Beyond amino acid sequence: disulfide bonds and the

origins of the extreme amyloidogenic properties of

insulin’s H-fragment

ski1,2 and Wojciech Dzwolak1
Robert Dec1, Michał Kolin
1 Faculty of Chemistry, Biological and Chemical Research Centre, University of Warsaw, Poland
2 Mossakowski Medical Research Centre, Polish Academy of Sciences, Bioinformatics Laboratory, Warsaw, Poland

Keywords The presence of disulfide bonds affects the protein stability and therefore
aggregation propensity; configurational tendency to misfold and form amyloid-like fibrils. Insulin’s three disulfide
entropy; disulfide bridges; loop entropy;
bridges stabilize the native state and prevent aggregation. Partial proteoly-
molecular dynamics
sis of insulin releases highly amyloidogenic and inherently disordered two-
Correspondence chain ‘H-fragment’ retaining insulin’s Cys7A-Cys7B and Cys6A-Cys11A
W. Dzwolak, Faculty of Chemistry, disulfide bonds. The abrupt self-association of H-fragment monomers into
Biological and Chemical Research Centre, fibrils is suppressed in the presence of disulfide-reducing agent. These cir-
University of Warsaw, 1 Pasteur Str., cumstances make the H-fragment an interesting model to study the impact
02-093 Warsaw, Poland of disulfide bonds on amyloidogenesis beyond the ‘stabilization-of-the-na-
Tel: +48 22 552 6567
tive-state’ paradigm. Here, we investigate fibrillization of various synthetic
E-mail: [email protected]
peptides derived from the H-fragment through modifications of Cys7A-
(Received 19 December 2018, revised 8 Cys7B/Cys6A-Cys11A bonds. In comparison to H-fragment, aggregation
March 2019, accepted 10 April 2019) of a two-chain ‘AB’ analog lacking Cys6A-Cys11A bond is decelerated,
while the alternative removal of Cys7A-Cys7B bond releases a non-aggre-
doi:10.1111/febs.14849 gating B-chain and a highly amyloidogenic ‘ACC’ fragment containing the
intrachain Cys6A-Cys11A bond. Our analysis, supported by calculations of
configurational entropy, suggests that Cys6A-Cys11A bond is a key factor
behind the explosive self-association of H-fragment. The bond restricts the
conformational space probed by nucleating monomers which is reflected by
an approximately 2.4 kJ
mol 1 K 1 decrease in entropy. The fact that the
intact Cys6A-Cys11A bond promotes fibrillization of the H-fragment is
remarkable in light of the previously established role of the same disulfide
bond in preventing formation of insulin fibrils. Our results imply that a sin-
gle disulfide bond within a folded protein and its fragment may play
entirely different roles in aggregation and that this role may evolve with
progressing phases of misfolding.

Introduction
Nowadays, the capacity of proteins to self-associate maladies including Alzheimer’s disease and type II dia-
into amyloid fibrils is recognized as a generic property betes mellitus [2]. Typically, amyloid fibrils are very
of polypeptides [1]. In vivo, the phenomenon of amy- stable both thermodynamically and mechanically [3],
loidogenesis has been linked to certain degenerative and these properties are often utilized by living

Abbreviations
AFM, atomic force microscopy; BI, bovine insulin; CD, circular dichroism; FT-IR, fourier transform infrared; GdnHCl, guanidine hydrochloride;
MD, molecular dynamics; PME, particle Mesh Ewald; RMSF, root-mean-square fluctuations; TCEP, Tris(2-carboxyethyl) Phosphine; ThT,
thioflavin T.

