Insights into neuroepigenetics through human histone deacetylase PET imaging

Abstract

Epigenetic dysfunction is implicated in many neurological and psychiatric diseases, including Alzheimer’s disease and schizophrenia. Consequently, histone deacetylases (HDACs) are being aggressively pursued as therapeutic targets. However, a fundamental knowledge gap exists regarding the expression and distribution of HDACs in healthy individuals for comparison to disease states. Here, we report the first-in-human evaluation of neuroepigenetic regulation in vivo. Using positron emission tomography (PET) with [¹¹C]Martinostat, an imaging probe selective for class I HDACs (isoforms 1–3), we found HDAC expression is higher in cortical gray matter than white matter with strikingly conserved regional distribution patterns within and between healthy individuals. Among gray matter regions, HDAC expression is lowest in the hippocampus and amygdala. Through biochemical profiling of postmortem human brain tissue, we confirmed [¹¹C]Martinostat selectively binds HDAC isoforms 1–3, the HDAC subtypes most implicated in regulating neuroplasticity and cognitive function. To relate the PET imaging signal to the epigenetic regulation of gene transcription, we measured mRNA expression changes elicited by Martinostat in human stem cell-derived neural progenitor cells. We demonstrate that genes closely associated with synaptic plasticity, including BDNF (brain derived neurotrophic factor) and SYP (synaptophysin), as well as genes implicated in neurodegeneration, including GRN (progranulin), were markedly increased at the transcript level in concert with increased acetylation at both histone H3 lysine 9 and histone H4 lysine 12. This study provides the first quantification of HDAC expression in the living human brain, and provides the foundation for gaining unprecedented in vivo epigenetic information in the healthy and diseased human brain.

Introduction

Disorders of the central nervous system (CNS), including Alzheimer’s disease (AD), schizophrenia, depression, and addiction, are increasingly recognized to involve the dysregulation of epigenetic machinery. Among all, histone deacetylases (HDACs) – a family of chromatin-modifying enzymes that dynamically regulates gene transcription, are the most frequently implicated (1, 2). A subset of HDACs has already been linked to neuronal development, synaptic plasticity, and cognition (3, 4). For example, postmortem human brain tissue analyses and in vivo rodent studies expose HDAC1, HDAC2 and HDAC3 as antagonists of learning and memory, and contributors to AD and mood disorders (3, 5–8). Genetic manipulations or pharmacologic inhibition of aberrant HDAC2 and HDAC3 activity can rescue behavioral defects in rodent models of both AD and mood disorders (6, 7, 9–13). HDAC inhibitors have also been proposed as targeted treatment of frontotemporal lobar degeneration due to mutations that cause haploinsufficiency of the progranulin-encoding gene GRN (13). Collectively, these studies indicate a direct relationship between the levels of class I HDAC (isoforms 1–3) and neuronal function.

In addition to the overall level of HDAC expression within the brain, spatially localized variation of HDACs is also highly impactful in neuronal plasticity, memory, and behavior. For example, intra-hippocampal injection of short-hairpin RNA (shRNA) against Hdac2 selectively normalizes HDAC2 levels and restores neuroplasticity-associated gene transcription, synaptic density, and cognitive behavior in a mouse model of AD (6). In contrast to the high level of hippocampal HDAC2 in animal models and postmortem human tissue from AD patients, deficient HDAC2 expression is observed in the frontal cortex of postmortem AD tissue, highlighting the importance of tightly regulated localized HDAC expression (14). Analogously, focal genetic deletion of Hdac3 in the hippocampus and the nucleus accumbens enhances long-term memory and acquisition of cocaine-associated place preference in mice, respectively (5, 15). While understanding of the full compendium of genes under HDAC-dependent regulation in defined regions of the brain is incomplete, HDAC2 chromatin immunoprecipitation studies in hippocampal tissue have identified a number of immediate-early genes (e.g. BDNF, CDK5) involved in learning and memory, as well as multiple genes involved in synaptic plasticity, (e.g. SYP, STY1) as downstream targets (3, 6, 16). Collectively, these studies provide unequivocal support that localized HDAC expression levels drive pivotal epigenetic mechanisms that modulate neuronal function.

Although there is strong evidence for localized HDAC dysfunction in CNS disease, epigenetic models cannot recapitulate dynamic human-environmental interactions and therefore may not accurately reflect in vivo human biology. Moreover, until now there was no method that enabled visualizing in vivo epigenetic mechanisms in humans. Here using positron emission tomography (PET) with our novel epigenetic imaging agent, [¹¹C]Martinostat (11, 17, 18), we report the first quantification of human epigenetic regulation.

Results

In vivo human PET imaging

To visualize, for the first time, HDAC expression in the living human brain, we performed [¹¹C]Martinostat PET imaging on eight healthy volunteers (4 males, 4 females, mean age ± SD: 28.6 ± 7.6 years) (Table S1). The uptake and retention of [¹¹C]Martinostat in the human brain was high, with regional heterogeneity easily observable at the individual subject level (Fig. 1 and Fig. S1). Quantitative analysis of the dynamic PET data using compartmental modeling in individual subjects showed that the distribution volume (V_T), a measure of radiotracer binding that is normalized to the activity present in circulating blood, and micro-parameters describing the pharmacokinetics of [¹¹C]Martinostat, can be determined robustly (Fig. S1, Supplementary Table 2 and Supplementary Table 3). V_T values appear stable beyond 50 min, with less than 10% variability when compared to the full length 90 min data. For example, a V_T value estimated with a 60 min scan duration is different from a 90 min scan by less than 2% in the superior frontal cortex (SFC) (Fig. S2). Regional standard uptake values from 60 to 90 min post-radiotracer administration (SUV_60-90min), an image-based analysis of binding, correlated positively with V_T values (Pearson r=0.98; p<0.0001) (Fig. 2). By calculating the standard deviation of the mean of V_T and SUV_60-90min across brain regions, we found that intersubject variability was smaller using SUV_60-90min analysis (Fig. 2B; paired t-test, p<0.0001). This supports that SUV_60-90min may be an appropriate surrogate outcome measurement for V_T and can be used in future studies to eliminate arterial blood sampling and reduce sample size because of its smaller variation. Although as with all surrogate measures, validation relative to a full treatment of the data using arterial blood in each patient population may be required to ensure a full, accurate interpretation.

