Abstract
Anticancer peptides (ACPs) are rising as a new strategy for cancer
therapy. However, traditional laboratory screening to find and identify
novel ACPs from hundreds to thousands of peptides is costly and time
consuming. Here, a sequential procedure is applied to identify
candidate ACPs from a computer‐generated peptide library inspired by
alpha‐lactalbumin, a milk protein with known anticancer properties. A
total of 2688 distinct peptides, 5–25 amino acids in length, are
generated from alpha‐lactalbumin. In silico ACP screening using the
physicochemical and structural filters and three machine learning
models lead to the top candidate peptides ALA‐A1 and ALA‐A2. In vitro
screening against five human cancer cell lines supports ALA‐A2 as the
positive hit. ALA‐A2 selectively kills A549 lung cancer cells in a
dose‐dependent manner, with no hemolytic side effects, and acts as a
cell penetrating peptide without membranolytic effects. Sequential
window acquisition of all theorical fragment ions‐proteomics and
functional validation reveal that ALA‐A2 induces autophagy to mediate
lung cancer cell death. This approach to identify ALA‐A2 is time and
cost‐effective. Further investigations are warranted to elucidate the
exact intracellular targets of ALA‐A2. Moreover, these findings support
the use of larger computational peptide libraries built upon multiple
proteins to further advance ACP research and development.
Keywords: anticancer peptides, cytotoxic screening, drug discovery,
lung adenocarcinoma, machine learning, peptide library, SWATH‐MS
__________________________________________________________________
This study describes a novel strategy for searching for a new
anticancer peptide (ACP) by integrating the computational peptide
library of alpha‐lactalbumin with downstream in silico machine
learning‐based screening and in vitro experimental validation. The
novel ACP, so‐called ALA‐A2, specifically kills lung adenocarcinoma
cells through autophagy mediated cell death.
graphic file with name GCH2-7-2200213-g002.jpg
1. Introduction
Worldwide, cancer is a leading cause of premature death and a growing
public health burden. Global cancer estimates indicate up to 19 million
new cases and 10 million cancer deaths in 2020.^[ [40]^1 ^]
Conventional cancer therapies, such as chemotherapy and radiation,
frequently result in cancer resistance and a lack of tumor selectivity.
More effective treatment agents are required to improve patient
outcomes.
Anticancer peptides (ACP) are sequences of amino acids of fewer than
25 amino acids in length that have a cytotoxic effect on cancer
cells.^[ [41]^2 ^] The modes of action of ACPs can be divided into
three categories: 1) membranolytic peptides that kill cancer cells
through pore formation on the cellular membrane; 2) cell penetrating
peptides that are rich in basic amino acids (arginine and lysine),
leading to translocation and access to intracellular compartments to
disturb cellular homeostasis; and 3) tumor‐targeting peptides that
directly interact with cancer‐specific molecules to mediate cell
death.^[ [42]^2 ^] Two major approaches of ACP discovery include
activity‐guided purification from biological or natural sources of
interest,^[ [43]^3 ^] and experimental screening of an established
peptide library, that is, antimicrobial peptides.^[ [44]^4 ^]
In our hands, a new strategy of ACP screening based on
bio–physico–chemical features and machine learning (ML) preference of
the input peptides proved to be time‐ and cost‐effective for
prioritizing candidates for experimental studies.^[ [45]^5 ^] A
naturally occurring peptide library from human milk, a promising source
for therapeutic peptide discovery,^[ [46]^5 , [47]^6 ^] was generated
by sequential peptide fractionation coupled with liquid
chromatography‐tandem mass spectrometry. Of 142 input peptides, in
silico ACP screening with subsequent experimental validation identified
an anti‐leukemic peptide that selectively kills four distinct leukemic
cell lines in vitro and three patient‐derived leukemic cells ex vivo.^[
[48]^5 ^] However, the bottleneck of the overall workflow lies in the
relatively small and limited peptide library. Unfortunately, expansion
of the peptide library using traditional approaches is associated with
significant time and budget constraints. To address this issue, we had
proposed a large peptide library generated by a computational method to
facilitate further discovery of ACP from human milk.^[ [49]^5 ^]
Alpha‐lactalbumin, a 16‐kDa milk protein with 142 amino acids, when in
a complex with oleic acid, selectively induces cancer cell death in
vitro and in vivo. Named “human alpha‐lactalbumin made lethal to tumor
cells (HAMLET)”, this complex is currently under investigation in an
early phase clinical trial.^[ [50]^7 ^] The monomeric form of
alpha‐lactalbumin was reported to be inactive for anticancer
activity,^[ [51]^8 ^] but alpha‐lactalbumin could nevertheless provide
a good lead for developing a computer‐generated peptide library,
followed by deployment of in silico ACP screening and experimental
validation of ACP candidates.
This study aimed to establish a workflow for discovery of novel ACPs
from potential alpha‐lactalbumin peptides. An R‐based script was
created to generate all possible peptides that could be derived from
the amino acid sequence of alpha‐lactalbumin. The ACP physicochemical
and structural properties, as well as the ACP probabilities predicted
by three different ML models, were used to prioritize the peptide
candidates from the 2688‐peptide library. Anticancer activities of the
candidate peptides were experimentally measured in vitro against human
cancer cell lines derived from lung, breast, colon, brain, and leukemia
cancers. Their hemolytic side effects were tested using a red blood
cell (RBC) lysis assay. Sequential window acquisition of all theorical
fragment ions (SWATH)‐proteomics was performed to determine a potential
mechanism of action of the peptide candidate, followed by functional
validation. This workflow resulted in discovery of a novel ACP that
induces autophagy‐associated cell death in a human lung cancer cell
line.
2. Results
This study investigated the utility of computer‐based peptide library
generation coupled with in silico ACP screening^[ [52]^5 ^] for
prioritizing peptide candidates. The top predicted ACP candidates were
tested in vitro for actual anticancer activity against various cancer
cell lines and also evaluated for potential toxicity using a RBC lysis
assay ex vivo. The overall workflow of this study is illustrated in
Figure [53] 1 .
