and those with AD 36 , a recent finding demonstrated that, similar to IFN-γ, higher CSF eotaxin concentrations were associated with slower cognitive decline in AD 31 . In the CNS, eotaxin is produced in choroid plexus epithelial cells in response to IFN-γ, as well as by activated astrocytes surrounding blood vessels. As with IFN-γ, both loci would be readily accessible to INI; further, the effects of eotaxin are thought to be modulated by IFN-γ, such that higher levels move eotaxin in an anti-inflammatory direction 37 . In vitro, eotaxin binding to its cognate receptor CCR3 promotes angiogenesis through the AKT/PI3K pathway, one of the primary insulin signaling pathways and thus its effects may be amplified by INI 38 . In a rodent model of multiple sclerosis, eotaxin was associated with increased BBB functioning, implicating this chemokine in vessel health, as well as with increased proliferation of oligodendrocyte proliferation and remyelination 39 . These results may have relevance for human studies of ischemic stroke, in which higher eotaxin levels predicted smaller stroke volume and better functional outcomes 3 and 12 months post-stroke 40 . Higher levels of CSF eotaxin have also been associated with slower functional decline in patients with amyotrophic lateral sclerosis 41 . Thus insulin-induced increases in eotaxin may benefit multiple pathways of relevance to AD and other CNS disorders.
The cytokine IL-2 showed trend increases in response to INI treatment, and is also closely interrelated with IFNγ 42 . It is produced by CD4+ and CD8+ T cells which as noted have been documented in CSF and perivascular white matter; it is also produced by astrocytes and neurons. In APP/PSE1 mice, AAV-induced IL-2 elevations resulted in diverse beneficial effects including recruitment of astrocytes to amyloid plaques, reduced brain Aβ42/ Aβ40 ratio and overall amyloid load, and improved spine density, synaptic plasticity, and memory 43 . Higher CSF IL-2 levels have been associated with slower cognitive decline in patients with MCI, leading to suggestions that IL-2 might be explored as a potential treatment for AD 44 . Clinical trials have targeted IL-2 for treatment of other diseases and confirmed its anti-inflammatory effects, which have been described as shifting the immune balance towards regulation over inflammation, and preserving T cell homeostasis 45,46 .
In addition to inducing increases in anti-inflammatory cytokines, we observed that insulin treatment reduced CSF levels of IL-6, a cytokine with multi-faceted roles in the transition from innate to acquired immunity, as well as in inducing neurotrophic responses 47,48 . It is expressed in astrocytes and endothelial cells in response to elevation in other cytokines such as IL-1β, and neuronal expression is increased in response to glutamate in rodent models of ischemia and epilepsy 48 . IL-6 has been proposed to have pro-inflammatory effects in AD; Aβ induces IL-6 expression in both astrocytes and microglia in vitro 49 , and in hippocampal neurons both Aβ and IL-6 induce synaptic dysfunction 50 . Several studies examined whether levels of CSF IL-6 are elevated in AD, but have produced conflicting results [51][52][53][54] . Given the multiple roles played by IL-6, and its complex signaling and trans-signaling activity future investigations will be needed to determine the mechanisms through which insulin may have lowered levels, and how this reduction may have impacted disease progression.