3194 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

organisms [4]. The susceptibility to amyloidogenic fibrils. All three bonds remain intact throughout the
aggregation clearly depends on the primary structure. fibrillization process [28]. It was shown that elimina-
For example, the tendency to form aggregates is often tion of any of insulin’s three disulfide bridges destabi-
enhanced by the presence of nonpolar amino acids lizes the native state and promotes fibrillization [29],
with b-sheet propensity; especially, in particular linear whereas introduction of a fourth noncanonical SS
arrangements with polar residues, whereas presence of bond between cysteine-substituted residues 10th of A-
proline residues or clusters of residues with uncompen- chain and 4th of B-chain prevents formation of amy-
sated electric charges is expected to have the opposite loid [30]. Huang et al. have shown that two isomers of
effect [5–7]. This level of consideration has become a human insulin with swapped SS bonds, namely: insu-
starting point for the development of many computa- lin-swap ([A7-A11, A6-B7, A20-B19]) and insulin-
tional tools predicting amyloidogenic properties of sin- swap2 ([A6-A7, B7-A11, A20-B19]) form fibrils more
gle chain peptides [8,9]. The prediction becomes easily than the unaltered protein [17]. Clearly, all the
significantly more challenging when a complex topol- three disulfide bridges present in native insulin con-
ogy of the main chain (typically due to the presence of tribute to the stability of the folded state. We have
disulfide bonds) has to be taken into account. In gen- identified earlier a highly amyloidogenic two-chain
eral, disulfide bonds stabilize the folded state by fragment of insulin (named ‘H-fragment’) consisting of
decreasing configurational entropy of the correspond- the 13 N-terminal residues of A-chain and 11 N-termi-
ing unfolded state [10,11] (a more accurate analysis nal residues of B-chain linked by the Cys7A-Cys7B
highlights, however, a complex interplay of various bond with the Cys6A-Cys11A intrachain bond remain-
factors [12,13]). From this perspective, reduction of ing intact [31] (Fig. 1). Spontaneous self-association of
disulfide bonds in a globular protein is expected to H-fragment monomers proved very sensitive to the
promote aggregation by increasing the population of presence of reducing agent indicating that H-frag-
(partly) disordered conformations, a view often sup- ment’s disulfide bonding is crucial for its amyloido-
ported experimentally—e.g., Ref. [14] (see [15] for a genicity. Importantly, the inherently disordered
review). Transition to an energetically suboptimal iso- character of H-fragment allows one to focus on the
mer by disulfide scrambling should lead to essentially effects of SS bonds in terms of restricting conforma-
similar outcome [16,17] which explains the possible tional space of aggregating monomers unobscured by
link between an in vivo activity of certain disulfide iso- barriers and traps of (partly) folded states. In this
merases and pathogenic protein misfolding [18]. It work, we investigate the role of Cys7A-Cys7B and
should be stressed here that removal or scrambling of Cys6A-Cys11A bonds in fibrillization of H-fragment
disulfide bonds not only accelerates aggregation but using several synthetic peptides (Table 1) and a set of
also often opens alternative self-assembly pathways biophysical and computational tools. The results sug-
leading to different amyloid polymorphs, possibly with gest that a single SS bond may switch between being
distinct biological activity [17,19–21]. Reduction of SS an aggregation-preventing and aggregation-promoting
bonds may facilitate fibrillization by increasing flexibil- factor depending on structural context within the amy-
ity of the polypeptide chain and diminishing the role loidogenic state.
of disulfide-stabilized kinetic traps [22]. On the other
hand, there are examples of non-native disulfide
Results and Discussion
bridges (either energetically frustrating the native state
[17], or artificially linking aggregation-prone mono-
Amyloidogenic properties of H-fragment and its
mers into dimers [23]) which, in fact, accelerate fibril-
AB analog: de novo and seed-induced
lization, even though the resulting amyloid structure
fibrillization
may be thermodynamically metastable vis- a-vis fibrils
accessible to the ‘wild-type’ precursor. The case of a Our preliminary investigation was focused on compar-
single Cys25-Cys80 intrachain bond in b2-microglobu- ative self-association kinetics of synthetic H-fragment
lin enhancing aggregation by reducing mobility of the and its ‘AB’ analog in which A-chain’s Cys6 and
unfolded state is worth mentioning here [24]. Cys11 are substituted with alanine residues (Table 1).
Insulin, a double chain peptide containing three A mildly alkalized 8 M GdnHCl solution was used for
disulfide bonds per monomer (Fig. 1) is an important initial solubilization of agglomerated commercial sam-
amyloidogenic model [25–27]. Despite the significant ples. Subsequently, thus prepared stock peptide solu-
constraints from the SS bonds, insulin readily under- tions were rapidly diluted and acidified (see Materials
goes conformational transition from the predominantly and Methods) triggering fibrillization whose progress
a-helical native structure to b-sheet-rich amyloid was monitored using ThT fluorescence assay. The

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3195
Disulfide bonds in amyloidogenic insulin fragment R. Dec et al.

Table 1. Amino acid sequences of different peptides analyzed in

this work. Disulfide bonds are indicated with solid black line

Name Sequence Exp. Calc.

H Yes Yes
GIVEQCCASVCSL
FVNQHLCGSHL

H-red GIVEQCCASVCSL No Yes

FVNQHLCGSHL

AB GIVEQACASVASL Yes Yes

FVNQHLCGSHL

A-CC GIVEQCAASVCSL Yes No

Fig. 1. The covalent structure and amino acid sequence of highly B FVNQHLCGSHL Yes No
amyloidogenic two-chain H-fragment color-marked within the
primary structure of bovine insulin (BI) monomer. The various
derivative peptides were obtained by Cys?Ala substitutions at the BB FVNQHLCGSHL Yes No
indicated positions of the A-chain (AB, A-CC), or through selective FVNQHLCGSHL
reduction of Cys6A–Cys11A or Cys7A–Cys7B disulfide bridges (H-
red, B, BB).

corresponding kinetic trajectories are presented in unbranched. Diameters of [H] specimen are typically
Fig. 2A. In comparison to intact bovine insulin (BI), within the 4–8 nm range, whereas [AB] tend to form
aggregation of either peptide is strongly accelerated. thicker fibers (approaching 14 nm according to height
The transition is particularly fast for synthetic H-frag- measurements) accompanied by rarer thin (2–3 nm in
ment, as it takes place without measurable lag phase diameter) possibly protofilament-like forms.
which is in accordance with the previously reported According to FT-IR spectroscopic measurements,
behavior of H-fragment samples obtained through main components of the amide I vibrational band are
aggregation-quenched partial proteolysis of insulin at 1625–1626 cm 1 for both [H] and [AB] pointing to
with pepsin [31]. Aggregation of H-fragment mono- predominantly b-sheet structure (Fig. 2C). There are,
mers is complete within the first 3 h. On the other however, fine fingerprint differences between the two
hand, de novo fibrillization of AB-fragment is signifi- types of fibrils: for [AB] the bands are broader and
cantly slower: after the approximately 6-h-long lag overlapped with lesser peaks at 1608 and 1663 cm 1
phase, the elongation phase sets in reaching plateau suggesting possible presence of populations of strongly
after the following 4 h. The visible difference between hydrogen-bonded sheets and loops, respectively. The
final ThT fluorescence intensity levels for H and AB FT-IR spectra of [H] and [AB] are also distinct in the
aggregation cannot be attributed to a lesser amount of range of stretching vibrations of protonated carboxyl
[AB] fibrils as the starting concentrations of H and AB groups above 1720 cm 1.
were identical while negligible fractions of monomers The prolonged lag phase observed upon sponta-
remained in solution at the end of aggregation. Also, neous aggregation of AB disappears when the peptide
as no amorphous (i.e., ThT-negative) aggregates were is seeded with preformed [AB], but also [H] fibrils, as
detected afterwards (see Fig. 2B) we conclude that the the data shown in Fig. 3 demonstrate. The elevated
fluorescence intensity difference is likely to be caused levels of ThT fluorescence at the onset of aggregation
by lower ThT-[AB] binding energy or a lower quantum arise from the residual fluorescence of the added seeds
yield of fluorescence of [AB]-bound ThT molecules, as (different for the same concentrations of [H] and [AB]
has been observed for certain types of amyloid fibrils for the reasons described above) combined with the
[32]. low enhancement of ThT emission by daughter [AB]
Morphological characterization of aggregates with fibrils. We note that daughter [AB] fibrils poorly
AFM revealed only fibrillar specimen (Fig. 2B). enhance ThT fluorescence regardless of whether the
Mature [H] and [AB] fibrils are straight and seeding was carried out using [AB] or [H] templates.