Figure 1.
Figure 1
[54]Open in a new tab
The entire workflow of ACP discovery from a computer‐generated peptide
library based on alpha‐lactalbumin, in vivo ACP screening, and in vitro
validation experiments. aa: amino acid; ACP: anticancer peptide.
2.1. Alpha‐Lactalbumin Inspired Peptide Library Generation and In Silico ACP
Screening
Human alpha‐lactalbumin (UniProt ID: [55]P00709) is comprised of
142 amino acids. Here, 2688 unique peptides, ranging from 5 to 25 amino
acids per peptide with stepping 1 amino acid each were generated from
alpha‐lactalbumin by using the peptide generator ALA2Pept. The
distributions of the 2688 peptides based on physicochemical,
structural, and ML probabilities are shown in Figure [56] 2 . As
expected, the numbers of computer‐generated peptides were inverse to
the peptide length (Figure [57]2a). The net charge of peptides ranged
from −6 to +3, with the electrically neutral peptides being the
majority of this library (Figure [58]2b). Most known ACPs have a
positive charge,^[ [59]^2 , [60]^5 ^] so the initial focus was on the
peptides with +3 charge. A secondary structure prediction was performed
on this subset of peptides with a +3 net charge. Of the 25 peptides
(≈1% of the total 2688 peptides) with +3 net charge, 19 were predicted
to have α‐helical conformation (ACP‐preferred structure) and 6 had a
coil/sheet structure (Figure [61]2c). Three distinct ML models
(ACPred‐FL, Anticp2.0, and mACPpred) revealed distinct, but
complementary, distributions of the predicted ACPs (Figure [62]2d). The
ACP candidates were ranked by the consensus score (geometric mean) of
the three ACP probabilities, and filtered against their physicochemical
and structural properties. All peptide sequences and the results of in
silico ACP screening of 2688 peptides are provided in Table [63]S1,
Supporting Information.
Figure 2.
Figure 2
[64]Open in a new tab
The distributions of various characteristics of the 2688 peptides
generated from alpha‐lactalbumin. a) Numbers of peptides by length
(5–25 amino acid residues). b) Numbers of peptides by the
physicochemical charge property. c) Secondary structure of 25 peptides
containing +3 net charges predicted by three ML models. d) ACP
probabilities as predicted by three distinct ML models, that is,
ACPred‐FL, Anticp2.0, and mACPpred.
2.2. Cancer Cytotoxic Screening Indicates ALA‐A2 as the Positive Hit
The list of candidates selected for the peptide synthesis and
experimental validations is shown in Table [65] 1 and Figure [66] 3a.
The top two peptide candidates, ALA‐A1 and ALA‐A2, are indicated by the
highest consensus score peptides with non‐redundant sequence, having
≥3 positive charges, and containing alpha‐helical secondary structure.
Based on the similarity of their amino acid sequences, ALA‐A3 and
ALA‐A4 are included as the lower‐ranked peptide comparators of
ALA‐A1 and ALA‐A2, respectively. Figure [67]3b illustrates the
positions of these peptides within the 3D structure of
alpha‐lactalbumin (PDB: 1b9o) generated by AlphaFold.^[ [68]^9 ^]
ALA‐A1 and ALA‐A3 peptides are derived from the linear N‐terminal
chain, while ALA‐A2 and ALA‐A4 are presented as part of the external
surface of the folded alpha‐lactalbumin.
Table 1.
Physicochemical, structural, and machine learning properties of
selected peptide ACP candidates
ID Amino acid sequence Length Charge Structure Machine learning model
Consensus score Prediction
ACPred‐FL AntiCp 2.0
mACP
pred
ALA‐A1 RFFVPLFLVGILFPAILAKQFTK 23 3+ coil–helix–coil–helix 0.980 0.620
0.964 0.836 ACP
ALA‐A2 KLWCKSSQVPQSR 13 3+ helix–coil 0.992 0.480 0.798 0.724 ACP
ALA‐A3 RFFVPLFLVGILFPAILAKQFTKC 24 3+ helix–coil–helix 0.980 0.590
0.977 0.826 ACP
ALA‐A4 LFQISNKLWCKSSQVPQSRN 20 3+ coil 0.944 0.460 0.075 0.319 non‐ACP
BMP‐S6 (positive control)^[ [69]^5 , [70]^10 ^] FKCRRWQWRMKKLGAPSITCVR
22 7+ coil–helix 0.606 1.000 0.9848 0.842 ACP
[71]Open in a new tab
Figure 3.
Figure 3
[72]Open in a new tab
In vitro cytotoxicity of the synthetic peptides against five human
cancer cell lines. a) The amino acid sequences with the predicted
secondary structures, net charges, and consensus scores of all
synthetic peptides included for experimental validations. b) The
positions of ALA‐A1, ‐A2, ‐A3, and ‐A4 peptides (black) on the
predicted structure (blue) and the crystal structure (pink; PDB:1b9o)
of alpha‐lactalbumin as visualized by UCSF chimera 1.15.^[ [73]^11 ^]
Note that ALA‐A3 and ALA‐A4 have amino acid sequences that partially
overlap with ALA‐A1 and ALA‐A2, respectively. c) Cancer cytotoxicity
screening against five human cancer cell lines, including lung cancer
(A549), colon cancer (HT29), breast cancer (MDA‐MB‐231), neuroblastoma
(SH‐SY5Y), and leukemia (K562) cells. The synthetic peptides (200 µm)
were incubated with cancer cells at 37 °C for 24 h before measuring
cell viability (MTT assay). Only ALA‐A2 exhibited a selective and
substantial anticancer effect against A549 and HT29 cells. This
experiment was performed in three biological replicates. *p < 0.05;
N‐ter: N‐terminal; C‐ter: C‐terminal.
