Ferulic acid amide derivatives with varying inhibition of amyloid-β oligomerization and fibrillization

Article abstract of DOI:10.1016/j.bmc.2021.116247

Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized, in part, by the misfolding, oligomerization and fibrillization of amyloid-β (Aβ). Evidence suggests that the mechanisms underpinning Aβ oligomerization and subsequent fibrillization are distinct, and may therefore require equally distinct therapeutic approaches. Prior studies have suggested that amide derivatives of ferulic acid, a natural polyphenol, may combat multiple AD pathologies, though its impact on Aβ aggregation is controversial. We designed and synthesized a systematic library of amide derivatives of ferulic acid and evaluated their anti-oligomeric and anti-fibrillary capacities independently. Azetidine tethered, triphenyl derivatives were the most potent anti-oligomeric agents (compound 2i: IC₅₀ = 1.8 μM ± 0.73 μM); notably these were only modest anti-fibrillary agents (20.57% inhibition of fibrillization), and exemplify the poor correlation between anti-oligomeric/fibrillary activities. These data were subsequently codified in an in silico QSAR model, which yielded a strong predictive model of anti-Aβ oligomeric activity (κ = 0.919 for test set; κ = 0.737 for validation set).

Full text of DOI:10.1016/j.bmc.2021.116247

Journal Pre-proofs
Ferulic Acid Amide Derivatives with Varying Inhibition of Amyloid-β Oligo-
merization and Fibrillization
Igri Kolaj, Yanfei Wang, Kailin Ye, Autumn Meek, S. Imindu Liyanage,
Clarissa Santos, Donald F. Weaver
PII:
S0968-0896(21)00255-8
https://doi.org/10.1016/j.bmc.2021.116247
BMC 116247
DOI:
Reference:
To appear in:
Bioorganic & Medicinal Chemistry
Received Date:
Revised Date:
Accepted Date:
25 February 2021
19 May 2021
28 May 2021
Please cite this article as: I. Kolaj, Y. Wang, K. Ye, A. Meek, S. Imindu Liyanage, C. Santos, D.F. Weaver,
Ferulic Acid Amide Derivatives with Varying Inhibition of Amyloid-β Oligomerization and Fibrillization,
Bioorganic & Medicinal Chemistry (2021), doi: https://doi.org/10.1016/j.bmc.2021.116247
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
© 2021 Published by Elsevier Ltd.

Manuscript File
Click here to view linked References
Ferulic Acid Amide Derivatives with Varying Inhibition
of Amyloid-β Oligomerization and Fibrillization
Igri Kolaja,b, Yanfei Wangb, Kailin Yeb,c, Autumn Meekb, S. Imindu Liyanageb, Clarissa Santosb,
Donald F. Weavera,b,c,
*
a Department of Chemistry, University of Toronto
80 St. George Street, Toronto, ON M5S 3H6, Canada
b Krembil Research Institute, University Health Network,
Krembil Discovery Tower, 60 Leonard Avenue, 4KD477, Toronto, ON M5T 0S7, Canada
c Department of Pharmaceutical Chemistry, University of Toronto
144 College Street, Toronto, ON M5S 3M2, Canada
Corresponding author:
Donald F. Weaver,
donald.weaver@utoronto.ca
60 Leonard Avenue, 4KD477, Toronto, ON M5T 0S7, Canada
Other authors:
Igri Kolaj: igri.kolaj@gmail.com; Yanfei Wang: yanfei.trv@gmail.com; Kailin Ye:
kailin.ye@mail.utoronto.ca; Autumn Meek: autumn.meek@gmail.com; S. Imindu Liyanage:
imindu.liyanage@mail.utoronto.ca; Clarissa Santos: clarissa.santos@uhnresearch.ca
Declarations of interest: None.
Funding: This research did not receive any specific grant from funding agencies in the public,
commercial, or not-for-profit sectors.
Key Words: Alzheimer’s disease; Aβ oligomerization; Aβ fibrillization; ferulic acid; curcumin;
amide derivatives.
1