Fig. 1. [¹¹C]Martinostat images of an individual subject show high cortical binding and distinct gray-white matter differences.

[¹¹C]Martinostat (injected dose: 4.7 mCi; specific activity: 1.1 mCi/nmol) images averaged from 60 to 90 min post-radiotracer injection (SUV_60-90min; SUV=radioactivity/injected dose/body weight) overlaid on the subject’s anatomical MR images. Panel on the bottom left shows the structure of the radiotracer, [¹¹C]Martinostat.

Fig. 2. Group [¹¹C]Martinostat SUV_60-90min images show small intersubject variation of localized regional [¹¹C]Martinostat binding.

(A) Mean images (upper left) and standard deviation images (inset on the lower right) of SUV_60-90min from healthy volunteers (n=8). The images are overlaid onto the MNI152 standard brain. (B) Correlation of regional distribution volume (V_T) values, derived from a two-tissue compartmental model using metabolite-corrected arterial plasma as an input function, and SUV_60-90min (n=6). The V_T values and SUV_60-90min were significantly correlated (Pearson r=0.98; N=14 brain regions; p<0.0001). (C) (Left) Regional SUV_60-90min of cortical (purple), subcortical (green), cerebellar (red), and white matter (blue) volumes of interest. *Individual pairs of brain regions that are significantly different (non-parametric Friedman test with Dunn’s multiple comparisons correction) from each other are the following: putamen vs. white matter and cerebellum vs. white matter (p<0.0001); occipital vs. white matter, putamen vs. hippocampus, putamen vs. amygdala, and cerebellum vs. hippocampus (p<0.001); insula vs. white matter, putamen vs. caudate, cerebellum vs. caudate, and cerebellum vs. amygdala (p<0.005); occipital vs. hippocampus and occipital vs. amygdala (p<0.05). (Right) Regional SUV_60-90min were normalized to the subject’s white matter SUV_60-90min (SUVR_60-90min). Each dashed line represents SUVR_60-90min from a single subject (n=8).

Group-level analyses showed that the average gray matter SUV_60-90min was 3.2 ± 0.4, whereas white matter was almost half of that value (1.7 ± 0.3) (Figs. 2 and S3). Among gray matter regions examined, heterogeneous binding was observed (Fig. 2; nonparametric Friedman test, p<0.0001). Comparisons between individual pairs are listed in the figure legend. Besides the white matter, the lowest [¹¹C]Martinostat SUV_60-90min were observed in the hippocampus (2.4 ± 0.3) and amygdala (2.4 ± 0.3) and the highest SUV_60-90min were observed in the putamen (3.7 ± 0.5) and cerebellum (3.6 ± 0.5) (Fig. 2 and S3). To facilitate intersubject comparison of regional HDAC distribution, we normalized regional SUV_60-90min to individual subject’s white matter SUV_60-90min as SUV_60-90min ratios (SUVR_60-90min). SUVR_60-90min showed that the regional distribution patterns of [¹¹C]Martinostat binding are remarkably consistent in all subjects (Figs. 2C and S4).

Ex vivo biochemistry of postmortem human brain tissue

To assess regional differences in [¹¹C]Martinostat binding, we biochemically profiled postmortem human brain tissue from a gray matter region (superior frontal gyrus, SFG) and a white matter region (corpus callosum, CC) (Supplementary Table 4). Quantitative protein levels of HDAC1, HDAC2, HDAC3 and HDAC6 were determined by western blotting and revealed significantly lower amounts of HDAC2 and HDAC3 in the CC relative to the SFG (Figs. 3, S5, and S6; unpaired t-test, **p< 0.01, *p<0.05). HDAC expression level differences between the SFG and the CC cannot be attributed to nuclear density (Fig. S7). The expression levels of HDAC2, HDAC3, and HDAC6 were similar in the SFG (0.12–0.16 pmol/mg total protein) with the notable exception of HDAC1 (1.7 pmol/mg total protein).

Fig. 3. HDAC2 and HDAC3 expression levels are higher in gray matter than white matter.

Whole cell lysates were prepared from postmortem human superior frontal gyrus (purple shading) and corpus callosum (blue shading) (n=3 replicate donor pools with 2 donors per pool). Equivalent amounts of total protein were compared to human recombinant HDAC standards through western blotting (Figs. S5, S6). HDAC immunoreactive band intensity values were normalized to GAPDH intensity values. HDAC expression levels were calculated per mg of total extracted protein. Solid lines represent mean expression values. Donor pools are denoted by black, gray, and open circles. Expression level differences in the superior frontal gyrus vs. corpus callosum are significant for HDAC2 and HDAC3 (unpaired t-test, ** p<0.01 and * p<0.05).

Thermal shift assays were performed with clarified brain homogenate and increasing concentrations of Martinostat. As expected, inhibitor-binding resulted in increased thermal stabilization (19, 20). Martinostat stabilized HDAC1, HDAC2 and HDAC3 in both the SFG and CC at nanomolar concentrations (Figs. 4A and S8–S10). No significant stabilization of either HDAC6 or HDAC8 (negative control) was observed. The former suggests differences between the accessibility of endogenous HDAC6 complex to Martinostat binding, relative to recombinant protein (17). To assess heterogeneity in HDAC isoform binding across gray matter regions, SFG binding was compared to the dorsolateral prefrontal cortex (DLPFC), hippocampus (Hipp) and anterior cingulate (Ant Cing). Based on thermal stabilization data, Martinostat exhibited a relatively uniform binding profile in gray matter with target engagement observed at concentrations around and above 0.16 μM (Figs. 4B and S11–13). Competition autoradiography was performed to compare the specific binding of [¹¹C]Martinostat in gray and white matter. [¹¹C]Martinostat binding in white matter is more biased by non-specific uptake (Fig. S14). Together, our in vivo imaging and ex vivo biochemistry results support that [¹¹C]Martinostat uptake reflects selective binding to a subset (HDAC1, HDAC2 and HDAC3) of class I HDACs across the human brain.

Fig. 4. Martinostat engages HDAC1, HDAC2, and HDAC3 in the human brain.