3196 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

Fig. 3. Effect of seeding with sonicated preformed [H] and [AB]

Fig. 2. Amyloidogenic properties of H- and AB-peptides. (A) fibrils on kinetics of aggregation of AB probed by ThT fluorescence.
Kinetics of spontaneous reassociation probed by ThT fluorescence Sample conditions: 2 mg
mL 1 peptide concentration in 0.05 M
at 2 mg
mL 1 peptide concentration in 0.05 M NaCl, 1.33 M NaCl, 1.33 M GdnHCl, H2O, pH = 1.9, at 37 °C, the mass ratio of
GdnHCl, H2O, pH = 1.9, at 37 °C. (B) AFM amplitude images of peptide monomers:seeds was 20 : 1; Orange and yellow plateaus
amyloid fibrils self-assembled from H- and AB-peptides under correspond to control data on ThT fluorescence in the presence of
the same conditions. (C) Corresponding FT-IR spectra (amide seeds only (at the concentrations used for seeding). (B)
I/I’ band region) of fibrils resuspended in D2O, 0.05 M NaCl, Morphologies of daughter AB fibrils induced by homologous seeds
pD 1.9. (left) and through cross-seeding with [H] fibrils (right) probed
by AFM. (C) Corresponding FT-IR spectra (amide I/I’ band
This suggests that no conformational memory effect region) of daughter AB fibrils resuspended in D2O, 0.05 M NaCl, pD
1.9.
(i.e., imprinting of structural features of mother seed
in daughter amyloid fibrils) takes place upon AB/[H]
cross-seeding. The observation is supported by mor- both samples, thick (approx. 14 nm in diameter) fibrils
phological (Fig. 3B) and spectroscopic (Fig. 3C) char- are observed along with very thin specimen. The broad
acterization of daughter amyloid samples. Namely, in distribution of fibril diameters is characteristic for

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3197
Disulfide bonds in amyloidogenic insulin fragment R. Dec et al.

mother [AB], rather than [H] fibrils. Also, the infrared

features of daughter fibrils appear to resemble more
closely those of mother [AB] amyloid. For example,
the broad main component of the amide I band is
flanked by the minor bands at 1608 and 1662 cm 1;
the flattened carboxyl band is blue shifted above
1730 cm 1.

Various roles of Cys6A-Cys11A and Cys7A-Cys7B

bonds in amyloidogenic properties of H-
fragment-derived peptides
According to the data presented so far, the removal
of the Cys6A-Cys11A bond through Cys?Ala substi-
tution dramatically prolongs the nucleation phase.
However, when this phase is no longer the rate deter-
mining step (in the presence of preformed [AB]
seeds), the aggregation becomes very fast. As the
intrachain Cys6A-Cys11A bond appears to strongly
contribute to H-fragment’s amyloidogenicity, it has
become of interest to screen fibrillization of various
H-fragment’s analogs under both disulfide-reducing
and ambient conditions. A set of model peptides
carefully selected for such a test could provide
insights into relationship between the remaining
Cys7A-Cys7B bond and the H-fragment’s proneness
to form fibrils (Table 1). We have employed TCEP, a
powerful disulfide-reducing agent active also at low
pH at which fibrillization of H-fragment is carried
out [21,33]. Along with H and AB, the study was
conducted on selected ‘A-CC’ and ‘B’/‘BB’ peptides.
The ‘A-CC’ retains H-fragment’s A-chain part (in-
cluding Cys6A-Cys11A bond) but due to the substitu- Fig. 4. (A) Comparison of kinetics of spontaneous aggregation of
tion of Cys7A with alanine it lacks the B-chain various H-derived peptides under ambient and reducing (marked
with ‘RED’ subscript) conditions. Kinetics of aggregation were
fragment (which was synthetized separately with
probed using ThT fluorescence assay under the typical conditions
Cys7B remaining in reduced form). We were also able (2 mg
mL 1 peptide in 0.05 M NaCl, 1.33 M GdnHCl, H2O,
to obtain samples of ‘BB’ peptide: a Cys7B-Cys7B pH = 1.9, at 37 °C). The reducing conditions were achieved
disulfide-stabilized homodimer of B. While this reper- through the addition of TCEP. (B) Far-UV CD spectra of various
toire of H-fragment’s analogs does not exhaust all peptides in the monomeric state compared with those
interesting possibilities, it provides critical reference corresponding to diluted suspensions of [H] and [AB] fibrils and
data necessary to illuminate the role of the peptide’s native BI. Concentration of all peptide samples in H2O, pH 1.9 was
0.2 mg
mL 1.
disulfide bonds in its self-assembling properties.
The outcome of the ThT-fluorescence-based mea-
surements shown in Fig. 4A should be interpreted in
qualitative terms, only, as determination of kinetic
H  A-CC > AB >> BB (under the ambient condi-
rates corresponding to different aggregation phases of
tions), and
these peptides was beyond the scope of this study and
HRED  A-CCRED > ABRED >> BBRED (under the
would also require experimentally inaccessible amounts
reducing conditions).
of samples. Yet, based on the kinetic trajectories pre-
sented in Fig. 4A, several key observations may be Secondly, the TCEP-induced reduction appears to
made. Firstly, according to approximate lengths of slow down aggregation of both H and A-CC peptides,
nucleation phase, the rate of fibrillization decreases as but accelerates fibrillization in AB sample. In the pres-
follows: ence of TCEP, an effective fast-acting but nonspecific