To determine whether the in silico predicted candidates actually
exhibit anticancer effects in vitro, the synthetic peptides were tested
against five cancer cell lines (Figure [74]3c). BMP‐S6, a
bovine‐derived ACP from previous studies,^[ [75]^5 , [76]^10 ^] was
included as the positive control of this study. ALA‐A2 (at 200 µm)
selectively induced cancer cytotoxicity of A549 human lung epithelial
adenocarcinoma (59.7 ± 1.1% cell viability) and HT29 human colorectal
adenocarcinoma (55.3 ± 2.7% cell viability) cell lines, but did not
affect K562 human lymphoblastic leukemia, MDA‐MB‐231 human breast
cancer, and SH‐SY5Y human neuroblastoma cell lines. ALA‐A1, ALA‐A3, and
ALA‐A4 did not exhibit a substantial anticancer effect against the five
cell lines examined. BMP‐S6 was the only peptide that exhibited cancer
cytotoxicity against K562 leukemic cells, consistent with previous
studies.^[ [77]^5 , [78]^10 ^] Accordingly, ALA‐A2 was designated the
positive hit from in silico and in vitro ACP screening, and subjected
to further validation.
2.3. Dose–Response Relationship, Hemolytic Side Effect, and Cell
Internalization of ALA‐A2
The dose dependency of ALA‐A2 against A549 and HT29 cells was measured
at twofold serial dilutions from 0 to 400 µm. Based on this dose
testing range, the IC50 of A549 is ≈300 µm and the IC50 of HT29 is
extrapolated to be greater than 1000 µm. The higher concentration of
ALA‐A2 doses resulted in lower cell viability of A549, and, to a lesser
extent, of HT29 cells (Figure [79] 4a).
Figure 4.
Figure 4
[80]Open in a new tab
Dose‐dependent cytotoxicity, hemolytic side effect, and cell
penetrating capability of ALA‐A2 peptide. a) ALA‐A2 dose–response
relationship in A549 lung cancer and HT29 colon cancer cell lines. The
cells were incubated with twofold dilutions of ALA‐A2 from 400 to 0 µm
for 24 h, and then cell viability was measured by the MTT assay. b) RBC
lysis assay for ALA‐A2 peptide toxicity testing. Human red blood cells
were incubated with the peptides for 1 h. Triton X‐100 at a final
concentration of 1% was used as a positive control. c) Confocal
microscopy showed that FITC‐tagged ALA‐A2 peptide (green) could
internalize into A549 lung cancer cells without altered membrane
integrity, whereas Triton X‐100 (0.5% v/v; the positive control) could
permeabilize cellular membrane allowing PI internalization (red).
Hoechst 33342 (blue) was used for the nuclear counter staining. Scale
bar: 50 µm.
Hemolysis is a major concern during the development of peptide‐based
drugs; hemolytic peptides do not progress to clinical trials.^[ [81]^12
^] Thus, hemolytic activity of the ACP ALA‐A2 was compared with those
of other peptides without anticancer activity. Figure [82]4b
illustrates that neither ALA‐A2 (200 µm) nor the other peptides exhibit
hemolytic effects against normal RBCs after 1 h incubation. The percent
hemolysis was ALA‐A1, 5.6% ± 0.6%; ALA‐A2, 7.8% ± 10.2%; ALA‐A3,
1.3% ± 1.2%; ALA‐A4, 0.7% ± 0.6%; and BMP‐S6, 6.0% ± 9.2%. Note that 1%
Triton X‐100, the positive control, completely lysed the normal RBCs
within 1 h.
The anticancer mechanism of ALA‐A2 was investigated in A549 lung cancer
cells. Membranolysis and cell penetrating peptides are both common, but
distinct, modes of ACP action.^[ [83]^2 ^] To explore the mode of
action of ALA‐A2, A549 lung cancer cells were treated with FITC‐tagged
ALA‐A2 peptide (200 µm) for 24 h, followed by confocal microscopy.
Figure [84]4c illustrates that the signals of FITC‐tagged ALA‐A2 were
present in the intracellular compartments of A549 cells. ALA‐A2 did not
disrupt membrane integrity. Propidium iodide (PI) is internalized
following treatment with Triton X‐100, the positive control for a cell
permeabilized agent,^[ [85]^13 ^] whereas it is not internalized
following treatment with ALA‐A2. Thus, ALA‐A2 penetrates the cell and
may be the trigger for cell death‐related signaling of A549 lung cancer
cells.
2.4. ALA‐A2 Triggered Autophagy‐Mediated Cell Death in A549 Lung
Adenocarcinoma Cells
SWATH‐proteomics was performed to investigate potential mechanisms of
the ALA‐A2 anticancer effect in A549 lung cancer cells. Of 286 proteins
detected and quantified across 18 data independent acquisition (DIA)
runs (full data sets in Table [86]S2, Supporting Information), heatmap
and volcano plot revealed 38 differentially expressed proteins
(12 upregulated and 26 downregulated) in ALA‐A2 treated A549 cells (as
illustrated in Figure [87] 5a,b and Table [88] 2 ). Reactome pathway
enrichment analysis was performed to deduce biological meaning from
this differential protein signature. Figure [89]5c demonstrates that
the autophagy pathways, particularly chaperone‐mediated autophagy, were
the best matched pathway with the lowest false discovery rate (details
in Table [90]S3, Supporting Information). This functional match mostly
reflected four downregulated proteins: ubiquitin‐60S ribosomal protein
L40 (fold‐change = 0.41, p = 0.027), heat shock 70 kDa protein 8
(fold‐change = 0.25, p = 0.006), and heat shock protein HSP90 alpha
(fold‐change = 0.26, p = 0.007) and beta (fold‐change = 0.40,
p = 0.004) (Figure [91]5a,b and Table [92]2 and Table [93]S3,
Supporting Information). To confirm this deduction, the ability of
ALA‐A2 peptide (200 µm) to induce autophagy was tested. As shown in
Figure [94]5d, ALA‐A2 treatment significantly increased the autophagic
activity in A549 cells (1.5‐fold of untreated cells) comparable to that
of rapamycin, a positive control known to induce autophagy. Taken
together, these findings identified ALA‐A2 as a cell penetrating
peptide that induces autophagy‐mediated cell death in A549 human lung
adenocarcinoma cells.