Abstract
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by
the misfolding, oligomerization and fibrillization of amyloid-β (Aβ). Evidence suggests that the
mechanisms underpinning Aβ oligomerization and subsequent fibrillization are distinct, and may
therefore require equally distinct therapeutic approaches. Prior studies have suggested that amide
derivatives of ferulic acid, a natural polyphenol, may combat multiple AD pathologies, though its
impact on Aβ aggregation is controversial. We designed and synthesized a systematic library of
amide derivatives of ferulic acid and evaluated their anti-oligomeric and anti-fibrillary capacities
independently. Azetidine tethered, triphenyl derivatives were the most potent anti-oligomeric
agents (compound 2i: IC50 = 1.8µM ± 0.73µM); notably these were only modest anti-fibrillary
agents (20.57% inhibition of fibrillization), and exemplify the poor correlation between anti-
oligomeric/fibrillary activities. These data were subsequently codified in an in silico QSAR
model, which yielded a strong predictive model of anti-Aβ oligomeric activity (κ = 0.919 for test
set; κ = 0.737 for validation set).
Word Count: 156
2

Graphical Abstract
3

1. Introduction
Aggregation of misfolded proteins is central to the leading mechanistic hypotheses for major
neurodegenerative disorders, including Parkinson’s disease, prion encephalopathies and
Alzheimer’s disease (AD). In AD, these aggregates induce neuroinflammation, neuronal death
and cerebral atrophy – manifesting as progressive declines in cognition, memory and executive
function.1 Despite exhaustive study, neither a definitive therapy nor a clear therapeutic target
have been discerned; rather an array of diverse and interconnected pathologies, ranging from
innate immunity to mitochondrial dysfunction, has been suggested as causative.1 However, the
misfolding and aggregation of neural proteins, be it symptomatic or etiological, remain the best
studied AD pathology and a continued focus of major therapeutic strategies.
AD is characterized by two dominant protein pathologies: the accumulation of neurofibrillary
tangles composed of hyperphosphorylated filaments of protein tau, and the formation of
extracellular amyloid plaques, composed of amyloid-β (Aβ) peptides.2, 3 Previously, Aβ
aggregation was thought to be the definitive step in the onset and progression of AD.3-7 This
amyloid hypothesis purported that soluble oligomeric species of Aβ triggered a cascade of
downstream events including neuronal damage, immune activation, tau phosphorylation,
microglial activation and cytokine/chemokine release, that ultimately culminate in neuronal
death and the clinical manifestations of AD.8 Though the etiology has since proven more
nuanced, with Aβ likely acting in complex synergies to yield the observed changes, it remains
the best studied target in AD. Agents interfering with the oligomerization and fibrillization
(collectively referred to as “aggregation”) of Aβ have therefore been among the most intensely
examined. To date, numerous small molecules, natural and synthetic, have been proposed with
4

some activity against the aggregation of Aβ. Among these, natural polyphenols, derived from
plants have repeatedly shown promise.9-11
In early work, the polyphenol curcumin, demonstrated notable anti-amyloidogenic activity as
well as antioxidative and anti-inflammatory properties in vitro,12 in vivo,13 and in silico.14,15 In
plants, it is formed from the dimerization of ferulic acid, itself a potent antioxidant,15-18 and a
potential multifunctional agent against AD.19-22 Recent studies have shown that administration of
ferulic acid, either alone or in combination, can rescue neurodegeneration and cognitive decline
in AD,23 as well as other disease indications.24 Prior derivatives of ferulic acid have also
conferred multifaceted protective associations in vitro,25 though their capacity as anti-
aggregative agents is disputed. Further, clinical trials have yet to show a significant effect on
cognition or AD progression, likely attributable to inadequate pharmacokinetic properties and
blood-brain barrier (BBB) permeability. Curcumin may also readily undergo oxidative
degradation within the aqueous physiology of the body, further complicating its utility as a
therapeutic.
More recently recognized biophysical discrepancies between the oligomerization and
fibrillization of Aβ may also be partly culpable.26-28 Though formerly considered synonymous
processes, data now suggest that the mechanisms underpinning the aggregation of Aβ into small
oligomeric species, and the subsequent aggregation into larger fibrils are distinct. Of particular
relevance to therapeutic strategies was the observation that oligomers appear highly neurotoxic,
while fibrils and plaques are likely benign, and may be a protective strategy to sequester harmful
oligomers. Therefore, agents which impede fibrillization, without targeting oligomerization, may
leave highly neurotoxic oligomers unaffected, and may deprive the brain of the ability to
sequester Aβ. Evidence also suggests that Aβ may be able to circumvent the inhibition of
5