(A) Whole cell lysates were prepared from postmortem human superior frontal gyrus and corpus callosum (n=3 replicate donor pools with 2 donors per pool). Thermal shift assays were performed with increasing concentrations of Martinostat (0, 0.0032, 0.016, 0.080, 0.40, 2.0, and 10 μM). Thermal stabilization of HDACs 1, 2, 3, 6 and 8 were compared through western blotting with scaled immunoreactive band intensity values represented as an averaged heat map (n=3). The imaging-derived dissociation constant (K_d) for [¹¹C]Martinostat in non-human primate brain is indicated by the black arrow (18). See Figs. S8–S10 for original western blotting data. (B) Whole cell lysates were prepared from postmortem human superior frontal gyrus (n=3 replicate donor pools with 2 donors per pool), and dorsal lateral prefrontal cortex, hippocampus, and anterior cingulate (n=3 replicate donor pools with 3 donors per pool). Thermal shift assays were performed with increasing concentrations of Martinostat (0, 0.16, 0.80, 4.0, 20, and 100 μM). Thermal stabilization of HDACs 1, 2, 3, 6 and 8 were compared through western blotting with scaled immunoreactive band intensity values represented as an averaged heat map (n=3). See Figs. S11–13 for original western blotting data.

In vitro biochemistry of human neural progenitor cells

To link [¹¹C]Martinostat uptake with downstream HDAC substrate signaling and gene expression, human stem cell-derived neural progenitor cells were treated with increasing concentrations of Martinostat. Acetylation levels of established class I HDAC substrates, histone H3 lysine 9 (H3K9) and histone H4 lysine 12 (H4K12), were quantitated through western blotting (3, 21). Treatment with 2.5 μM Martinostat increased H3K9 and H4K12 acetylation levels as compared to DMSO (Figs. 5A and S15A; repeated measures two-way ANOVA with Dunnett’s multiple comparisons correction, **p<0.01, ****p<0.0001). mRNA transcript levels of memory, neuroplasticity, and neurological disease-related genes were quantitated through qPCR (3, 6, 16, 21). Treatment with 2.5 μM Martinostat increased BDNF, SYT1, SYP, and GRN expression levels as compared to DMSO (Figs. 5B and 15B; repeated measures two-way ANOVA with Dunnett’s multiple comparisons correction, *p<0.05, **p<0.01, ****p<0.0001).

Fig. 5. Martinostat increases histone acetylation and gene expression levels in human neural progenitor cells.

Human neural progenitor cells were treated with DMSO, Martinostat (MSTAT 0.5 μM, 2.5 μM) and SAHA for 24 hrs. (A) Whole cell lysates were prepared (n=3 replicates). Equivalent amounts of total protein were compared through western blotting (Fig. S15A). Histone acetylation immunoreactive band intensity values were normalized to GAPDH intensity values. Error bars represent standard deviation. Histone H3 lysine 9 and histone H4 lysine 12 acetylation levels were significantly increased as compared to DMSO with 2.5 μM Martinostat and SAHA treatments (repeated measures two-way ANOVA with Dunnett’s multiple comparisons correction, α=0.05, ** p<0.01, **** p<0.0001). (B) RNA was extracted (n=3 replicates) and converted into cDNA. mRNA transcript levels of memory/neuronal plasticity (blue) and monogenic neurological disorder-related genes (green) were compared through qPCR (n=9 technical replicates) and normalized to GAPDH mRNA levels (Fig. S15B). Error bars represent standard error of the mean. BDNF, SYT1, SYP, and GRN levels were significantly increased as compared to DMSO with 2.5 μM Martinostat treatment (repeated measures two-way ANOVA with Dunnett’s multiple comparisons correction, α=0.05, * p<0.05, ** p<0.01, **** p<0.0001). BDNF, SYP, and GRN levels were significantly increased as compared to DMSO with SAHA treatment (repeated measures two-way ANOVA with Dunnett’s multiple comparisons correction, α=0.05, ** p<0.01, **** p<0.0001). Gene abbreviations used are as follows: BDNF, brain-derived neurotrophic factor; CDK5, cyclin-dependent kinase 5; SYT1, synaptotagmin-1; SYP, synaptophysin; GRN, progranulin; FXN, frataxin; GAPDH, glyceraldehyde-3-phosphate dehydrogenase.

Discussion

The first-in-human epigenetic imaging study with [¹¹C]Martinostat establishes that HDACs are highly expressed throughout the healthy brain with region-specific distribution (Figs. 1 and 2). Based on our in vitro profiling with recombinant HDACs (17) and our ex vivo profiling with post mortem human brain tissue (Fig. 4), we define [¹¹C]Martinostat signal in the brain to be derived from class I HDACs (isoforms 1–3), which are relevant to cognition, memory, and mood regulation (3, 12, 15, 22). Consistent with in vitro recombinant inhibition data, Martinostat thermally stabilizes HDAC1, HDAC2 and HDAC3 in all regions that we analyzed (Fig. 4). Notably, Martinostat stabilizes these isoforms at a concentration of ~0.1 μM (Fig 4), which is consistent with the imaging-derived dissociation constant (K_d) for [¹¹C]Martinostat in non-human primate brain (18). Somewhat in contrast with in vitro recombinant inhibition data, Martinostat does not appear to stabilize HDAC6 in the brain regions that we assessed (Fig. 4), although it is worth noting that the recombinant assay provided a nearly 10-fold lower IC₅₀ for HDAC6 when compared to isoforms 1–3 (17). The minor discrepancy between the recombinant and tissue-based assays confirms the importance of physiological context (e.g. where HDAC complexes may be intact) for inhibitor-binding studies. At high concentrations α-tubulin acetylation may be increased by Martinostat, thus implicating potential HDAC6 binding at therapeutic-relevant concentrations.

Our imaging data reveal that in vivo HDAC expression is higher in cortical gray matter than white matter, which is confirmed for HDAC2 and HDAC3 by post mortem human tissue analysis (Fig. 3). We hypothesize that ex vivo HDAC1 data may be strongly upregulated as a consequence of tissue death and may not reflect physiological expression levels (23, 24). The imaging bias toward gray matter [¹¹C]Martinostat uptake suggests that HDAC1 levels may indeed be over-estimated from ex vivo samples. Martinostat binds HDACs strongly in the gray matter but more weakly in the white matter as indicated by autoradiography (Fig. S14). As expected, the overall imaging signal appears to relate to both protein density and differences in target affinity with density in this case being the more pronounced discriminator. We postulate that HDAC complexes in each tissue type, such as gray vs. white matter, may impact the selectivity of Martinostat and other HDAC inhibitors, including those currently used as FDA-approved drugs. HDAC complex-directed selectivity of HDAC inhibitors has been previously shown through chemoproteomic approaches (25, 26), and additional work will be required to elucidate the HDAC complexes most represented by the [¹¹C]Martinostat signal.