Figure 5.
Figure 5
[95]Open in a new tab
SWATH‐proteomic and functional analyses suggest that cell death is
caused by autophagic induction. a) Heatmap and b) volcano plot
identified the significantly differentially expressed proteins between
ALA‐A2 peptide‐treated versus untreated A549 cells (n = 9 per group;
three technical replicates of three biological specimens) (full
expression datasets in Table [96]S2, Supporting Information). c) This
reactome was subjected to biological pathway analyses. The stronger
yellow accent indicates lower false‐discovery rate. Chaperone mediated
autophagy was predicted as the most significant pathway elicited in
ALA‐A2 peptide‐treated A549 cells (details in Table [97]S3, Supporting
Information). d) Autophagy activity assay. Rapamycin, an autophagic
inducer, was the positive control. This experiment was performed in
three biological replicates. *p < 0.05.
Table 2.
Significantly differentially expressed proteins between ALA‐A2 treated
versus untreated A549 lung adenocarcinoma cells
SwisProt ID Gene name Protein name %Coverage Number of matched peptides
MW
[kDa]
pI
Fold‐change [treated/
untreated]
p‐value
Proteins whose expression was upregulated in ALA‐A2 treated cells
[98]Q13148 TADBP TAR DNA‐binding protein 43 2.90 1 44.74 5.85 4.07
0.0348
[99]Q9UKM9 RALY
RNA‐binding protein Raly
(Autoantigen p542)
6.86 1 32.463 9.20 3.57 0.0209
[100]P20073 ANXA7 Annexin A7 8.61 1 52.739 5.52 3.29 0.0011
[101]O00299 CLIC1 Chloride intracellular channel protein 1 11.20 2
26.923 5.09 3.02 0.0212
[102]P09525 ANXA4 Annexin A4 18.81 1 35.883 5.84 2.98 0.0074
[103]Q02878 RL6 60S ribosomal protein L6 19.10 2 32.728 10.59 2.87
0.0319
[104]P26639 SYTC Threonine‐tRNA ligase 1 4.70 1 83.435 6.23 2.61 0.0100
[105]P50991 TCPD T‐complex protein 1 subunit delta 13.36 3 57.924 8.13
2.39 0.0255
[106]Q99729 ROAA Heterogeneous nuclear ribonucleoprotein A/B 5.42 1
36.225 8.21 2.35 0.0248
[107]O00410 IPO5 Importin‐5 6.38 3 123.63 4.83 2.22 0.0010
[108]P42704 LPPRC Leucine‐rich PPR motif‐containing protein 14.92 11
157.905 5.53 2.06 0.0091
[109]P14923 PLAK Junction plakoglobin 7.52 2 81.745 5.75 2.05 0.0041
Proteins whose expression was downregulated in ALA‐A2 treated cells
[110]Q6FI13 H2A2A Histone H2A type 2‐A 62.31 19 14.095 10.90 0.48
0.0438
[111]Q9NZL4 HPBP1 Hsp70‐binding protein 1 8.36 2 39.303 5.13 0.48
0.0067
[112]P49368 TCPG T‐complex protein 1 subunit gamma 10.83 3 60.534 6.10
0.46 0.0336
[113]P14618 KPYM Pyruvate kinase PKM 54.43 32 57.937 7.95 0.43 0.0258
[114]P37837 TALDO Transaldolase 17.80 3 37.54 6.36 0.43 0.0089
[115]P62987 UBA52 Ubiquitin‐60S ribosomal protein L40 52.34 4 14.728
6.56 0.41 0.0272
[116]P08238 HSP90AB1 Heat shock protein HSP90 beta 37.15 25 83.264 4.96
0.40 0.0039
[117]P14625 ENPL Endoplasmin 17.68 10 92.469 4.73 0.38 0.0301
[118]P29401 TKT Transketolase 39.33 20 67.878 7.58 0.35 0.0004
[119]P05783 K1C18 Keratin, type I cytoskeletal 18 53.02 42 48.058 5.34
0.35 0.0149
[120]P08727 K1C19 Keratin, type I cytoskeletal 19 23.75 7 44.106 5.05
0.34 0.0327
[121]P84103 SRSF3 Serine/arginine‐rich splicing factor 3 12.80 1 19.33
11.64 0.33 0.0102
[122]P16402 H13 Histone H1.3 25.34 2 22.35 11.02 0.32 0.0227
[123]P04406 G3P Glyceraldehyde‐3‐phosphate dehydrogenase 60.00 32
36.053 8.58 0.32 0.0088
[124]P26641 EF1G Elongation factor 1‐gamma 20.14 4 50.119 6.27 0.31
0.0123
[125]O60218 AK1BA Aldo‐keto reductase family 1 member B10 71.52 26
36.02 7.66 0.29 0.0120
[126]P07437 TBB5 Tubulin beta chain 53.15 39 49.671 4.78 0.27 0.0041
[127]P62805 H4 Histone H4 57.28 15 11.367 11.36 0.26 0.0302
[128]Q06830 PRDX1 Peroxiredoxin‐1 29.65 10 22.11 8.27 0.26 0.0190
[129]P07900 HSP90AA1 Heat shock protein HSP90 alpha 37.02 24 84.66 4.94
0.26 0.0069
[130]P22626 ROA2 Heterogeneous nuclear ribonucleoproteins A2/B1 24.93 3
37.43 8.97 0.25 0.0281
[131]P11142 HSPA8 Heat shock 70 kDa protein 8 28.79 21 70.898 5.37 0.25
0.0056
[132]P08865 RSSA 40S ribosomal protein SA 17.97 3 32.854 4.79 0.25
0.0285
[133]Q5VTE0 EF1A3 Elongation factor 1‐alpha‐like 3 35.93 19 50.185 9.15
0.15 0.0061
[134]P63261 ACTG Actin, cytoplasmic 2 63.20 53 41.793 5.31 0.09 0.0030
[135]P21359 NF1 Neurofibromin 1.02 1 319.372 7.12 0.07 0.0030
[136]Open in a new tab
3. Discussion
The novel approach to ACP identification described in this study
demonstrates that a computer‐generated peptide library with downstream
in silico ACP screening and in vitro experimental validation could
accelerate anticancer research and development. Alpha‐lactalbumin, a
milk protein known for anticancer properties,^[ [137]^7 , [138]^14 ^]
inspired this study design. The library of 2688 distinct peptides
containing 5–25 amino acids in length were generated from the sequence
information of alpha‐lactalbumin rapidly (Figure [139]2). As different
ML models were established upon various data sets, algorithms, and
tuning parameters, their predictive performance can vary.^[ [140]^15 ^]
Rather than relying on only one ML model, we utilized the consensus
score, which is a geometric mean of ACP probabilities of three
different ML models, plus the physicochemical and structural
properties, to obtain better ACP candidates. After in silico ACP
screening of 2688 peptides, a few top‐ranked candidates were achieved
within days for the custom‐ordered peptide synthesis and in vitro
experimental validation (Figures [141]3, [142]4, [143]5). This study
identified ALA‐A2 as a novel ACP in a time‐ and cost‐effective manner.