3198 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

reducing agent, all accessible disulfide bonds are

A
promptly reduced (prior to aggregation all dissolved
synthetic fragments of BI were in the random coil con-
formation as far-UV CD demonstrates – see Fig. 4B).
Therefore, it follows that H-fragment monomers con-
vert into an equimolar mixture of reduced B- and A-
CC fragments when TCEP is added at the onset of
measurement. Meanwhile, the parallel measurements
clearly indicate that reduced B-fragment and its cova-
lent BB dimer are non-amyloidogenic (at least under
the conditions and the experimental timescales exam-
ined here). Hence, the elsewhere reported pro-amy-
loidogenic effect of disulfide bonds crosslinking
monomers ([23]) is not observed for the B/BB pair.
Indeed, no aggregates were found in B/BB samples
even after prolonged incubation. This contrasts with
the explosive fibrillization of A-CC peptide, especially
under ambient conditions (Fig. 4A).
B
The morphological and spectroscopic characteriza-
tion of [ACC] aggregates is summarized in Fig. 5. These
straight fibrils, typically 6–13 nm in diameter, reveal
periodic twists also observed for [H] and [AB] specimen.
However, the infrared fingerprint is more characteristic.
The main component of the amide I band at 1624 cm 1
along with the well-resolved minor band at 1660 cm 1
are very narrow and symmetric reminiscent more of [H]
fibrils. These sharp features and the overall simplicity of
the infrared spectrum indicate presence of highly
ordered and homogenous structures. The A-CC case
clearly supports the previous assertion that the Cys6A-
Cys11A bond enhances formation of fibrils. The appear-
ance of the HRED trajectory in Fig. 4A may be
explained by the presence of in situ released reduced A-
CC fragment undergoing fibrillization on its own (some-
how decelerated due to Cys6A-Cys11A bond reduction) Fig. 5. (A) Amplitude AFM image of [A-CC] aggregate formed upon
a 48 h/37 °C incubation of 2 mg
mL 1 ACC dissolved in 0.05 M
with the reduced non-amyloidogenic B-fragment
NaCl, 1.33 M GdnHCl in H2O, pH 1.9. (B) The corresponding
remaining in solution. Only half of the overall peptide solvent-subtracted FT-IR transmission spectrum of [A-CC]. The
content (in terms of mass) participates in the fibrilliza- precipitate was washed several times with portions of 0.05 M NaCl
tion under these conditions which contributes to the vis- in D2O, pD 1.9 prior to being resuspended in this solution for
ible reduction of ThT fluorescence intensity also spectral measurement.
observed when enzymatically derived H-fragment was
made to reassociate in the presence of TCEP ([31]). In
the case of AB, reduction of the single Cys7A-Cys7B bond. During aggregation of H-fragments, the other-
bond can only enhance aggregation by untethering the wise non-amyloidogenic B-chain fragment appears to
amyloidogenic A-fragment from the aggregation-resis- be ‘dragged along’ through the Cys7A-Cys7B tether
tant B-fragment. Unsurprisingly, while the fibrillization and forced to attain the b-sheet conformation (remark-
kinetics of AB and A-CC under ambient conditions are ably, infrared spectra of [BI], [H], and [ACC] fibrils
very different, they become similar in the presence of indicate high and similar b-sheet contents (Figs 2 and
TCEP. 5). We have shown previously ([31]) that the Tango
These experimental results provide strong evidence algorithm fails to predict the scale of amyloidogenic
that the driving forces behind the powerful amyloido- potential of fully reduced A-chain part of the H-frag-
genic properties of H-fragment originate from its A- ment. In most cases, topological constrains from disul-
chain part and the intrachain Cys6A-Cys11A disulfide fide bridges cannot be even taken into account by such

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3199
Disulfide bonds in amyloidogenic insulin fragment R. Dec et al.