We also found that macrophage-derived chemokine (MDC, CCL22) concentrations tended to increase in the placebo group and decreased slightly with INI treatment. MDC and its receptor CCR4 play a role in inflammation and homeostasis; reduced microglia activation in mice with a TREM2 genetic mutation that increases risk of AD is accompanied by increased MDC concentrations 55  www.nature.com/scientificreports/ context of human AD, with very few studies assessing CSF levels. Interestingly, in a previous study of participants with MCI and AD, CSF MDC concentrations were modulated by treatment with resveratrol 56 . Taken together, we observed increases in several markers of inflammation and immune function in the insulin-treated group, who showed clinical benefit on measures of cognition and daily function and improved AD biomarker profiles in the original trial. These observations are for the most part consonant with observational studies, which have generally reported increased levels of cytokines such as IFN-γ and eotaxin to be associated with slower progression in AD. At first glance, such associations may appear counter-intuitive. However, an emerging conceptualization of inflammation and immune function as not intrinsically harmful or beneficial has emerged, with greater understanding that the impact of such changes depends on the concentrations of molecules, their location, their co-occurrence with other inflammatory/immune molecules, and with other influential aspects of the neurobiological milieu. Accordingly, a concept has emerged of a compensatory immune response system invoked by chronic inflammation in CNS disorders 57 ; higher levels of such markers may reflect individual capacity to mount such a response, which in turn may enable resilience to pathological changes resulting in slower clinical progression. The anti-inflammatory effects of insulin have been noted in multiple peripheral systems, linked to inhibition of NFkβ and the NLRP3 inflammasome 58,59 . Our results suggest that intranasal insulin treatment may invoke these anti-inflammatory properties in the CNS, with corresponding therapeutic benefit in AD.
In subsequent exploratory analyses we examined the relationships between changes in markers of immune function, inflammation, and vascular integrity and changes in key outcomes from the original trial, including classical AD CSF biomarkers, imaging biomarkers, and cognitive/functional measures for the insulin-treated and placebo groups. A striking difference in the overall pattern of relationships in the two groups emerged; the insulin and placebo groups had nearly completely different associations, with only one correlation overlapping between groups. This divergence suggests that INI might alter the typical progression of AD as demonstrated by the placebo group.
Inspection of specific correlations showed frequent associations involving several markers of inflammation and immune function in the insulin-treated group. IL-1β has been found to be elevated in the CSF of adults with AD 18 . In the insulin-treated group, increased IL-1β, eotaxin-3 and IL-8 were associated with reduced Aβ42, Aβ40, and P-tau. A speculative interpretation of these relationships is that insulin treatment enhanced the immune/ inflammatory response which in turn improved microglial phagocytosis and degradation of amyloid, preventing increased CSF concentrations, and amyloid-induced elevations in P-tau. This possibility is also consistent with the finding that increased eotaxin-3 levels correlated with less loss of hippocampal gray matter volume.
Several markers of vascular injury were also related to outcomes from the parent trial for the insulin-treated group. Serum amyloid A (SAA) is secreted in response to inflammation and associated with damage to blood vessels 60 . We found that decreased SAA following insulin treatment correlated with greater preservation of function measured by the ADL-MCI scale. We also found that insulin-induced increases in VEGF, a growth factor that stimulates blood vessel health, angiogenesis, and wound repair 61 , predicted better performance on a memory composite score. Two other markers of angiogenesis, tyrosine kinase-2 (Tie-2) and placental growth factor (PlGF), positively correlated with Aβ42/40 and Aβ42/T-tau ratios respectively. PlGF is notable for its involvement in maturation and stabilization of blood vessels under hypoxic conditions. Both PlGF and Tie-2 have been shown to modulate BBB function 62,63 ; increased blood vessel stability could enhance clearance of Aβ42 from the brain and into the CSF. Although not well-studied in AD, one study has reported increased concentrations in CSF PlGF from adults with AD or other neurodegenerative disorders, and a second showed lower levels of Tie2 in AD 64,65 . As the BBB is known to be disrupted in AD, insulin-induced modulation of PlGF and Tie-2 could improve BBB function and thereby aid in the clearance of Aβ.
Our study has several limitations. It is an exploratory study with post hoc analyses in a small sample and therefore not powered to account for multiple comparisons. Future larger studies are needed to validate changes in these neuroinflammatory/immune/vascular biomarkers in response to intranasal insulin treatment and examine their relationship to AD progression. Larger studies are also needed to further understand inter-relationships among these markers, perhaps through application of more sophisticated multivariate or machine learning approaches. Although our results demonstrate robust effects in response to 12 months of INI administration, an important question remains concerning the brain's response to longer insulin exposure. Despite these weaknesses, we believe these findings provide strong rationale for future examination of INI in AD and MCI using well-validated delivery technologies.