oligomerization by forming disordered species that can nucleate and aggregate into larger
plaques.28, 29
To date, studies have relied largely on the inhibition of easily visualized fibrillization (or
protofibrils), rather than oligomerization to determine the efficacy of compounds. Yet the
differing aggregation of oligomers and fibrils suggest that the functional moieties conferring
anti-oligomeric activity may be distinct from those required to prevent fibrillization. The failure
to make a clear distinction, and directly target the inhibition of neurotoxic oligomers, may partly
explain the failure of amyloid-based therapeutic strategies thus far.
Herein, we systematically evaluate the anti-amyloidogenic activities (both anti-fibrillary and
anti-oligomeric) of amide derivatives of ferulic acid. Plant derived amides are highly active
within biological systems. Replacement of carboxylic acids with amides is also known to
improve BBB penetration, while preserving many of the desirable molecular characteristics.
Moreover, amides are amenable to chemical modification, and are therefore a reasonable drug-
design platform. We further observed that amide substitution of ferulic acid’s carboxylate moiety
yielded a 10-fold increase in anti-amyloid activity (Table 1). We assayed these novel amide
derivatives specifically as anti-oligomeric agents against Aβ; relevant compounds were further
assayed for their anti-fibrillary functionality. These results were used in the construction of a
binary QSAR model, which proved capable of accurately predicting the activity of novel
compounds.
6

2. Results and Discussion
2.1 Synthesis of Ferulic Acid Derivatives
A diverse library of amide-based ferulic acid derivatives was constructed. These
derivatives were developed to explore differences in electronic, geometric and physicochemical
properties in influencing the ferulate analogues to inhibit Aβ oligomer and/or fibril formation.
Collectively, this library systematically explores three structural “faces” of ferulic acid: (i) the
carboxylic acid moiety – converted to an amide, (ii) the hydroxyl group on the phenol ring, and
(iii) the rigid, planar backbone between the carboxylic acid (substituted amide) and aromatic
ring.
7

Scheme 1. Synthesis of Ferulic Acid Analogs
Reagents and conditions: (a) malonic acid, pyridine, 110˚C, 14 h; (b), HATU, i-Pr2NEt, amine,
DMF, room temperature (rt), 16− 24 h; (c) LiOH, THF/H2O, 16-24 h; (d) HATU, i-Pr2NEt,
amine, DMF, 35˚C or 50˚C, 16− 24 h.
8

2.2 Inhibition of Aβ Oligomerization
To assess anti-oligomeric activity, an enzyme-linked immunosorbent assay (ELISA),
previously described by Levine et al.,30 was adapted for this study. Unlike assays of fibrillization,
this method directly quantifies the abundance of oligomers (relative to monomers) with high
sensitivity and minimal interference. Specifically, bio-tinylated Aβ42 (bio-Aβ42), was allowed to
oligomerize in the presence of compounds in a decreasing concentration gradient, and the
resulting oligomers were visualized employing a NeutrAvidin/streptavidin protocol. If
compounds were functional, a sigmoidal curve of inhibition would be formed and used to
estimate the concentration required to inhibit 50% oligomerization (IC50). Compounds with no
discernable difference in the abundance of oligomers, or those which did not exhibit any pattern
of inhibition were deemed to have no activity (N.A.).
2.2.1 Carboxylic Acid Derivatives
In initial testing, the dimer of ferulic acid (curcumin; compound 0a) proved substantially
more effective against Aβ oligomerization (IC50 = 3.38µM ± 0.11µM) than monomeric ferulic
acid (compound 0; IC50 = 226.23µM ± 22.26µM). This suggests a relatively large area of
interaction between Aβ monomers, which may best be targeted with sterically bulky
substitutions on the carboxylic acid. In the first screen of derivatives (Table 1), the parent alkyl
amide (modified with a benzyl group at the phenyl hydroxide; section 2.5) was substituted with a
variety of alkyl, aromatic, halide and cyclic substituents.
9