Beyond regional differences in HDAC distribution, the most striking observation was the consistency of [¹¹C]Martinostat binding patterns between individual subjects (Fig. 2). As epigenetic machinery, and thus HDAC expression, is a highly dynamic process, we did not fully expect a spatially conserved pattern of HDAC expression between individuals. This result not only suggests that HDAC expression is tightly regulated and may represent a state function, but also reiterates the importance of localized levels of HDACs as they directly relate to gene transcription (1). We postulate that regional [¹¹C]Martinostat uptake differences between healthy and diseased individuals will be detectable given the conversed baseline expression that we have measured. The use of [¹¹C]Martinostat imaging may eventually enable precision medicine approaches for disease stratification and treatment based on epigenetic aberrations in the human brain.

Since we envision and will apply [¹¹C]Martinostat to measure HDAC expression in patient populations, it is critical that the outcome measurements are reliable, reproducible, and non-invasive. As shown in the current study, [¹¹C]Martinostat binding can be reliably quantified with and without arterial blood data, and the two measurements are positively correlated (Fig. 2B). The image-based SUV_60-90min has less intersubject variability (11.2%–19.2% across brain regions) than the blood data derived V_T values (22.0%–39.2%) and in preliminary test/retest scans (3 hours apart) in the same individual has less than 3% variability (Fig. S4). These results support the use of SUV_60-90min in future studies to eliminate arterial blood sampling when patient enrollment would be limited by the invasiveness and risk of this procedure. Perhaps as important, PET studies with [¹¹C]Martinostat may be sufficiently powered with a smaller sample size when SUV_60-90min is chosen as the outcome measurement instead of V_T. However, validation studies may be required to evaluate whether SUVs are appropriate surrogates for V_T values in different patient populations.

Since HDACs generally inhibit gene expression, brain regions with low levels of [¹¹C]Martinostat uptake from our imaging data, such as the hippocampus, may be more transcriptionally active allowing for dynamic chromatin-mediated plasticity. In fact, this hypothesis is supported by preclinical in vivo rodent studies showing that i) deficient hippocampal histone acetylation at learning/memory-associated gene loci is associated with aging-related memory impairment (21), ii) HDAC inhibitors restore hippocampal histone acetylation, gene expression, and behavior in aging, neurodegeneration, and mood disorder models (9, 11, 21), and iii) overexpression of HDAC2 in the hippocampus decreases neuronal plasticity and memory formation, while HDAC inhibitors enhance neuronal plasticity by increasing glucose utilization, dendritic spine density, and synaptic transmission (3, 6, 9, 27). As hippocampal HDAC2 overexpression has been found in postmortem brain tissue from AD patients (6), [¹¹C]Martinostat PET imaging holds great potential for detecting aberrant hippocampal HDAC expression and assessing novel HDAC therapeutics in AD patients.

To begin to attribute HDAC imaging with [¹¹C]Martinostat to human gene regulation in the brain, we compared mRNA transcript level changes elicited by pharmacologically relevant doses of Martinostat in human stem cell-derived neural progenitor cells. The experiment was designed to detect differences in gene expression known to be under the control of HDAC-mediated processes (e.g. H3K9 and H4K12 acetylation) (3, 6, 21), which we confirmed were regulated by Martinostat (Fig. 5A). Memory and neuroplasticity-related genes BDNF, SYT1, and SYP were positively regulated by 2.5μM Martinostat treatment (Fig. 5B). We benchmarked the level of mRNA transcript increases to the well-characterized HDAC inhibitor, SAHA. At a fourth of the concentration of SAHA (2.5 vs 10 μM), Martinostat was able to increase BDNF and SYP to an equivalent or greater level (approximately 20-fold and 10-fold, respectively). These data suggest that regions where [¹¹C]Martinostat binding in the human brain is lowest, the levels of HDAC-regulated genes, such as BDNF, are greatest. These results are consistent with our hypothesis that HDAC levels measured by [¹¹C]Martinostat will be inversely correlated with chromatin-mediated plasticity. Indeed, previous studies have consistently shown the hippocampus, which has the lowest [¹¹C]Martinostat uptake in the gray matter (Fig. 2C), to be enriched in BDNF(16, 28–31).

Besides genes implicated in memory and neuroplasticity, Martinostat enhanced the mRNA expression of GRN encoding the glycoprotein progranulin, mutations in which are a major cause of autosomal dominant frontotemporal lobar degeneration (Fig 5B) (13). Since progranulin-deficient frontotemporal lobar degeneration may conceivably be treated by elevating progranulin levels, the demonstration here that Martinostat treatment increases GRN mRNA levels supports the value of HDAC-targeted epigenetic therapies as a disease-modifying treatment for this type of dementia. Moreover, since HDAC inhibitors are the subject of current clinical investigation for frontotemporal lobar degeneration, measuring HDAC levels in human brain with [¹¹C]Martinostat imaging may provide a critically needed tool for determining optimal doses of therapeutics and for patient stratification should levels of HDACs change in the disease state.

In conclusion, the first-in-human epigenetic imaging study herein reveals that HDACs are highly expressed throughout the healthy brain with a conserved regional distribution. Our study is the first to show tissue-specific variations in HDAC inhibitor binding, which we postulate is due to differences in HDAC complex identities in those tissues/regions. Together, our human neuroimaging and biochemical experiments provide a critical foundation for how to accomplish HDAC inhibition in the CNS as a therapy for human brain disorders – a remarkably powerful but yet to be realized concept.