ALA‐A2 was relatively selective to lung adenocarcinoma cells, acting as
a cell penetrating peptide to induce autophagy‐mediated cell death
without provoking hemolysis. The sequence of ALA‐A2 as KLWCKSSQVPQSR
was predicted for its half‐life in blood using PlifePred
([144]https://webs.iiitd.edu.in/raghava/plifepred/, accessed on October
30, 2022) that remained ≈20 min in blood.^[ [145]^16 ^] Because of a
relatively short half‐life of ALA‐A2 peptide, it should be modified,
for example, by amidation at N‐ or C‐terminus,^[ [146]^4 , [147]^17 ^]
to improve the stability of ALA‐A2 for further clinical applications.
Moreover, the 200 µm ALA‐A2 concentration was chosen to observe lung
cancer cell responses because the effect of this dose was obvious on
A549 cells. This high dose has no hemolytic effect (Figure [148]4b)
while exhibiting lung cancer cytotoxicity through autophagy‐induced
cell death (Figure [149]5).
Brisuda et al.^[ [150]^7a ^] reported that the α‐helical peptide
(residues 1–39) derived from alpha‐lactalbumin in complexes with oleate
(so‐called alpha1‐oleate) had a tumoricidal effect against bladder
cancer, but beta‐sheet‐oleate (residue at 40–80) complexes lacked
anticancer properties. Both ALA‐A1 and ALA‐A3 partially overlapped with
the alpha1‐oleate, while ALA‐A2 and ALA‐A4 were part of the
beta‐sheet‐oleate. However, the data reported herein indicate that only
ALA‐A2 exhibits the anticancer effect. This discrepancy may be due to
the absence of oleate, the shorter length of ALA‐peptides, and the
different cancer cell lines examined in this study. Our computational
workflow did not generate any peptide longer than 25 amino acid
residues. By default, the bare peptides of alpha1‐oleate (39 amino
acids in length) and beta‐sheet‐oleate (41 amino acids in length) were
not evaluated in this study. The 5 amino acid peptide length is the
minimal number that mass spectrometry can detect and is the shortest
peptide length for ACP prediction available (mACPred [≥5 amino acids],
Anti‐CP2.0 [4–50 amino acids], and ACPred‐FL [10–50 amino acids]).
However, the shortest peptide in the protein data bank is seven amino
acids long as this size is needed to form the crystalline structure of
the alpha‐helical peptide.^[ [151]^18 ^] A peptide length of 25 amino
acids or greater may suffer from peptide aggregation and poor
solubility. Moreover, the lengths of peptides containing 13–24 amino
acid residues were effectively similar in cell uptake.^[ [152]^19 ^]
Thus, 25 amino acids length fulfilled this requirement.
The mechanism of ALA‐A2 is as a cell penetrating peptide with the
ability to induce autophagy‐mediated cancer cell death.^[ [153]^20 ^]
ALA‐A2 passes through the plasma membrane to the intracellular
compartment of A549 lung cancer cells (Figure [154]4c), and disturbs
cancer cell homeostasis. ALA‐A2‐treated A549 cells exhibit
downregulation of chaperone proteins involved in autophagic processes,
including heat shock 70 kDa protein 8 (HSPA8), heat shock protein
HSP90 alpha (HSP90AA1), and beta (HSP90AB1) (Figure [155]5 and
Table [156]2). Although the exact intracellular targets of ALA‐A2 have
yet to be determined, the mode of anticancer actions of ALA‐A2 has
shared with previous studies.^[ [157]^21 ^]
Ding et al.^[ [158]^21a ^] showed that the engineered peptide
Trx‐pHLIP‐Beclin1 could induce autophagic cell death in SKOV3 ovarian
cancer cells. Pelaz et al.^[ [159]^21b ^] reported that a designed
peptide, TAT‐Cx43266‐283, deregulated autophagy and induced
glioblastoma stem cell death. Recently, Dent et al.^[ [160]^22 ^]
reported that the formation of autophagosomes was induced in
HCT116 colon cancer following the inhibition of ATPase activities of
HSP90 and HSP70. Thus, ACP‐induced autophagy cancer cell death is
probably explained, at least in part, by disturbing the expression and
function of heat shock proteins. This is consistent with a report by
Zhao et al.,^[ [161]^21c ^] in which a novel HSP90 inhibitor, DPB,
could inhibit A549 lung cancer growth via inducing apoptosis and
autophagy. HSP90 is known as the cancer chaperone which plays a crucial
role in maintaining the stability of oncogenic drivers in lung
adenocarcinoma.^[ [162]^23 ^] Suppression of HSP90 leads to the
degradation of oncogenic drivers and a loss of lung cancer cell
viability.^[ [163]^23 ^] Overexpression of HSP90 has been correlated
with a poorer prognosis and chemoresistance in patients with
non‐small‐cell lung carcinoma.^[ [164]^23 , [165]^24 ^] The findings
reported herein support ALA‐A2 inducing autophagy through
downregulation of HSP90 expression in lung adenocarcinoma
(Figure [166]5 and Table [167]2). Future studies on the molecular
interaction between ALA‐A2 and HSP90 would be promising. Ultimately,
ALA‐A2 peptide could be tested as an adjuvant therapy to attenuate
chemoresistance in non‐small‐cell lung carcinoma.