Table 1
Calculated IC50 against Aβ oligomerization for carboxyl–derivatives of ferulic acid amides
bio-Aβ42
IC50 (μM)
bio-Aβ42
IC50 (μM)
Comp. Structure
Comp. Structure
Ferulic Acid
Curcumin
226.23 ±
22.26
3.85 ± 0.11
0
0a
24.60 ±
15.64
27.94 ± 3.35
1
1f
39.27 ± 0.85
10.64 ±
3.47
1a
1fa
34.93 ±
23.38
18.65 ± 2.28
1aa
1fb
119.16 ±
2.65 μM
45.78 ±
17.18
1ab
1ac
1fc
1g
78.07 ±
12.04
N.A.
12.38 ± 4.13
79.87 ± 2.2
N.A.
1b
1h
N.A
N.A.
N.A.
1c
1d
1e
1i
106.82 ±
16.39
1j
1k
65.09 ± 8.61
N.A. denotes no discernable activity; uncertainty denotes standard deviation. All compounds
were assessed in at least three independent samples, with a minimum of three measurements.
10

A simple alkyl side-chain, cyclic or acyclic, proved most effective against
oligomerization (compounds 1: IC50 = 24.60µM ± 15.64µM; 1b: IC50 = 12.38µM ± 22.26µM;
and 1f: IC50 = 27.94µM ± 22.26µM). Polar bonds in the region immediately adjacent to the
amide proved highly ineffective – suggesting interactions among hydrophobic residues in this
space. The carbonyl group of the amide was also critical; when replaced with a thiocarbonyl, all
activity was lost. Since amides and thioamides are isosterically similar, it is possible that
hydrogen bonding between the carbonyl and an associated residue facilitates interaction with Aβ.
However, when the carbonyl was removed entirely (compound 1f: IC50 = 27.94µM ± 3.35µM vs.
1fa: IC50 = 10.64µM ± 3.47µM) the efficacy of the compound rose. Therefore, polar bonds in
this region likely hinder interactions with Aβ, yet the carbonyl oxygen may overcome this via
hydrogen bonding.
A subsequent series of compounds (Tables 2-3), built upon this initial carbonyl/alkyl
spacer, was designed to mimic curcumin’s elongated, π-rich structure (Scheme 1). Longer spans
were therefore incorporated to add surface area to the molecules, and thus maximize exposure to
the wide amyloidogenic residues between Aβ monomers. (S)- and (R)-proline (Table 2) as well
as β-alanine (Table 3) were selected as scaffolds, to match the conformational arrangement of
curcumin. Among these, the proline derivatives proved moderately effective, yet were generally
weaker inhibitors than the initial shorter amides. The presence of a polar amide group in relative
proximity to the ferulic amide may be partly explain the lack of improvement.
11

Table 2
Calculated IC50 against Aβ oligomerization for proline carboxyl–derivatives of ferulic acid
amides
bio-Aβ42
IC50 (μM)
bio-Aβ42
IC50 (μM)
Comp. Structure
Comp. Structure
17.08 ±
4.36
19.24 ± 2.06
2a
2d
18.72 ±
3.74
16.83 ± 1.17
8.59 ± 2.65
2b
2c
2e
2f
17.75 ±
3.63
Uncertainty denotes standard deviation. All compounds were assessed in at least three
independent samples, with a minimum of three measurements.
12