In conclusion, our results showed that intranasal insulin treatment modulated markers of immune function, inflammation, and vascular integrity in the cohort who showed clinical benefit in the parent trial in a manner that suggests activation of a compensatory immune response system. Within the insulin-treated group, a pattern of associations was observed between changes in these markers and changes in CSF, imaging and clinical outcomes from the parent trial that diverged strikingly from patterns observed for the placebo group, suggesting that insulin is altering the progression of AD in fundamental ways. Our results also highlight the complexity of interpretation of immune/inflammatory profiles, in which negative or beneficial effects are influenced by the neurobiological milieu at each stage of disease. Further research is needed to identify the pathways through which these immune/inflammatory/vascular mediators act and the specific mechanisms through which they are affected by insulin. Finally, our findings provide additional evidence that INI is a promising therapeutic option for AD.

Methods
This study consisted of a 12-month placebo-controlled double-blinded phase, in which participants were rand- www.nature.com/scientificreports/ registered with ClinicalTrials.gov (NCT01767909; first date of registration 15/01/2013). Eligibility and recruitment methods for this study have been described elsewhere 12 . The parent study used two devices to deliver insulin intranasally. The initial device for the trial had been used in previous INI studies, but new adaptations introduced by the manufacturer caused sporadic reliability issues, resulting in compliance that was slightly less than the desired 80% target (compliance for both placebo and insulin groups averaged about 73%); thus a second device from a different manufacturer was adopted after trial onset which had not been used in previous AD studies. However, only the cohort using the initial device demonstrated changes in AD CSF biomarkers and cognitive benefits over the 18-month long study, and the cohort that used that device is the subject of these retrospective analyses. Study protocols and informed consent were approved by the institutional review boards at the individual study sites which were all academic research centers, and at the University of California, San Diego, the University of Southern California, and the Wake Forest School of Medicine. All methods were performed in accordance with relevant guidelines and regulations. Written informed consent was obtained from participants and study partners. The study was conducted under local institutional review board supervision. Briefly, participants with amnestic MCI (n = 18) or AD (n = 31) were recruited from 12 sites. Participants received baseline testing including the ADAS-Cog13, Mini-Mental State Exam (MMSE), Clinical Dementia Rating (CDR), a lumbar puncture, and an MRI, then were randomized on a 1:1 basis to receive either placebo (n = 25) or 20 IU intranasal insulin (n = 24) twice daily for 12 months using a weighted algorithm as previously described 12 . After 12 months the cognitive battery was readministered, another lumbar puncture was performed, and another MRI was obtained. A total of 38 participants (placebo n = 20; insulin n = 18) with CSF data that passed quality control measures at baseline and month 12 were included in analyses.