computational methods. According to the far-UV CD of entropy of AB- and H-red fragments are similar
data, the presence of pro-amyloidogenic Cys6A- [8.974 kJ
mol 1 K 1, and 8.992 kJ
mol 1 K 1, respec-
Cys11A bridge does not entail stabilization of the tively (see Table 2)], but significantly larger than
main chain conformation: both H and AB are per- entropy of the disulfide-containing unaltered H-peptide
fectly disordered. There is, however, one unavoidable (6.571 kJ
mol 1 K 1).
consequence of formation of this bond: the loss of It should be stressed that an interplay of kinetic and
entropy due to restriction of conformational space strictly thermodynamic factors determines the net result
accessible to the peptide fragment involved in disul- of introducing a disulfide loop into an amyloidogenic
fide-closed loop (‘loop entropy’). The approximate peptide. In a manner analogous to disulfide-induced
magnitude of this effect scales with natural logarithm destabilization of unfolded state, the loop entropy effect
of the number of residues in the loop [34]. The caused by Cys6A-Cys11A is expected to increase the
approach chosen in this study enables a more accurate absolute value of Gibbs free energy change accompany-
estimation of the entropy difference between disulfide- ing the self-assembly of H-monomers into amyloid fibrils
bonded and non-bonded peptide chains. In the follow- [10,11,34]. For a disordered amyloidogenic fragment, a
ing part, these entropic effects were calculated and dis- decrease in configurational entropy may facilitate nucle-
cussed for H-, and AB-fragment, as well as H-red ation of self-propagating amyloid structure and shorten
fragment with selectively reduced Cys6A-Cys11A bond the lag phase, as is the case of H-fragment. Comparison
(this entity exists only in silico due to instantaneous of RMSF parameter calculated for each residue of three
exchange of near –SH groups and SS bonds). analyzed peptides again shows very similar values for
both AB- and H-red peptides ranging from ~0.37 to
~1,15 nm (see Fig. 6B). For H-fragment, RMSF range
Configurational entropy of H-, H-red-, and AB-
from ~0,11 to ~0.69 nm indicating a decreased confor-
fragments
mational flexibility. Strikingly, the attenuation of con-
We have employed MD simulations to investigate the formational fluctuations caused by the Cys6A-Cys11A
role of the Cys6A-Cys11A disulfide bridge in restricting bond within H-fragment (compared to AB and H-red)
the conformational flexibility of the H-fragment. Abso- extends well beyond the immediate vicinity of the bridge.
lute configurational entropy of H-, AB-, and H-red frag- For example, the disulfide bond appears to damp fluctu-
ments has been estimated using Schlitter formula [35]. ations of A-chain’s N and C termini by a factor of 4–5,
The RMSF parameter values for every residue of each whereas for the B-chain’s termini this effect is much less
simulated peptide were also calculated. For each system, pronounced. This observation suggests that a synergistic
30 independent MD simulations have been performed; effect between the aggregation-prone A-chain’s part and
each starting with random initial velocities and lasting the local restriction of chain’s fluctuations by the Cys6A-
100 ns. The resulting MD trajectories were then succes- Cys11A bond may play a role in determination of the
sively combined, one after the other, into one long accu- amyloidogenic properties of H-fragment. A visual
mulated trajectory lasting 3 ls for each analyzed inspection of trajectories unveils further differences in
system. We have chosen to perform large number of structure and dynamics of analyzed peptides (Fig. 6C).
independent simulations in order to achieve sufficient A comparison of representative set of 500 conformers
sampling of the peptide conformational space. Even in a extracted every 6 ns from cumulative MD trajectory for
very long single simulation run, the modeled system each system indicates that the fragments deprived of the
could get trapped in a local potential minimum and a Cys6A-Cys11A bond are more prone to strong fluctua-
satisfactory level of convergence would be not achieved. tions than H-fragment. Taken altogether, these in silico
It has been demonstrated by Caves et al. [36] that multi- results indicate that the intrachain Cys6A-Cys11A disul-
ple runs allow for a better sampling of accessible confor- fide bond significantly restricts the conformational flexi-
mational space than a single simulation of equivalent bility of H-peptide, and of the aggregation-prone A-
length. For the systems investigated in the present work, chain part in particular.
the plot presenting values of absolute entropy accumu- In summary, we have demonstrated for the first time
lated over simulation time shows good convergence (see that the elusive origins of the explosive amyloidogenicity
Figure 6A). After 2500 ns of each cumulative trajec- of insulin H-fragment are intimately linked to its A-
tory, the calculated entropy value is practically constant chain part and the single Cys6A-Cys11A disulfide bond
and this value (after averaging over the entropy values therein. Despite being refractory to aggregation on its
obtained using accumulated trajectories ranging from own, the B-chain fragment follows the amyloidogenic
2500 ns to 3000 ns) is in each case put forward as the self-assembly process apparently initiated within the A-
absolute configurational entropy. The calculated values chain region. The pro-amyloidogenic Cys6A-Cys11A

3200 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

Fig. 6. (A) Entropy values calculated

according to the Schlitter formula for H-
fragment with (red), and without Cys6A–
Cys11A disulfide bond (removed through
reduction, blue, or double Cys?Ala
substitution, black). (B) RMSF parameter
values calculated for the three peptides at
different residues. (C) Trace representation
of structures of the three peptide models.
For each system, a total of 500 snapshots
were extracted from MD trajectory every
6 ns and superimposed on the starting
conformation. A- and B-chains are marked
in blue and red, respectively.

Table 2. The list of conducted MD simulations and estimated by Pepscan (Lelystad, The Netherlands). BI and other non-
absolute entropy values due to the Schlitter formula [35] deuterated chemicals were purchased from Sigma-Aldrich
1
(St. Louis, MO, USA). D2O (‘99.8 atom % D’ grade) and
Peptide Simulation time Entropy [J
mol K 1]
DCl were from ARMAR Chemicals (D€ ottingen, Aargau,
H (30 9 100 ns) 3000 ns 6571.6  8.5 Switzerland). Solid samples of the synthetic peptides pro-
H-red (30 9 100 ns) 3000 ns 8992.4  4.7 vided by the manufacturer as trifluoroacetic acid salts were
AB (30 9 100 ns) 3000 ns 8974.7  5.5 initially dissolved in 8 M GdnHCl, pH 9.0, at a
12 mg
mL 1 concentration. For the measurements of kinet-
ics of peptide fibrillization, thus obtained fresh liquid stock
bridge can be conceptualized as a loop restricting the samples were swiftly diluted with 0.1 M NaCl, H2O, and
conformational space probed by nucleating H-fragment HCl to give final 2 mg
mL 1 peptide solutions in 0.05 M
monomers which is reflected by an approximately ¼ NaCl, 1.33 M GdnHCl, pH 1.9 additionally containing
amyloid-specific fluorophore: ThT at 20 lM concentration.
decrease in the absolute configurational entropy com-
The reference measurements of kinetics of spontaneous BI
pared to peptide analogs lacking this bond. The disul-
fibrillization were collected under the same solution condi-
fide-induced damping of molecular fluctuations is not
tions. The reducing conditions of the kinetic fibrillization
uniform but affects mostly the amyloidogenic A-chain
experiments depicted in Fig. 4 were achieved by an addi-
part. Our study depicts an interesting example of disul-
tion of TCEP to its final concentration of 6.8 mg
mL 1.
fide bridge whose role in fibrillization becomes redefined We use brackets to label mature aggregated amyloid-like
with the changing structural context (i.e., shifting from forms of peptides (e.g., ‘[AB]’ corresponds to amyloid-like
a folded monomer to a disordered fragment). The obser- fibrils self-assembled from AB monomers).
vation that disulfide-induced decrease in fluctuations
within an amyloidogenic sequence may dramatically
increase its amyloidogenic potential requires further Fibrillization kinetics
attention; especially given the emerging necessity to con- For ThT-fluorescence-based measurements (kex. 440 nm/
sider local topological constrains in further development kem. 485 nm) of fibrillization kinetics, a CLARIOstar
of amyloidogenicity-predicting tools. plate reader from BMG LABTECH (Offenburg, Germany)
and 96-well black microplates were used. Typically, wells
were filled with 150 lL volumes of diluted peptide samples
Materials and methods
containing ThT. Measurements were carried out at 37 °C
without agitation for 48 h; the sections shown in figures
Samples
correspond to the first 16 h of kinetic experiments which
All insulin fragments as listed in Table 1 (except for the encompassed all the significant changes. Each kinetic trace
inherently unstable H-red peptide) were custom-synthesized was calculated as an average from three independently