Materials and Methods

Experimental Design

Our main research objective was to quantitate in vivo regional HDAC expression in the healthy human brain, using [¹¹C]Martinostat PET. As a first-in-human PET imaging study, a cohort of 8 individuals was included to evaluate intra-subject and inter-subject variability of [¹¹C]Martinostat uptake. These are critical information for appropriate power calculations when designing future studies. No data from the 8 subjects were excluded as outliers for [¹¹C]Martinostat uptake values. Regional volume of distribution (V_t,) and standard uptake values (SUVs) were the image-based endpoints assessed (see section Image Analysis). We furthered our imaging findings through ex vivo biochemistry to ascribe HDAC subtype selectivity to Martinostat using human and non-human primate brain tissue. Thermal shift and HDAC expression level assays included 3 biological replicates with lysates pooled from 2–3 human donors per replicate. Nuclear density and autoradiographic assays included a minimum of 4 non-human primate brain slices per region from 1 Papio anubis. We also furthered our imaging finding through in vitro biochemistry to determine the effects of Martinostat treatment on substrate acetylation and gene expression levels. Acetylation and mRNA profiling assays included 3 biological replicates of human neural progenitor cells. Imaging and biochemical studies were not blinded.

Participants

Eight participants (4 females and 4 males, mean age ± SD: 28.6 ± 7.6 years) were included in this study (eIND #123154). Participants were healthy volunteers with no history of hepatic, renal, neurological or psychiatric disease and were not taking any prescription medication, as evaluated by medical examinations. Participants had not smoked tobacco products within the past five years and were not using any illicit drugs, as assessed by a urine drug test (Discover 12 Panel Test Card, American Screening Corp., Shreveport LA). Additionally, a serum pregnancy test (Sure-Vue serum hCG STAT, Fisher Healthcare, Houston TX) was performed for female participants to ensure no pregnancy at the time of the scan. Participants provided written informed consent to take part in the study, which was approved by the Institutional Review Board and the Radioactive Drug Research Committee at Massachusetts General Hospital. Volunteers were compensated for their participation in the study.

Radiosynthesis of [¹¹C]Martinostat

[¹¹C]CO₂ (1.0–1.2 Ci) was obtained via the ¹⁴N (p, α) ¹¹C reaction with 11 MeV protons (Siemens Eclipse cyclotron), and trapped in a TRACERlab FX-MeI synthesizer (General Electric). [¹¹C]CH₄ was obtained by the reduction of [¹¹C]CO₂ in the presence of Ni/hydrogen at 350°C and recirculated through an oven containing I₂ to produce ¹¹CH₃I. [¹¹C]Martinostat was synthesized using 2 mg of the desmethyl precursor in 300 μL of DMSO in a TRACERlab FX-M synthesizer. [¹¹C]MeI was introduced to the vessel, which was then sealed and heated to 100°C for 5 min. At the end of the reaction time, the solution was diluted with 1.2 mL of water and loaded onto a semi-preparative HPLC (Agilent Eclipse XDB-18 column, 5μm, 9.4x250 mm. PN: 990967-202). Purification using this column was achieved with 40% 0.01M NaOH/60% EtOH HPLC as the mobile phase (2 mL/min, 254 nm). [¹¹C]Martinostat injection was manufactured under cGMP guidelines for intravenous administration. [¹¹C]Martinostat was collected from the HPLC into a sterile and pyrogen-free glass bulb that was pre-charged with sterile water (14.3 mL), saline (0.7 mL, 23.4% sodium chloride) and 2N HCl (9 μL). The average radiochemical yield was 1–2% at the end of synthesis (EOS) (non-decay corrected to trapped [¹¹C]CH₃I). Chemical and radiochemical purities were ≥ 95 %.

Specific radioactivity (SA) of [¹¹C]Martinostat was calculated from an analytical HPLC sample of 100 μL, employing an Agilent Eclipse XDB-CN column (5 μm, 4.6x150 mm. PN: 993967-905). Eluents: A = MeCN and B = 0.1% TFA (v/v) in water; gradient: 0–7 min, 10–30% A; 7–8 min, 30% A; 2 mL/min) with an UVVis detector (280 nm) followed by a flow-through gamma detector connected in series. A calibration curve of known mass quantity versus HPLC peak area (254 nm) was used to calculate the mass concentration of the sample. The specific activity was 1.3 ± 0.7 mCi/nmol) at the time of injection (TOI).

MR/PET Imaging

Participants had no MRI or PET contraindications to safely undergo brain imaging. An arterial (A-line) was placed in the radial artery of one arm, and an intravenous (IV) catheter was placed in the antecubital vein of the other arm. A licensed nuclear medicine technologist administered [¹¹C]Martinostat into the IV as a manual bolus, and an experienced nurse practitioner drew blood samples from the A-line during the scan to determine plasma radioactivity and radioactive metabolites. Participants were instructed to remain still for the total duration of each scan.

PET and MRI images were acquired on a 3T Siemens TIM-Trio with a BrainPET insert (Siemens, Erlangen, Germany). A PET-compatible CP transmit coil and an 8-channel receive array coil were used for MRI data acquisition. A high-resolution anatomical scan using multi-echo MPRAGE sequence (TR = 2530ms, TE1/TE2/TE3/TE4 = 1.64/3.49/5.35/7.21 ms, TI = 1200 ms, flip angle =7°, and 1 mm isotropic resolution) was acquired.

Dynamic PET image acquisition was initiated concomitant with the start of IV bolus injection of ~5 mCi (4.8 ± 0.4 mCi for the 8 scans) of [¹¹C]Martinostat to the subject. PET data were acquired for 90 min, stored in list mode format and binned into 28 frames of progressively longer duration (10 sec x 6, 20 sec x 3, 30 sec x 2, 60 sec x 1, 120 sec x 1, 180 sec x 1, 300 sec x 8, 600 sec x 5). The corresponding images were reconstructed using the 3D OP-OSEM algorithm with detector efficiency, decay, dead time, attenuation, and scatter corrections applied. The attenuation correction map was derived using an SPM-based, pseudo-CT method (32), which combines segmentation and atlas-based approaches. Simultaneously collected MR sequences consisting of an EPI read-out were used to measure subject motion during the scan and a MR-based motion-correction was applied to the PET data (33). The final PET images were reconstructed into 153 slices with 256 × 256 pixels and a 1.25-mm isotropic voxel size, in the units of radioactivity concentrations (Bq/mL) and standardized uptake values (SUV; SUV = mean radioactivity/injected dose/weight). Three subjects completed a second PET scan, which was accomplished three hours after the first scan on the same day, using identical imaging methods.