Table 3
Calculated IC50 against Aβ oligomerization for β-alanine carboxyl–derivatives of ferulic acid
amides
bio-Aβ42
IC50 (μM)
bio-Aβ42
IC50 (μM)
Comp. Structure
Comp. Structure
12.78 ± 0.17
21.83 ± 0.74
10.55 ± 0.84
3a
3f
3.65 ± 0.33
3b
3c
3d
3e
3g
3h
3i
4.78 ± 1.67
3.61 ± 0.59
35.67 ±
2.42
1.80 ± 0.73
3.13 ± 1.87
8.17 ± 1.70
54.21±
2.96
4
5b
5c
17.91 ±
17.02
5a
81.56 ± 9.91
N.A. denotes no discernable activity; uncertainty denotes standard deviation. All compounds
were assessed in at least three independent samples, with a minimum of three measurements.
13

Use of a 3-azetidinecarboxylic acid scaffold proved more effective; incorporation of a
benzyl-carbamoyl azetidine yielded significant inhibition (compound 3i; IC50 of 1.8µM ±
0.73µM). This triple-phenyl molecule displays a characteristic bent L-shaped conformation (Fig.
1A), which simulates curcumin’s spatial arrangement (Fig. 1B), yet is elongated at both ends.
The added lateral distance may contribute towards the improved inhibition of Aβ
oligomerization. The conformational constraint of the azetidine moiety also appears to be
critical. Unlike simple alkyls or higher order rings, the strained azetidine minimizes torsional
variation in the backbone, forming a planar link between the ferulic acid and side chain. The
success of the azetidine derivatives over the others suggests this conformational restriction
facilitates better interaction with Aβ.
Further modification, such as varying the azetidine chain’s phenyl ring, did not yield
significant changes in function. Even replacement of the phenyl with a furan (compound 3h) did
not alter the IC50 substantially – suggesting that π interactions broadly aid the inhibition of Aβ
oligomerization in this region. Other major modifications (Table 4), such as incorporation of a
tetrazole or oxazole decreased activity, compared to the parent alkyls. Use of an imidazole tether
was more effective, achieving an IC50 less than 10μM (compounds 11 and 12), but inhibition was
inferior comparable to azetidines. Amongst all tested tethers, the most efficient interaction with
Aβ repeatedly occurred in molecules in which the ferulic amide was attached to a 2 carbon alky
spacer (ideally rigid), then connected to an additional amide. This polar-nonpolar-polar transition
within a localized region appears to facilitate interactions with Aβ.
14

A
B
C
D
Fig 1. A & B: Electron density representation of (A) compound 3i; phenyl azetidine substituted
ferulic acid derivative (IC50 = 1.8µM ± 0.73µM); (B) compound 0; curcumin (IC50 = 3.85μM ±
0.11µM). Red denotes electron dense regions, blue denotes electron deficient regions. C & D:
docking characteristics of (C) compound 3i and (D) curcumin with Aβ.
15

Table 4
Calculated IC50 against Aβ oligomerization for carboxylate–replaced derivatives of ferulic acid
amides
bio-Aβ42
IC50 (μM)
bio-Aβ42
IC50 (μM)
Comp. Structure
Comp. Structure
N.A.
N.A.
6
10
98.46 ±
1.63
7
11
12
7.76 ± 1.78
9.91 ± 1.04
N.A.
8
9
N. A.
N.A. denotes no discernable activity; uncertainty denotes standard deviation. All compounds
were assessed in at least three independent samples, with a minimum of three measurements.
16