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3201
Disulfide bonds in amyloidogenic insulin fragment R. Dec et al.

collected trajectories (the error bars correspond to the stan- Simulations

dard deviations). Once the kinetic measurements were com-
pleted, samples of peptide aggregates were collected from Configurational entropy can be calculated using the Boltz-
wells and after appropriate dilution/removal of excess of mann’s formula by integrating the multidimensional proba-
GdnHCl were subjected to AFM imaging and FT-IR spec- bility density function over the coordinate space. However,
troscopic measurements. such an approach becomes impractical even for small
molecular systems due to large number of degrees of free-
dom and insufficient sampling during the simulation which
Atomic force microscopy
prevents direct evaluation of the integral [37]. Karplus and
Samples of aggregates formed during the plate-reader-as- Kushick [38] applied a quasi-harmonic analysis and devel-
sisted kinetic measurements are subsequently subjected to oped a method to approximate differences in configura-
AFM imaging. A small portion of aggregate suspension tional entropy employing determinants of covariance
collected from a plate well was initially washed several matrices of atomic fluctuations. In this method, internal
times with slightly acidified 0.05 M NaCl aqueous solution coordinates were usually used for derivation of matrices.
in order to remove excess of GdnHCl. Subsequently, sus- While this approach was further improved and applied by
pension of fibrils was diluted 40 times with acidified water other groups [39,40], another method introduced by Schlit-
(without salt). A small droplet (10 lL) of fibrils suspension ter [35] enabled estimation of absolute and relative entro-
was swiftly deposited onto freshly cleaved mica and left to pies. In this approach, entropy formulation is based on a
dry overnight. AFM tapping-mode measurements were car- quantum harmonic oscillator model and represents the
ried out using a Nanoscope III atomic force microscope upper limit for the quantum entropy. Here, conversion to
from Veeco Instruments (Plainview, NY, USA) and internal coordinates is not required because Cartesian coor-
TAP300-Al sensors (res. frequency 300 kHz) from Bud- dinates are used for covariance matrices derivation. Other
getSensors (Sofia, Bulgaria). Other experimental parameters groups used this approach to study various systems includ-
were the same as in earlier studies [31]. ing fluids and protein–protein complexes [41,42]. The
method was also used to estimate entropy of folding pro-
tein molecules [43], and is also implemented in software
FT-IR spectroscopy packages dedicated to MD simulations [44]. We have
Centrifuged samples of aggregates collected from the plate employed the same approach to study configurational
reader were washed five times with excessive amounts of entropy of the H-peptides.
0.05 M NaCl in D2O, pD 1.9 (according to pH-meter readout Starting conformations of the H-peptide were derived
with adjustment) which allowed us to effectively remove both from the crystallographic structure of T6 hexameric bovine
GdnHCl and H2O (whose vibrations obscure peptides’ amide insulin PDB ID: 2A3G [45]. Residues 14 to 21, and 12 to
I band region). For transmission FT-IR measurements at 30 were removed from insulin monomer’s chain A and
25 °C, Nicolet iS50 FT-IR spectrometer from Thermo Fisher chain B, respectively. AB- and H-red fragment starting con-
Scientific (Waltham, MA, USA) equipped with a DTGS formations were, in turn, obtained from the H-fragment
detector and a CaF2 transmission cell with a 0.025 mm model either by introducing appropriate Cys?Ala point
Teflon spacer were used. Typically, for a single spectrum 128 mutations, or by replacing Cys6A-Cys11A bond with free
interferograms of 2 cm 1 resolution were co-added. During reduced cysteine side chains. The resulting peptide fragment
measurements, the sample chamber was continuously purged was inserted into a simulation box (6.9 nm 9 6.9 nm 9
with dry CO2-depleted air. All spectra were corrected by sub- 6.9 nm) and surrounded with 10 819 water molecules. To
tracting the proper amount of D2O and water vapor spectra make the system neutral, 24 chloride ions and 20 sodium
prior to being baseline-adjusted. Data processing was per- ions were added (necessary for PME procedure). MD simu-
formed using GRAMS software (Thermo Fisher Scientific). lations of H-, AB-, and H-red fragments (Table 1) were
carried out. The simulation protocol consisted of three
CD measurements stages. First, geometry of the system was optimized during
2000 steps of steepest descent energy minimization to
Liquid samples were obtained by dissolving soluble pep- remove initial steric clashes between atoms. Next, sur-
tides, and BI or suspending insoluble aggregates in water rounding solvent molecules were equilibrated during a 10-
pH 1.9 at 0.2 mg
mL 1 peptide concentration. The freshly ns-long MD simulation with weak position restrains
prepared samples were subsequently transferred to quartz imposed on the protein backbone atoms. Finally, the prin-
cuvette with 1 mm optical pathway. Far-UV CD spectra cipal computational stage consisted of 30 independent all
corrected for the buffer signal were carried out at room atom MD simulations of a single peptide molecule. Each
temperature by accumulation of five independent spectra MD run started with random initial velocities and lasted
(at 200 nm
s 1 scanning rate) on a J-815 S spectropolarime- 100 ns. Resulting 30 MD trajectories were subsequently
ter from Jasco Corp. (Tokyo, Japan). merged into a single cumulative trajectory that was further