Image Analysis

Dynamic PET data were motion corrected to a late time point image (frame 20; 39–44 min post-radiotracer injection) of the time series using rigid body linear registration (6 degrees of freedom) implemented in FSL (MCFLIRT) (34). A PET mean image from the motion-corrected time-series was calculated for each subject and registered and resampled to the subject’s T1-weighted structural scan (MPRAGE) using spmregister from FreeSurfer (http://surfer.nmr.mgh.harvard.edu) (35). The PET mean image was further registered to the MNI space using a linear (FLIRT) and a nonlinear (FNIRT) algorithm implemented in FSL (FMRIB’s Software Library, http://www.fmrib.ox.ac.uk/fsl) (36). Finally, dynamic PET images (in both radioactivity concentration and SUV units) were normalized to the MNI space, using a combined transformation matrix derived from the PET mean image, for further analysis.

Kinetic modeling was performed using PMOD 3.4 (PMOD Technologies Ltd., Zurich, Switzerland). Twenty-eight volumes of interest (VOIs) were defined according to the Automated Anatomical Labeling (AAL) human brain atlas distributed with PMOD (37). A two-tissue compartmental model was applied to the regional time-activity curves extracted from the VOIs and using the metabolite-corrected arterial plasma as input function to derive volume of distribution (V_T) and micro-parameters describing the pharmacokinetics of the radiotracer (Supplementary Table. 3). Minimum scan duration requires for stable V_T values estimation was also evaluated (Figure. S2). An averaged SUV image (SUV_60-90min) was calculated from 60 to 90 min post-radiotracer injection. Regional cortical V_T and SUV_60-90min values were combined for cortical lobes using a weighted average to reduce the total number of VOIs (resulting in a total of 14 VOIs). Voxel-wise, group mean and standard deviation maps of the SUV_60-90min were calculated and overlaid on a MNI152 template brain after spatial smoothing with a 6-mm FWHM Gaussian filter (Fig. 2). In addition, SUV_60-90min were normalized to individual subject’s white matter SUV_60-90min (SUVR_60-90min) to evaluate intersubject variability for different VOIs (Fig. 2C).

Arterial Plasma and Metabolite Analysis

Blood samples were drawn from the arterial line at 10 sec intervals beginning at the start of the scan for 3 min (~2 mL each), followed by additional samples at 5, 10, 20, 30, 45, 60, and 80 min (~6 mL each) post-radiotracer injection for plasma and metabolite analyses. Blood analysis was carried out using methods published previously (18). In short, the collected samples were centrifuged to obtain plasma, which was then removed (200μL for samples collected during the first 3min; 600μL for all later samples) and placed in an automatic gamma counter that was cross-calibrated with the PET scanner. The analysis of radiolabeled metabolites was conducted on a custom automated robot, fitted with Phenomenex Strata-X 500 mg SPE cartridges that were primed with ethanol (2 mL) and deionized water (20 mL). Each sample was counted in a WIZARD2 Automatic Gamma Counter to determine the presence of radiolabeled metabolites. Final total plasma radioactivity was interpolated linearly and corrected for the fraction of radiometabolites. The metabolite-corrected plasma activity curve was used as the arterial input function for kinetic modeling.

Human Tissue Samples

Postmortem frozen human brain tissue was obtained from the NIH NeuroBioBank; specifically, tissue was obtained from the Harvard Brain Tissue Resource Center, the University of Miami Brain Endowment Bank, the Human Brain and Spinal Fluid Resource Center, and Brain Tissue Donation Program at the University of Pittsburgh Medical Center. For all donors, informed consent was obtained from next-of-kin. Donor brains had a neuropathology diagnosis of normal. Sample details are provided in Supplementary Table 4.

Tissue Lysate Preparation

Human brain tissue from the superior frontal gyrus (SFG), corpus callosum (CC), dorsal lateral prefrontal cortex (DLFPC), hippocampus (Hipp), and anterior cingulate (Ant Cing) was suspended in lysis buffer containing PBS, 0.15% NP-40, and a protease inhibitor cocktail (Roche 04693159001) at 50 mg/mL. Tissue was homogenized for 120 sec with an electric pestle homogenizer and sonicated on ice at 50% power (Fisher Scientific Model CL-18) for thirty 1 s pulses. Lysate was plunged through a syringe 10 times to shear chromosomal DNA. Lysate was incubated with rotation at 4°C for 20 min and centrifuged 18,000x g at 4°C for 20 min. Supernatant was collected and total protein concentration was measured using a BCA protein assay (Pierce 23227). For SFG and CC replicates, equivalent amounts of total protein were pooled from 2 donors per replicate. For DLFPC, Hipp, and Ant Cing replicates, equivalent amounts of total protein were pooled from 3 donors per replicate.

HDAC Expression Levels

Known concentrations of recombinant HDAC enzymes (Reaction Biology Corp KDA-21-365, KDA-21-277, and KDA-21-213; Abcam ab82071) were diluted in 2-fold increments and compared to 12 μg total protein from human brain lysates (n=3 replicate pools per region) through western blotting. Notably, recombinant HDAC2 contained a GST tag, which increased its detected size. Immunoreactive band intensity was quantified with ImageJ (Image Processing and Analysis in Java, NIH, Bethesda, MD)(38). HDAC band intensity values were normalized to GAPDH intensity values to control for equal loading. Standard curves were calculated for each recombinant HDAC isoform with GraphPad Prism software and the concentrations of HDACs per lane of lysate were determined. Data are reported in picomoles/mg of total protein.

Thermal Shift Assay

Lysates were prepared from human SFG, CC, DLPFC, Hipp, and Ant Cing (n=3 replicate pools per region). Lysates (~1.5–2.0 mg/mL) were divided into 115 μL aliquots and treated with 1% DMSO containing gradients of cold (non-radiolabeled) Martinostat (0, 0.0032, 0.016, 0.080, 0.40, 2.0, 10 μM) and (0, 0.16, 0.80, 4.0, 20, 100 μM). Aliquots were incubated for 30 min at room temperature with rotation. Aliquots were sub-divided into 50 μL aliquots, heated to 55°C or 60°C for 3 min, and cooled to 25°C for 3 min. Note, input aliquots contained 1% DMSO without Martinostat and were kept at room temperature. Aliquots were immediately centrifuged 18,000x g at 4°C for 20 min. Supernatants were collected, 3X SDS-PAGE loading buffer containing 125 mM DTT was added, and samples were heated to 70°C for 10 min. Note, based on empirically determined melting temperatures, 55°C samples were used for HDAC 1/2/6/8 western blots and 60°C samples were used for HDAC3 western blots.