2.2.2 Planar Backbone Derivatives
Unlike the characteristics of the amide, the conformation and saturation of the alkene of
ferulic acid is not critical to activity. Saturation and cyclopropanation of the alkene (Table 1:
compounds 1aa, 1fb,1fc) did not yield significant changes in activity - suggesting that the α,β-
unsaturated backbone, and the resulting structural conformation, are not crucial for binding to the
Aβ peptide. Furthermore, the IC50 of compounds 1fc (45.78μM ± 17.18μM), 1aa (34.93μM ±
23.38μM) and 1fa (18.65μM ± 2.28μM), were similar to their parent compounds, 1 (24.6μM ±
15.64μM), 1a (39.27μM ± 0.85μM) and 1f (27.94μM ± 3.35 μM) respectively, suggesting that
the activity is not due to a false positive caused by nucleophilic addition to the β-position of the
Michael acceptor. Ultimately, the replacement of the double bond may be advantageous in the
design of therapeutics, as metabolic stability and brain penetrance may be enhanced by the
removal a potential Michael acceptor.
2.2.3 Phenyl Substitutions
The phenyl group of ferulic acid was generally not amenable to substitution. Two simple
alkyl-amides (with moderate activity) were selected as parent molecules, and substituted with
several side chains. In both cases, replacing the hydroxide with a phenyl ether, substantially
increased anti-oligomeric activity (Table 5: 13a vs. 13b; 14a vs. 14b). As with the carboxyl-arm,
it is likely the added benzene facilitates further π interactions with Aβ, enabling more
energetically favorable interactions between the molecule and peptide. However, most
modifications thereafter substantially decreased activity. Adding either electron withdrawing
(14e) or donating (14d) groups to the phenyl ring eliminated all discernable activity. Likewise,
17

lengthening the alkyl chain, or modifying the phenyl ring also decreased activity. Introduction of
a toluenesulfonyl (tosyl) group was the sole variation associated with a gain in activity
(compound 13c; IC50 = 10.64µM ± 3.47µM). It is notable that the tosyl substituent preserves
many elements of the general spatial arrangement of the phenyl ether; yet, its susceptibility to
nucleophilic attack and other modification renders it impractical as a potential therapeutic agent.
The polarity of the sulfate group would also decrease BBB penetrance, and increase the
metabolic burden on the drug. This investigation therefore employed the benzylether for all
subsequent molecules.
18

Table 5
Calculated IC50 against Aβ oligomerization for hydroxyl–derivatives of ferulic acid amides
Benzyl
Substitution
bio-Aβ42
IC50 (μM)
Benzyl
Substitution
bio-Aβ42
IC50 (μM)
Parent molecule Comp.
Comp.
13a
(1a)
39.27 ± 0.85
N.A.
N.A.
13d
13e
205.46 ± 8.18
10.64 ± 3.47
13b
13c
14a
(1b)
12.38 ± 4.13
N.A.
14d
98.79 ± 26.28
45.31 ± 25.0
N.A.
14b
14c
14e
14f
. N.A.
N.A. denotes no discernable activity; uncertainty denotes standard deviation; ® on parent
molecule denotes site of substitution. All compounds were assessed in at least three independent
samples, with a minimum of three measurements.
19

2.3 Inhibition of Aβ Fibrillization
Compounds with demonstrated anti-oligomeric activity were tested against fibrillization
employing a standard Thioflavin T (ThT) assay. Unlike the direct approach of the oligomer
assay, the ThT assay relies on changes in the fluorescence of thioflavin when associated with a
large β-pleated sheet structure to quantify the presence of fibrils. Inhibition was calculated as a
percent difference in comparison to a non-inhibitory control, averaged after approximately 100
hrs of incubation (Fig. 2, summarized in Table 6). To facilitate aggregation over the
experimental time frame, a relatively high concentration of Aβ (20 μM) was employed;
compounds were therefore tested at corresponding concentration of 100μM. To mitigate
potential anomalies in fluorescence at these concentrations, the relative change in florescence
(over time) was tracked, and was quantified as a percent relative to a negative sample of Aβ
containing only DMSO. If more than 10% inhibition was observed, they were retested at 20μM.
20

Fig 2. Fibrillization of Aβ, visualized by the fluorescence of thioflavinT (ThT), under inhibition
by 100µM of curcumin (+), and ferulic acid (×). No inhibition control (●) was treated with 0.4uL
of DMSO to control for the volume of DMSO employed in dissolving compounds.
21