4. Conclusion
This study describes a novel strategy for searching for new ACPs. To
validate this strategy, ALA‐A2 was discovered to be a new ACP by
integrating the computational peptide library of alpha‐lactalbumin with
downstream in silico ACP screening and in vitro experimental
validation. ALA‐A2 specifically inhibited lung adenocarcinoma cells
through autophagy induction and cancer cell death. This approach, when
applied to an expanded computational peptide library based on a large
number of proteins, has great potential to minimize the time and cost
of therapeutic peptide research.
5. Experimental Section
Cell Culture
All cell lines were obtained from the American Type Culture Collection
(ATCC, Manassas, VA, USA). The A549 lung adenocarcinoma (ATCC
CCL‐185TM), MDA‐MB‐231 triple‐negative breast adenocarcinoma (ATCC
HTB‐26TM), and SH‐SY5Y neuroblastoma (ATCC CRL‐2266TM) cell lines were
cultured in Dulbecco's Modified Eagle's Medium (DMEM)‐high glucose
(Gibco, Thermo Fisher Scientific, MA, USA). HT29 colon adenocarcinoma
(ATCC HTB‐38) cell lines were maintained in DMEM/Nutrient Mixture F‐12
(Gibco). K562 chronic myeloid leukemia (ATCC CRL‐3344) were grown in
Roswell Park Memorial Institute Medium 1640 (Gibco). The cells were
cultured in medium supplemented with 10% fetal bovine serum (Gibco) and
1× penicillin/streptomycin (Gibco) and incubated in a 5%
CO[2] incubator with saturated humidity.
Computational Generation of Alpha‐Lactalbumin‐Inspired Peptide Library
The amino acid sequence of human alpha‐lactalbumin retrieved from
UniProt ID [168]P00709 (last accessed on 14 September 2020) served as
the template for generating the in silico peptide library. A peptide
generator, ALA2Peptide, was coded to generate all possible peptides of
alpha‐lactalbumin ranging in the optimal length of ACP (5–25 amino
acids)^[ [169]^2 ^] with stepping 1 amino acid each (from N‐terminal to
C‐terminal). This program can be accessed via
[170]https://github.com/schuti/ALA2Pept. The output peptide library
containing 2688 distinct peptide sequences is provided in Table
[171]S1, Supporting Information.
In Silico ACP Screening
The in silico ACP screening was performed as described^[ [172]^5 ^]
with a minor modification. Three different ML models: mACPpred (a
support vector machine‐based predictor;
[173]http://www.thegleelab.org/mACPpred/),^[ [174]^25 ^] ACPred‐FL (a
sequence‐based predictor; [175]http://server.malab.cn/ACPred-FL/),^[
[176]^26 ^] and AntiCP2.0 (an ensemble tree classifier‐based predictor;
[177]https://webs.iiitd.edu.in/raghava/anticp2/),^[ [178]^27 ^] were
used to predict physicochemical and anticancer properties. The
geometric mean of ACP probabilities was calculated from the three ML
models to serve as the consensus score as well as net charge
prediction. The PEP‐FOLD3 de novo peptide structure prediction
webserver^[ [179]^28 ^] was used to predict the secondary structure of
the peptides with ≥3 positive charges (25 out of 2688). Top‐ranked
peptide candidates (ALA‐A1, ALA‐A2) were defined as the top ranked
peptides by the consensus score with non‐redundant sequence, having
≥3 positive charges, and containing alpha‐helical secondary structure.
ALA‐A3 and ALA‐A4, the lower‐ranked positively charged peptides with
redundant sequence to ALA‐A1 and ALA‐A2, were included as comparators.
BMP‐S6, a known bovine milk‐derived ACP,^[ [180]^5 , [181]^10 ^] served
as the positive ACP control, while the untreated condition was the
blank (negative) control of this experiment.
Synthetic Peptides
Five synthetic peptides (purity >98%) were custom‐made by GL Biochem
(GL Biochem [Shanghai] Ltd., Shanghai, China). These peptides included
ALA‐A1, NH[2]‐RFFVPLFLVGILFPAILAKQFTK‐COOH; ALA‐A2,
NH[2]‐KLWCKSSQVPQSR‐COOH; ALA‐A3, NH[2]‐RFFVPLFLVGILFPAILAKQFTKC‐COOH;
ALA‐A4, NH[2]‐LFQISNKLWCKSSQVPQSRN‐COOH; and BMP‐S6 (the positive
control),^[ [182]^5 , [183]^10 ^] NH[2]‐FKCRRWQWRMKKLGAPSITCVR‐COOH.
The peptides were resuspended in fresh media to the final concentration
before use.