Band Intensity Analysis

Western blot images were converted from Image Lab 5.2.1 (.scn) files to 600 dpi Tif files. Tif files were opened in Image J. Images were converted to 8-bit and background was subtracted with a rolling ball radius of 50.0 pixels. Images were inverted and mean band intensity was quantified with the measurement tool.

Thermal Stabilization Analysis

Mean immunoreactive band intensities from each replicate were quantified with ImageJ. Thermal stabilization of HDAC isoforms conferred by incubation with a concentration ‘x’ Martinostat (μM) was calculated as a function of immunoreactive band intensity ‘IR’ using the following equation:

$$\%\mathit{stabilization}=\frac{I{R}_{x[\mathrm{\mu }M]}-I{R}_{0[\mathrm{\mu }M]}}{I{R}_{\mathit{input}}-I{R}_{0[\mathrm{\mu }M]}}$$

The average % stabilization (n=3 replicate pools per region) of each HDAC isoform was plotted as a heat map, using a 2-color scale (minimum= lowest value, white; maximum= highest value, red). The individual replicate % stabilizations of each HDAC isoform were also plotted as heat maps. SFG and CC values are displayed on the same color scale (Figs. 4A, S8, S9, S10). SFG, DLFPC, Hipp, and Ant Cing values are displayed on the same color scale (Figs 4B, S11, S12, S13).

Western Blotting

Proteins were separated on Criterion Stain-Free 4–20% gels (Biorad 567-8095) at 200V for 50 min. Proteins were transferred to Low Fluorescence PVDF membrane (Biorad 162-0264) at 0.14 amps for 60 min. Gels and membranes were imaged with a Chemidoc XRS system (Biorad 170-8265) for quality control purposes. Membranes were processed as follows: blocked in Tris buffered saline + Tween 20 (TBST, 0.1% Tween 20) containing 5% milk (Biorad 170-6404) at room temperature for 1 hr (note-rest of the protocol performed at room temperature), washed in TBST, incubated with primary antibodies in TBST containing 1% milk (HDAC1: Thermo Fisher PA1-860 1:5000, HDAC2: Abcam ab124974 1:5000, HDAC3: Abcam ab32369 1:5000, HDAC6: Santa Cruz sc11420 1:5000, HDAC8: Abcam ab187139 1:5000, GAPDH: Abcam ab8245 1:50000, acetyl histone H3 lysine 9: EMD Millipore 06-942-S 1:4000, acetyl histone H4 lysine 12: EMD Millipore 07-595 1:4000) for 1 hr, washed in TBST, incubated with secondary antibody in TBST containing 1% milk (anti-rabbit-HRP: Cell Signaling #7074S 1:5000, anti-mouse-HRP: Cell Signaling #7076S 1:5000) for 1 hr, washed in TBST, developed with ECL prime western blotting detection reagent (GE RPN2232), and visualized with a Chemidoc XRS system.

Nuclear Density

Frozen baboon (Papio anubis n=1) brain tissue spanning regions SFG to CC was sectioned (10 microns) with a cryostat (Thermo Scientific HM550) directly onto ColorFrost Plus microscope slides (Fisher Scientific 12-550-18). Sections (n=4) were fixed with 4% paraformaldehyde in phosphate buffer (Boston Bioproducts, BM-698) for 10 min and washed in phosphate buffered saline (PBS) (Fisher Scientific, bp399-1) two times for 5 min. Tissue was permeabilized (PBS + 0.2% TritonX-100) for 10 min and washed quickly in PBS. Mounting media containing DAPI (Life Technologies, P-36931) was applied and dried overnight. Images (SFG n=16, CC n=16) were taken at 20X magnification with an epi-fluorescence microscope (Nikon Eclipse E400, RTke Diagnostic Instruments Inc., Spot Software version 4.6). Images were stacked in ImageJ with equivalent thresholds for brightness, particle size, and circularity. Nuclear density (count of nuclei per field of view), nuclear size (area per individual nuclei), and total nuclear area (nuclear density x nuclear size) were measured. Replicate mean values and standard deviation are reported.

Autoradiography

Frozen baboon (Papio anubis n=1) brain tissue spanning gray matter and white matter was sectioned (10 microns) with a cryostat (Thermo Scientific HM550) directly onto ColorFrost Plus microscope slides (Fisher Scientific 12-550-18). Sections were fixed with 4% paraformaldehyde in phosphate buffer (Boston Bioproducts, BM-698) containing 2% ethanol at 4°C for 30 min, and washed in 10 mM TrisHCl pH 7.5 at 4°C. Slices were co-incubated with ~100 μCi [¹¹C]Martinostat, and either 0 μM or 2 μM of cold (non-radiolabeled) Martinostat in 5% DMSO (n=22 0 μM slices; n=10 2 μM slices) at room temperature for 12 min. Slices were washed in 10 mM TrisHCl pH 7.5 for 10 min and dried under vacuum at 25°C for 30 min. Slices were incubated with multi-sensitive screens (Perkin Elmer, 7001723) for 60 min and imaged with a Cyclone Plus Storage Phosphor system (Perkin Elmer). Images were smoothed with ImageJ using a Gaussian blur (3.0 radius) and colored using the Royal lookup table with equivalent thresholds for brightness. Raw intensity values from gray and white matter were quantified with the ImageJ measurement tool, with replicate mean values and standard deviation reported.

Human Neural Progenitor Cell Culture

Human iPSC-derived neural progenitor cells from a healthy control subject fibroblast cell line GM08330 (Coriell Institute for Medical Research, Camden, NJ) were generated as described in Sheridan et al. 2011 (39) and were cultured as previously described Zhao et al. 2012 (40). Media used for cell culture was Neural Sphere (NS) media composed of 70% DMEM (Dulbecco’s modified Eagle’s Medium, Gibco #11995), 30% Ham’s F12 with L-glutamine (Modified Gellgro/Mediatech #10-080-CV), 1X penicillin/streptomycin, and 1X B27 Supplement (50X, Gibco #17504-044). Just before use, the media was supplemented with EGF (20 ng/mL, Epidermal Growth Factor, Sigma, prepared as 1000X stock in DMEM), bFGF (20 ng/mL, basic Fibroblast Growth Factor, Sigma, prepared as 1000X stock in PBS), and heparin (5 μg/mL, Sigma, prepared as 1000X stock in Ham’s F12). For compound treatments (0, 0.5, 2.5, 5.0 μM Martinostat in DMSO and 10 μM SAHA (Medicilon) in DMSO), neural progenitor cells were plated at 0.5×10⁶ cells per well in a poly-ornithine/laminin double-coated 6-well plastic, sterile plate (Falcon #353046) as previously described Zhao et al. 2012 (40), in 2 mL NS media with the above growth factors. The cells were incubated at 37°C for 48 hrs to grow to 95–100% confluency. The media was aspirated and the cells were treated with compound in fresh NS media at 37°C for 24 hrs.