3202 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

used for entropy calculation. The CHARMM36 force filed sequence determinants of amyloid structure using
parameter set [46] was used, simulation step was 2 fs, tra- position-specific scoring matrices. Nat Methods 7, 237-
jectory frames were recorded every 10 ps. The explicit U109.
TIP3P water model was used for solvent molecules [47]. 7 de la Paz ML & Serrano L (2004) Sequence
Hydrogen bonds were restrained using LINCS algorithm determinants of amyloid fibril formation. Proc Natl
[48]. Long-range electrostatic interactions were taken into Acad Sci USA 101, 87–92.
account using PME method [49] with a cutoff of 12 A.  All 8 Meric G, Robinson AS & Roberts CJ (2017) Driving
MD simulations were performed at temperature of 310 K forces for nonnative protein aggregation and
and under pressure of 1013 hPa. All calculations and data approaches to predict aggregation-prone regions. Annu
analysis were completed using GROMACS version 5.1.4. [50] Rev Chem Biomol 8, 139–159.
program and VMD molecular graphics viewer [51]. 9 Oliveberg M (2010) Waltz, an exciting new move in
amyloid prediction. Nat Methods 7, 187–188.
10 Fass D & Thorpe C (2017) Chemistry and enzymology
Acknowledgements of disulfide cross-linking in proteins. Chem Rev 118,
This work was supported by the National Science Cen- 1169–1198.
tre of Poland, grant no. 2015/17/B/NZ1/00832. The 11 Honda R (2018) Role of the disulfide bond in Prion
study was carried out at the Biological and Chemical protein amyloid formation: a thermodynamic and
Research Centre, University of Warsaw, established kinetic analysis. Biophys J 114, 885–892.
within the project cofinanced by EU from the Euro- 12 Betz SF (1993) Disulfide bonds and the stability of
globular proteins. Protein Sci 2, 1551–1558.
pean Regional Development Fund under the Opera-
13 Banach M, Kalinowska B, Konieczny L & Roterman I
tional Programme Innovative Economy, 2007–2013.
(2016) Role of disulfide bonds in stabilizing the
conformation of selected enzymes—an approach based
Conflicts of interests on divergence entropy applied to the structure of
hydrophobic core in proteins. Entropy 18, 67.
The authors declare no conflict of interest. 14 Cao A, Hu D & Lai L (2004) Formation of amyloid
fibrils from fully reduced hen egg white lysozyme.
Author contributions Protein Sci 13, 319–324.
15 Li Y, Yan J, Zhang X & Huang K (2013) Disulfide
RD performed the experiments. MK carried out in sil- bonds in amyloidogenesis diseases related proteins.
ico calculations. RD, MK, and WD analyzed the Proteins: Struct, Funct, Bioinf 81, 1862–1873.
results. WD conceived and designed the study. MK 16 Hua QX, Jia W, Frank BH, Phillips NF & Weiss MA
helped to write the manuscript. WD wrote the manu- (2002) A protein caught in a kinetic trap: structures and
script. stabilities of insulin disulfide isomers. Biochemistry 41,
14700–14715.
17 Huang K, Maiti NC, Phillips NB, Carey PR & Weiss
References MA (2006) Structure-specific effects of protein topology
1 Dobson CM (2017) The amyloid phenomenon and its on cross-b assembly: studies of insulin fibrillation.
links with human disease. CSH Perspect Biol 9, Biochemistry 45, 10278–10293.
a023648. 18 Hoffstrom BG, Kaplan A, Letso R, Schmid RS, Turmel
2 Chiti F & Dobson CM (2017) Protein misfolding, GJ, Lo DC & Stockwell BR (2010) Inhibitors of protein
amyloid formation, and human disease: a summary of disulfide isomerase suppress apoptosis induced by
progress over the last decade. Annu Rev Biochem 86, misfolded proteins. Nat Chem Biol 6, 900–906.
27–68. 19 Yang M, Dutta C & Tiwari A (2015) Disulfide-bond
3 Knowles TPJ & Buehler MJ (2011) Nanomechanics of scrambling promotes amorphous aggregates in lysozyme
functional and pathological amyloid materials. Nat and bovine serum albumin. J Phys Chem B 119, 3969–
Nanotechnol 6, 469–479. 3981.
4 Fowler DM, Koulov AV, Balch WE & Kelly JW (2007) 20 Wineman-Fisher V, Tudorachi L, Nissim E & Miller Y
Functional amyloid–from bacteria to humans. Trends (2016) The removal of disulfide bonds in amylin
Biochem Sci 32, 217–224. oligomers leads to the conformational change of the
5 Tzotzos S & Doig AJ (2010) Amyloidogenic sequences ‘native’amylin oligomers. Phys Chem Chem Phys 18,
in native protein structures. Protein Sci 19, 327–348. 12438–12442.
6 Maurer-Stroh S, Debulpaep M, Kuemmerer N, de la 21 Zako T, Sakono M, Hashimoto N, Ihara M & Maeda
Paz ML, Martins IC, Reumers J, Morris KL, Copland M (2009) Bovine insulin filaments induced by reducing
A, Serpell L, Serrano L et al. (2010) Exploring the disulfide bonds show a different morphology, secondary

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3203
Disulfide bonds in amyloidogenic insulin fragment R. Dec et al.