In Vitro Cancer Cytotoxicity Screening
The protocol previously described by Chiangjong et al. and Baindara
et al.^[ [184]^5 , [185]^29 ^] was modified for use in determining the
cytotoxicity of peptides on the various cancer types. Adherent cells
were seeded at 1 × 10^4 cells/well into a 96‐well plate 24 h prior to
replacing the media containing the peptide of interest. Suspended cells
were seeded at 1 × 10^4 cells/well in the media containing the peptide
of interest. Five human cell lines, including lung cancer (A549), colon
cancer (HT29), breast cancer (MDA‐MB‐231), neuroblastoma (SH‐SY5Y), and
leukemia (K562) cells, were treated with the culture media containing
the peptides at 200 µm for 24 h. Cell viability was then determined
using Cell Proliferation Kit I, MTT (Roche Diagnostics GmbH, Mannheim,
Germany): MTT labeling reagent (10 µL) was added to each well and
incubated at 37 °C in a 5% CO[2] atmosphere for 4 h. After incubation,
100 µL solubilizing solution was added to each well and incubated at
37 °C in 5% CO[2] overnight. The absorbance at 570 nm (A [570]) was
measured for each well. The percentage of cell viability was calculated
as:
[MATH:
% Ce
ll Viability=[A
570 of trea
ted cells − A570 of<
mtext> mediaA570 of untreated cells−A570 of media] × 100 :MATH]
(1)
To evaluate dose‐dependency, ≈1 × 10^4 cells/well were cultured in
media containing ALA‐A2 at 0, 6.25, 12.5, 25, 50, 100, 200, and 400 µm
in a 96‐well plate for 24 h. Cell viability was determined by the MTT
assay, as above.
Ex Vivo Hemolysis Assay
The hemolysis assay was as described previously.^[ [186]^30 ^] After
informed consent, peripheral blood (3 mL) was collected from three
healthy individuals into heparin tubes (BD Biosciences, NJ, USA). The
specimen was centrifuged at 2000 × g for 5 min at room temperature, and
the plasma was discarded. The RBCs were washed twice with PBS and then
resuspended in 0.9% normal saline solution at a volume equivalent to
that of the original plasma. Approximately 0.8% v/v RBC suspension was
incubated with the peptides at 200 µm (100 µL final volume) in a
U‐bottom 96‐well plate at room temperature for 1 h with agitation. 1%
v/v Triton X‐100 was applied as a positive control. The untreated
condition served as the blank control. After 1 h incubation, the
supernatants were collected by centrifuging at 2000 × g for 5 min at
room temperature. The absorbance of hemoglobin in the supernatant was
measured at 415 nm. The hemolysis percentage was calculated as follows:
[MATH:
% He
molysis = [sample absorbance<
mo>−blank controlpositive control<
mo>−blank control] × 100 :MATH]
(2)
This study was conducted in accordance with the Declaration of
Helsinki, and the protocol was approved by the Human Research Ethics
Committee, Faculty of Medicine Ramathibodi Hospital, Mahidol University
(Protocol ID: MURA2020/759; with approval of amendment on May 5, 2021).
Confocal Microscopy
In 6‐well plates containing a glass cover slip, A549 lung
adenocarcinoma and HT29 colon adenocarcinoma cells were seeded at
2 × 10^5 cells/2 mL medium/well and incubated for 24 h. The medium was
replaced with fresh medium with or without 200 µm FITC‐tagged
ALA‐A2 peptide and incubated at 37 °C in 5% CO[2] for 24 h. All cells
were then incubated with PI (1:2000) and Hoechst 33342 (1:5000) in the
culture medium at 37 °C in 5% CO[2] for 15 min. The cells were washed
with PBS twice before being fixed with 4% paraformaldehyde (PFA) in
PBS. The fixed cells were washed twice with PBS and mounted in 20%
glycerol in PBS. After being fixed in PFA, the positive PI staining
control cells were permeabilized with 0.5% v/v Triton X‐100 for 10 min.
Thereafter, the cells were stained with PI for 15 min at room
temperature, washed with PBS, and then mounted with 20% glycerol in
PBS. Internalization was visualized by confocal fluorescence microscopy
(Nikon Instruments, Inc., Melville, NY, USA).
SWATH‐Proteomic and Bioinformatic Analysis
Targeted label‐free proteomic analysis using SWATH/DIA was performed as
described previously^[ [187]^31 ^] to explore the potential mechanisms
of ALA‐A2‐induced A549 lung adenocarcinoma cell death. In brief,
A549 lung adenocarcinoma cells were seeded at a density of
1 × 10^5 cells/well in a 24‐well plate. After overnight incubation, the
media were changed with fresh media with or without 200 µm ALA‐A2 and
incubated at 37 °C in 5%CO[2] for 24 h. The cells were washed twice
with PBS before lysis with Laemmli lysis buffer (62.5 mm Tris‐HCl, pH
6.8, 10% v/v glycerol, 2% w/v SDS, and 2.5% v/v beta‐mercaptoethanol).
Total protein was quantified using the Bradford protein assay (Bio‐Rad,
Hercules, USA). 10 µg of protein was resolved by 12% SDS–PAGE before
staining with Coomassie brilliant blue G‐250, and the gel was cut into
small pieces for in‐gel tryptic digestion. An amount of peptide sample
corresponding to 2.5 µg of total protein was resolved in an Eksigent
nanoLC ultra nanoflow high performance liquid chromatography in tandem
with a TripleTOF 6600+ mass spectrometer (ABSciex, Toronto, Canada) set
for information‐dependent acquisition (IDA) and DIA modes. The peptides
were loaded onto a C18 column trap (Nano Trap RP‐1, 3 µm 120 Å,
10 mm × 0.075 mm; Phenomenex, CA, USA) at a flow rate of 3 µL min^−1 of
0.1% formic acid in water for 10 min to desalt and concentrate the
sample, which was then resolved by HPLC using a stationary phase of a
C18 analytical column (bioZen Peptide Polar C18 nanocolumn,
75 µm × 15 cm, particle size 3 µm, 120 Å; Phenomenex) with mobile phase
gradients at a flow rate of 300 nL min^−1 of 3–30% acetonitrile/0.1%
formic acid for 60 min, 30–40% acetonitrile/0.1% formic acid for
10 min, 40–80% acetonitrile/0.1% formic acid for 2 min, 80%
acetonitrile/0.1% formic acid for 6 min, 80–3% acetonitrile/0.1% formic
acid for 2 min, and 3% acetonitrile/0.1% formic acid for 25 min. The
eluate was ionized and sprayed into the mass spectrometer using
OptiFlow Turbo V Source (Sciex). Ion source gas 1, ion source gas 2,
and curtain gas were set at 19, 0, and 25 vendor arbitrary units,
respectively. The interface heater temperature was 150 °C and ion spray
voltage was 3.3 kV.