Table 6
Percent inhibition of Aβ Fibrillization for amide derivatives of ferulic acid
% inh.
% inh.
% inh.
% inh.
Comp. Structure
Aβ40
Aβ40
Comp. Structure
Aβ40
Aβ40
(100μM) (20μM)
(100μM) (20μM)
Ferulic Acid
Curcumin
0
0a
36.83%
9.75%
78.15%
65.48%
1
1f
0%
13.42%
13.26%
8.81%
7.28%
0%
1a
1aa
1b
1fb
1fc
1fa
0%
6.49%
16.96%
9.35%
22.31%
0.29%
5.50%
3e
6
3i
7
48.23%
9.56%
14.85%
20.57%
0%
9
11
12
44.11%
27.53%
0%
13.82%
0%
5.98%
10
2.21%
1ab
11d
91.88%
73.08%
3.62%
1.95%
All compounds were assessed in at least three independent samples, with a minimum of three
measurements.
22

We observed significant discrepancies in fibrillization and oligomerization activities of
individual compounds. Derivative 1ab, for example, strongly inhibits Aβ fibril formation (91.8%
at 100μM, 73.1% at 20μM) but only weakly interferes with early oligomerization events (IC50 =
119.16μM ± 2.65μM). More broadly, when fibrillization was plotted against oligomerization
(inverse IC50), the lack of correlation is clear (Fig. 3). While a weak (R2 = 0.06) positive trend
can be discerned, the majority of compounds effective against one facet of aggregation was
ineffective against another. Therefore, compounds which inhibit the assembly of large Aβ fibrils
do not necessarily interfere with early nucleation and oligomerization events.
23

0.6
0.5
0.4
0.3
0.2
0.1
0
0
20
40
60
80
100
Inhibiton of Fibrillization (%)*
Fig 3. Fibrillization of Aβ, assayed by percent inhibition of thioflavinT (ThT) fluorescence,
plotted against inhibition oligomerization (assayed by outlined ELISA protocol).
24

Likewise, compounds effective against oligomerization, but ineffective against
fibrillization were observed (Table 5: compound 1, 1fa, 1b, 1f, 2i). Though the precise
mechanism of this assembly is unclear, it may arise from the ability of Aβ to form disordered
oligomers, which can survive to form less stable, but still viable fibrils. Assays of fibrillization
and oligomerization should therefore be treated as distinct measures, and not conflated as a
definitive assay of amyloid aggregation. Moreover, optimizing a compound to increase anti-
fibrillary activity may be ineffective at preventing oligomerization, and vice-versa. As soluble
oligomers, not large fibrils or plaques, are culpable for the neurotoxicity of Aβ, we suggest
prioritizing the optimization of anti-oligomeric activity – though ideally both should be
considered concurrently.
2.4 Design of Anti-Aβ Aggregates: Lessons from Ferulic Acid Derivatives
The activity of derivatives tested in this investigation suggest that interactions between
monomers of Aβ occur over a large surface, ideally targeted by bulky, heavily substituted
molecules. However, as a compound grows, solubility, metabolism and BBB penetrance tend to
become less favorable. Balancing the dual considerations of increased surface area and
optimizing brain bioavailability should therefore be a consideration in future drug development.
In this investigation, the most potent analogs in preventing Aβ oligomerization were 3e
(IC50 = 3.13 ± 1.87 μM), 3f (IC50 = 3.65 ± 0.33 μM), 3g (IC50 = 4.78 ± 1.67 μM), 3h (IC50 = 3.61
± 0.59 μM), and 3i (IC50 = 1.8 ± 0.73 μM). These molecules share an azetidine tether between
the ferulic acid and their elongated side chains (derived from β-alanine). They further have
multiple π-rich systems on both termini of the molecule, and are sharply bent into a
25