Histone acetylation changes in human neural progenitors with Martinostat and SAHA

Cell pellets from human neural progenitor cells were lysed in radio immunoprecipitation assay (RIPA) buffer (Boston BioProducts #BP-115) with EDTA-free protease inhibitors (Sigma #4693159001) and rocked at 4°C for 30 min. The lysate was centrifuged at 14,000 rpm at 4°C for 25 min and the supernatant was collected. Protein quantification was determined by the Pierce bicinchoninic assay (Thermo Scientific #23227). Lysates were diluted to 800 ng/μL in RIPA buffer and stored at −80°C until ready for use. H3K9 and H4K12 acetylation levels were measured by western blotting. Mean immunoreactive band intensities from each replicate were quantified with ImageJ. Acetylation band intensity values were normalized to GAPDH intensity values to control for equal loading. Replicate mean values and standard deviation are reported.

Gene expression changes in human neural progenitors with Martinostat and SAHA

RNA was generated from each well of a 6-well treated plate of human neural progenitor cells. After media aspiration, wells were washed with 1 mL of PBS and then lysed with 1 mL TRIzol (Thermo Fisher). The cells were incubated at room temperature for 5 min and RNA was extracted with the DirectZol RNA MiniPrep Kit (Zymo Research #R2052). RNA was stored at −80°C until ready for use. cDNA was generated from 1200ng RNA using the High Capacity cDNA Synthesis Kit with RNase inhibitor (Thermo Fisher #4368814). cDNA was used immediately or stored at −20°C until ready for use. Before use, cDNA was diluted 1:4 with water. qPCR was conducted on the Roche 480 Light Cycler in a 384-well plate. Into each well 5 μL of TaqMan Gene Expression Master Mix (Thermo Scientific #4369510), 0.5 μL of 20X commercial TaqMan primer probe for each gene described (Thermo Scientific: GRN, Hs00963707_g1; FXN, Hs00175940_m1; BDNF, Hs00380947_m1; EGR1, Hs00152928_m1; CDK5, Hs00358991_g1; SYT1, Hs00194572_m1; SYP, Hs00300531_m1), 0.5 μL of DNase/RNase free water, and 4 μL of above diluted cDNA was added. C_T values (n=9 technical replicates) were calculated with LightCycler 480 software. Results were normalized to GAPDH. Replicate mean values and standard error of the mean are reported.

Statistical Analyses

Statistical tests were performed using GraphPad Prism (Prism6, GraphPad Software Inc., La Jolla, CA, USA). For PET imaging analysis, a non-parametric Friedman test (α=0.05 with Dunn’s multiple comparisons correction) was carried out to compare SUV_60-90min between brain regions (Fig 2). A Pearson correlation analysis was performed between V_T and SUV_60-90min values for the 14 VOIs (Fig. 2) to evaluate whether an image-based outcome measurement (SUV_60-90min) is an appropriate surrogate to that estimated with the full kinetic modeling data (V_T). Differences in HDAC expression levels as well as differences in nuclear density, size, and total area between SFG and CC were evaluated with an unpaired t-test (Figs. 3, S6A, and S7B). Differences in HDAC expression levels between DLFPC, Hipp, and Ant Cing were evaluated with an ordinary one-way ANOVA (α=0.05 with Tukey’s multiple comparisons correction) (Fig. S6B). Differences in histone acetylation and gene expression levels as compared to DMSO were evaluated with a repeated measures two-way ANOVA (α=0.05 with Dunnett’s multiple comparisons correction) (Figs. 5 and S15). In autoradiographic assays, differences between [¹¹C]Martinostat baseline and blocking intensity values, in gray matter and white matter, were evaluated with an ordinary two-way ANOVA (α=0.05 with Sidak’s multiple comparisons correction (Fig. S14B).

Supplementary Material

Supp

Table S1. Biometric information for PET imaging participants.

Table S2. Goodness of fit of dynamic PET data to compartmental models. One- and two-tissue compartmental models (1TCM and 2TCM) were evaluated with Akaike information criteria (AIC) and Model selection criteria (MSC).

Table S3. Kinetic rate constants and regional distribution volume (V_T) for [¹¹C]Martinostat (n=6) were estimated with a two-tissue compartmental model (mean ± SD). COV(%) represents the standard error of the fit (group mean).

Table S4. Sample information for postmortem human brain tissue. All donors had a neuropathology diagnosis of normal.

Fig. S1. Time activity curves (TACs) and compartmental model fitting (two-tissue compartmental model) results for superior frontal cortex and white matter.

Fig. S2. Stability of outcome measurement (V_T) as a function of scan duration.

Fig. S3. Regional SUV_60-90min from all volumes of interests analyzed.

Fig. S4. Same day test-retest reproducibility of [¹¹C]Martinostat.

Fig. S5. Assessment of HDAC expression levels in postmortem human brain tissue.

Fig. S6. HDAC expression levels in postmortem human brain tissue.

Fig. S7. Nuclear density and nuclear size are region-specific in postmortem baboon brain tissue.

Fig. S8. Martinostat thermal shift assay in human superior frontal gyrus and corpus callosum biological replicate 1.

Fig. S9. Martinostat thermal shift assay in human superior frontal gyrus and corpus callosum biological replicate 2.

Fig. S10. Martinostat thermal shift assay in human superior frontal gyrus and corpus callosum biological replicate 3.

Fig. S11. Martinostat thermal shift assay across human gray matter biological replicate 1.

Fig. S12. Martinostat thermal shift assay across human gray matter biological replicate 2.

Fig. S13. Martinostat thermal shift assay across human gray matter biological replicate 3.

Fig. S14. [¹¹C]Martinostat binding is non-specific in postmortem baboon white matter.

Fig. S15. Assessment of Martinostat treatment in human neural progenitor cells.