Mass spectrometry was operated in the positive ion mode set for
3500 cycles per 105 min gradient elution. Each cycle performed one time
of flight (TOF) scan (250 ms accumulation time, 350–1250 m/z window
with a charge state of +2) followed by IDA of the 100 most intense
ions, while the minimum MS signal was set to 150 counts. The MS/MS scan
was operated in high sensitivity mode with 50 ms accumulation time and
50 ppm mass tolerance. Former MS/MS candidate ions were excluded for a
period of 12 s after their first occurrence to reduce the redundancy of
identified peptides. DIA mode was performed in a range of 350–1500 m/z
using a predefined mass window of 7‐m/z with an overlap of 1‐m/z for
157 transmissible windows. MS scan was set at 2044 cycles, where each
cycle performed one TOF‐MS scan type (50 ms accumulation time across
100–1500 precursor mass range) acquired in every cycle for a total
cycle time of 3.08 s. MS spectra of 100–1500 m/z were collected with an
accumulation time of 96 ms per SWATH window width. Resolution for
MS1 was 35 000 and SWATH‐MS2 scan was 30 000. Rolling collision energy
mode with collision energy spread of 15 eV was applied. The IDA and DIA
data (.wiff) were recorded by Analyst‐TF v.1.8 software (ABSciex).
A total of 18 wiff files of IDA experiments (two groups;
three biological replicates per group; three technical replicates per
biological sample) were combined and searched using Protein Pilot
v.5.0.2.0 software (ABSciex) against the Swiss‐Prot database (UniProtKB
2022_01) Homo sapiens (20385 proteins in database) with the searching
parameters as follows; alkylation on cysteine by iodoacetamide, trypsin
enzymatic digestion, one missed cleavage allowed, monoisotopic mass,
and 1% false discovery rate. The group file (Protein Pilot search
result) was loaded into SWATH Acquisition MicroApp v.2.0.1.2133 in
PeakView software v.2.2 (Sciex) to generate a spectral library. The
maximum number of proteins was set as the number of proteins identified
at 1% global FDR from fit. RT alignment was performed by the high
abundance endogenous peptides covering the chromatographic range. SWATH
data extraction of 18 DIA files (two groups; three biological
replicates per group; three technical replicates per biological sample)
was performed by SWATH Acquisition MicroApp (Sciex) using the following
parameters; 5‐min extraction window, 25 peptides/protein,
6 transitions/peptide, excluding shared peptides, peptide confidence
>99%, FDR <1%, and XIC width of 20 ppm. SWATH extraction data,
including the identities and quantities of peptides and proteins, was
exported into an Excel file for further analysis.
Comparative proteomic analysis was performed by using R program as
described previously,^[ [188]^31b ^] including log2 transformation, VSN
normalization, missing value imputation by median, differential
expression analysis, heatmap, and volcano plot at the thresholds of 2×
fold change and p‐value <0.05. The pathway enrichment analysis was
analyzed by Reactome v.78^[ [189]^32 ^] (last accessed on February 2,
2022) using the differential expressed proteins as the input, where a
matched pathway with FDR <0.05 was considered statistically
significant.
Autophagy Activity Assay
A549 cells (1 × 10^4 cells/well in a 96‐well plate) were treated with
200 µm ALA‐A2. Treated cells were maintained for 24 h before
determining autophagy activity using an autophagy assay kit (ab139484,
Abcam, Milpitas, CA). To prepare for the staining, the media was
removed, and the cells were washed once with 1× assay buffer. The dual
color detection solution was added to the wells and incubated for
30 min. The cells were washed and 1× assay buffer added to each well.
The nuclear signal was excitation of 340 nm and emission of 480,
whereas autophagosomes and autophagolysosomes were detected by
excitation of 480 nm and emission of 530 nm. Autophagy activity was
calculated by normalizing the autophagosome/autophagolysosome signal to
the nuclear signal. Rapamycin (2 µm), a known autophagy inducer, was
the positive control.^[ [190]^33 ^] This experiment was performed in
three biological replicates.
Statistical Analysis
Data were presented as the number, percentage, or mean ± SD as
appropriate. One‐way ANOVA with Tukey's post hoc tests for multiple
comparisons was performed. A p‐value <0.05 was considered statistically
significant.
Ethical Approval Statement
The study was conducted in accordance with the Declaration of Helsinki,
and the protocol was approved by the Human Research Ethics Committee,
Faculty of Medicine Ramathibodi Hospital, Mahidol University (Protocol
ID: MURA2020/759; with approval of amendment on May 5, 2021).
Informed Consent Statement
Informed consent was obtained from all subjects involved in the study.
Conflict of Interest
The authors declare no conflict of interest.
Author Contributions
T.L. and P.O.‐y. contributed equally to this work. Conceptualization:
So.C.; Methodology: T.L., P.O.‐y., W.C., So.C.; Validation: W.C.,
So.C.; Formal analysis: T.L., P.O.‐y., Se.C., W.S., W.C., So.C.;
Investigation: T.L., P.O.‐y., Se.C., W.S., W.C., So.C.; Resources:
D.S.N., A.L.M., S.H., So.C.; Software: P.O.‐y., So.C.; Writing—original
draft preparation: T.L., P.O.‐y.; Writing—review and editing: Se.C.,
W.S., D.S.N., A.L.M., S.H., W.C., So.C.; Visualization: T.L., P.O.‐y.,
So.C.; Supervision: D.S.N., A.L.M., S.H., W.C., So.C.; Funding
acquisition: So.C. All authors have read and agreed to the published
version of the manuscript.
Supporting information
Supporting Information
[191]Click here for additional data file.^ (1.4MB, pdf)
Acknowledgements