structure, and cell toxicity from intact insulin amyloid 34 Pace CN, Grimsley GR, Thomson JA & Barnett BJ
fibrils. Biophys J 96, 3331–3340. (1988) Conformational stability and activity of
22 Serebryany E, Woodard JC, Adkar BV, Shabab M, ribonuclease T1 with zero, one, and two intact disulfide
King JA & Shakhnovich EI (2016) An internal disulfide bonds. J Biol Chem 263, 11820–11825.
locks a misfolded aggregation-prone intermediate in 35 Schlitter J (1993) Estimation of absolute and relative
cataract-linked mutants of human cD-crystallin. J Biol entropies of macromolecules using the covariance
Chem 291, 19172–19183. matrix. Chem Phys Lett 215, 617–621.
23 Yamaguchi T, Yagi H, Goto Y, Matsuzaki K & 36 Caves LS, Evanseck JD & Karplus M (1998) Locally
Hoshino M (2010) A disulfide-linked amyloid-b peptide accessible conformations of proteins: multiple molecular
dimer forms a protofibril-like oligomer through a dynamics simulations of crambin. Protein Sci 7, 649–666.
distinct pathway from amyloid fibril formation. 37 Hensen U, Grubmuller H & Lange OF (2009) Adaptive
Biochemistry 49, 7100–7107. anisotropic kernels for nonparametric estimation of
24 Katou H, Kanno T, Hoshino M, Hagihara Y, Tanaka absolute configurational entropies in high-dimensional
H, Kawai T, Hasegawa K, Naiki H & Goto Y (2002) configuration spaces. Phys Rev E Stat Nonlin Soft
The role of disulfide bond in the amyloidogenic state of Matter Phys 80, 011913.
b2-microglobulin studied by heteronuclear NMR. 38 Karplus M & Kushick JN (1981) Method for
Protein Sci 11, 2218–2229. estimating the configurational entropy of
25 Ahmad A, Uversky VN, Hong D & Fink AL (2005) macromolecules. Macromolecules 14, 325–332.
Early events in the fibrillation of monomeric insulin. J 39 Di Nola A, Berendsen HJ & Edholm O (1984) Free
Biol Chem 280, 42669–42675. energy determination of polypeptide conformations
26 Ivanova MI, Sievers SA, Sawaya MR, Wall JS & generated by molecular dynamics. Macromolecules 17,
Eisenberg D (2009) Molecular basis for insulin fibril 2044–2050.
assembly. Proc Natl Acad Sci USA 106, 18990–18995. 40 Edholm O & Berendsen H (1984) Entropy estimation
27 Loksztejn A & Dzwolak W (2010) Vortex-induced from simulations of non-diffusive systems. Mol Physics
formation of insulin amyloid superstructures probed by 51, 1011–1028.
time-lapse atomic force microscopy and circular 41 Sch€afer H, Mark AE & van Gunsteren WF (2000)
dichroism spectroscopy. J Mol Biol 395, 643–655. Absolute entropies from molecular dynamics simulation
28 Kurouski D, Washington J, Ozbil M, Prabhakar R, trajectories. J Chem Phys 113, 7809–7817.
Shekhtman A & Lednev IK (2012) Disulfide bridges 42 Hsu ST, Peter C, van Gunsteren WF & Bonvin AM
remain intact while native insulin converts into amyloid (2005) Entropy calculation of HIV-1 Env gp120, its
fibrils. PLoS One 7, e36989. receptor CD4, and their complex: an analysis of
29 Li Y, Gong H, Sun Y, Yan J, Cheng B, Zhang X, Huang configurational entropy changes upon complexation.
J, Yu M, Guo Y, Zheng L et al. (2012) Dissecting the Biophys J 88, 15–24.
role of disulfide bonds on the amyloid formation of 43 Sch€afer H, Daura X, Mark AE & van Gunsteren WF
insulin. Biochem Bioph Res Co 423, 373–378. (2001) Entropy calculations on a reversibly folding
30 Vinther TN, Norrman M, Ribel U, Huus K, Schlein M, peptide: changes in solute free energy cannot explain
Steensgaard DB, Pedersen TA, Pettersson I, Ludvigsen folding behavior. Proteins: Struct, Funct, Bioinf 43, 45–56.
S, Kjeldsen T et al. (2013) Insulin analog with 44 Hikiri S, Yoshidome T & Ikeguchi M (2016)
additional disulfide bond has increased stability and Computational methods for configurational entropy
preserved activity. Protein Sci 22, 296–305. using internal and Cartesian coordinates. J Chem
31 Piejko M, Dec R, Babenko V, Hoang A, Szewczyk M, Theory Comput 12, 5990–6000.
Mak P & Dzwolak W (2015) Highly amyloidogenic 45 Smith GD, Pangborn WA & Blessing RH (2005) The
two-chain peptide fragments are released upon partial structure of T6 bovine insulin. Acta Crystallogr D 61,
digestion of insulin with pepsin. J Biol Chem 290, 5947– 1476–1482.
5958. 46 Huang J & MacKerell AD (2013) CHARMM36 all-
32 Cloe AL, Orgel JP, Sachleben JR, Tycko R & atom additive protein force field: validation based on
Meredith SC (2011) The Japanese mutant Ab (DE22- comparison to NMR data. J Comput Chem 34, 2135–
Ab1 39) forms fibrils instantaneously, with low- 2145.
Thioflavin T fluorescence: seeding of wild-type Ab1 47 Mahoney MW & Jorgensen WL (2000) A five-site
40 into atypical fibrils by DE22-Ab1 39. Biochemistry model for liquid water and the reproduction of the
50, 2026–2039. density anomaly by rigid, nonpolarizable potential
33 Babenko V, Piejko M, W
ojcik S, Mak P & Dzwolak functions. J Chem Phys 112, 8910–8922.
W. Vortex-induced amyloid superstructures of insulin 48 Hess B (2008) P-LINCS: A parallel linear constraint
and its component A and B chains. Langmuir 29, 5271– solver for molecular simulation. J Chem Theory Comput
5278. 4, 116–122.

3204 The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies
R. Dec et al. Disulfide bonds in amyloidogenic insulin fragment

49 Darden T, York D & Pedersen L (1993) Particle mesh der Spoe DL et al. (2013) GROMACS 4.5: a high-
Ewald: An N
log (N) method for Ewald sums in large throughput and highly parallel open source molecular
systems. J Chem Phys 98, 10089–10092. simulation toolkit. Bioinformatics 29, 845–854.
50 Pronk S, P
all S, Schulz R, Larsson P, Bjelkmar P, 51 Humphrey W, Dalke A & Schulten K (1996) VMD:
Apostolov R, Shirts MR, Smith JC, Kasson PM, van visual molecular dynamics. J Mol Graphics 14, 33–38.

The FEBS Journal 286 (2019) 3194–3205 ª 2019 Federation of European Biochemical Societies 3205