Imaging biomarkers. T1 and Fluid Attenuated Inversion
Recovery images were collected with 1.5 or 3 T MRI. FreeSurfer 6.0.0 was used to produce participant-specific gray matter volume by processing T1 weighted images. FLAIR images were segmented by the lesion growth algorithm 66 as implemented in the LST toolbox version 3.0.0 (www. stati stica lmode lling. de/ lst. html) for SPM. The algorithm first segments the T1 images into the three main tissue classes (CSF, GM and WM). This information is then combined with the coregistered FLAIR intensities in order to calculate lesion belief maps. By thresholding these maps with a pre-chosen initial threshold (κ = 0.3) an initial binary lesion map is obtained which is subsequently grown along voxels that appear hyperintense in the FLAIR image. The result is a lesion probability map. The lesion probability maps were then warped to MNI space and lobar volume was extracted using Mayo Clinic Adult Lifespan Template 67 . Global WMHV was then calculated by excluding brainstem and cerebellum. T1 weighted images were used to analyze gray matter volume. Bilateral entorhinal, inferior temporal, middle temporal, inferior parietal, fusiform, and precuneus were used to construct a temporal-parietal GMV meta-ROI 68 . Hippocampal and entorhinal volumes were extracted and analyzed individually. CSF biomarkers. CSF was collected in the morning after an overnight fast. CSF aliquots were immediately placed on dry ice and either stored at − 80 °C for later analysis or shipped overnight to a central biomarker laboratory for analysis of AD biomarkers Aβ42, Aβ40, total tau, and tau phosphorylated at threonine 181 using the Meso Scale Discovery platform as previously described 12 . Neuroinflammation/immune/vascular markers were assessed en bloc on first thawed CSF samples using Meso Scale Discovery Neuroinflammation Panel 1 multiplex kits on a MESO Quickplex SQ 120 (Meso Scale Diagnostics, Rockville, MD, USA) at the Wake Forest School of Medicine. This kit is a compilation of 5 individual multiplex kits as follows: Cytokine Panel 1, Chemokine Panel 1, Proinflammatory Panel 1, Vascular Injury Panel 2, and Angiogenesis Panel 1. These panels included 37 analytes which have been identified as frequently abnormal using human CSF from AD, Parkinson's Disease and other neurological disorders, as well as healthy controls, and is promoted as a tool for discovery (for a description of this panel and its development, see https:// www. mesos cale. com/ ~/ media/ files/ produ ct% 20ins erts/ neuro infla mmati on% 20pan el% 201% 20hum an% 20ins ert. pdf). Protocols were modified to extend the standard calibration curve to increase the number of samples with analyte concentrations above the lower limit of detection. Reliably detectable concentrations could not be obtained for IL-4 which was then eliminated from further analysis. Remaining analytes are listed in Table 2  . Mean intraassay CVs for internal control analytes were as follows: Cytokine Panel 1 (4.2%), Chemokine Panel 1 (4.7%), Proinflammatory Panel 1 (9.3%), Vascular Injury Panel 2 (2.4%), and Angiogenesis Panel 1 (3.7%). Analyte concentrations below the LLOD were assigned a value of one half the value of LLOD. APOE genotyping was performed on DNA isolated from whole blood using an established protocol and participants were dichotomized into E4 carriers and non-carriers 12 .
Cognitive and functional testing consisted of the ADAS-Cog, CDR-SOB, Alzheimer Disease Cooperative Study ADL, and a memory composite defined as the sum of z scores from the Free and Cued Selective Reminding Test and immediate and delayed story recall [69][70][71] . Together cognitive and functional tests, CSF Aβ and Tau concentrations, and MRI measurements of WMHV and GMV are referred to as markers of AD progression. www.nature.com/scientificreports/ Cross sectional analysis was performed to assess group differences at baseline in age, cognitive status, and sex using chi squared tests or general linear models when appropriate. Change in GMV and WMHV were constructed as previously described 13 . In the first phase of analysis, to determine whether there were treatmentrelated changes in neuroinflammatory/immune/vascular markers over the 12 month treatment period, general linear modeling was performed in SAS v 9.4 to conduct repeated measures analysis of variance, with treatment group (insulin vs. placebo) as the independent factor and time (baseline vs. 12 months) as the within group factor, and with inclusion of the following covariates: baseline concentration of the analyte, cognitive diagnosis, age, ApoE4 status, baseline MMSE, and sex. Non-contributing covariates (p > 0. 15) were dropped from the model. Following the identification of a significant time by treatment interaction, post hoc comparisons of least square adjusted means were conducted. In the second exploratory phase of analysis, Pearson's r correlations were used to determine relationships between changes in neuroinflammatory/immune/vascular variables, and cognitive and AD biomarker changes separately for insulin and placebo groups. Due to the exploratory nature of the analyses, no corrections for multiple comparisons were applied, and all correlation p values < 0.05 were considered nominally significant. All correlations with p < 0.05 were then plotted in a heat map as shown in Fig. 3 to enable the appreciation of differences in correlation patterns between the insulin and placebo groups. www.nature.com/scientificreports/