perpendicular three-dimensional structure. Notably, these molecules are not efficient anti-
fibrillary agents, with only modest inhibition of fibrillization at 48.23% and 20.57% for
compounds 3e and 3i respectively. As such, these compounds may be able to inhibit the
neurotoxic oligomerization of Aβ, but may allow for the aggregation into benign plaques to
proceed. It is as yet unclear what the impact of this selective inhibition may be on overall disease
progression and outcomes, but in light of emerging data on the differing aggregative properties
of Aβ, should be a priority for future study.
2.5 Quantitative Structure–Activity Relationship (QSAR) Model
The observations of this investigation were used in the creation of a QSAR model to
predict the activity of novel compounds. A total of 79 compounds with measured IC50 values
were randomly divided into a training set of 54 molecules and a validation set of 25 molecules,
each with a varying range of structural features and bio-activities. Using Molecular Operating
Environment software, 318 descriptors pertaining to various molecular properties were
calculated for the training set of molecules. These data proved insufficient to allow for accurate
prediction of IC50 values; thus a binary prediction model of active vs. inactive was adopted.
The model was developed by sequentially eliminating individual descriptors, based on
relative importance, to identify the key factors that would predict compound activity. A binary
pIC50 (-log[IC50]) threshold of 3.7 was selected to distinguish between the active and inactive
compounds. A training set of 36 compounds was created of which 18 were classified as active.
The binary QSAR model predicted 34/36 as active (with a cross-validation of 30/36) and 18/18
as inactive (with a cross-validation of 14/18). When the model was applied to the validation set
26

of 25 molecules, 15 of the 15 active molecules were correctly predicted active, and 7 of the 10
inactive compounds were correctly predicted inactive. The model produced a Cohen's kappa (κ)
coefficient of 0.919; and κ = 0.737 for the validation set – suggesting strong agreement among
the predicted and actual results. The descriptors used in the model include the number of carbon
atoms present, atomic polarizability, the shape of the molecule, capacity for forming hydrogen
bonds, along with the hydrophilic volume of the molecule. Further details of the QSAR model
are described in the supplemental information.
27

3. Conclusion
A diverse library of “curcumin-like” small molecules modelled on amide-derivatives of
ferulic acid was constructed to discern the molecular characteristics relevant for the inhibition of
Aβ aggregation. Collectively, this library addresses three structural components of ferulic acid: i)
the carboxylic acid/amide, ii) the phenyl hydroxyl group, and iii) the rigid, planar backbone
between the carboxylic acid and aromatic ring.
Our findings suggest that modifications to these three prominent features have a significant
effect on the inhibitory potency. Substitution of the hydroxyl group with a benzyl ether generally
increased activity, though subsequent modification of the benzene’s electron dynamics lost any
gains obtained from the benzene. The amide moiety (formerly the ferulic carboxylic acid) was
also crucial for function. Extensive substitutions revealed that the most effective compounds
employed a rigid alkyl tether, followed immediately by a second highly polar amide. This
regional hydrophobic/hydrophilic transition may provide a specific target for future therapeutics
to consider. Extending these molecules with an additional phenyl ring further increased activity,
and mimics the extended, terminal-phenyl arrangement of ferulic acid. In general, lengthening
the molecule with bulky substitutions increased activity, suggesting a wide amyloidogenic region
of interaction between monomers of Aβ; however these may impede the penetrance and
solubility of an eventual drug in the brain. As such, balancing the bulk of a molecule to optimize
both its interaction with Aβ and its bioavailability will be a critical consideration in future drug
design.
Studies should also account for differences in the oligomerization and fibrillization of Aβ. As
observed, compounds effective against the oligomerization of Aβ may be ineffective against the
28

fibrillization of Aβ, and vice-versa. Though compounds effective against both mechanisms of
aggregation are the traditional choice, it is conceivable that a compound which inhibits
oligomerization, but which allows for the formation of benign fibrils and plaques may both
prevent the accumulation of highly toxic soluble oligomers, and allow for the sequestration of
Aβ. It is as yet unclear whether this may be a viable therapeutic consideration. This study
concluded by employing these data to generate a successful QSAR model capable of making a
binary prediction, with high sensitivity and specificity, as to whether a compound was active or
not.
With acknowledgement to prior work on ferulic acid and curcumin, we note that the
identification of potent anti-Aβ oligomeric agents (compounds 3e-i) will likely be insufficient for
a definitive therapy against AD. Further studies will require more comprehensive analysis within
the evolving biochemical paradigm of AD pathology. Moreover, details studies on the
degradation and pharmacology of compounds within the human body, including transit of the
BBB are necessary. The intention herein was therefore not to identify a novel therapy, but rather
to employ a novel library of ferulic acid derivatives to elucidate relevant molecular
characteristics to consider in future drug design.
29