Abstract

   Preeclampsia is still the leading cause of morbidity and mortality in
   pregnancy without a cure. There are two phenotypes of preeclampsia,
   early-onset (EOPE) and late-onset (LOPE) with poorly defined pathogenic
   differences. This study aimed to facilitate better understanding of the
   mechanisms of pathophysiology of EOPE and LOPE, and identify specific
   biomarkers or therapeutic targets. In this study, we conducted an
   untargeted, label-free quantitative proteomic analyses of plasma
   samples from pregnant women with EOPE (n = 17) and LOPE (n = 11), and
   age, BMI-matched normotensive controls (n = 18). Targeted proteomics
   approach was also employed to validate a subset of proteins (n = 17).
   In total, there were 26 and 20 differentially abundant proteins between
   EOPE or LOPE, and normotensive controls, respectively. A series of
   angiogenic and inflammatory proteins, including insulin-like growth
   factor-binding protein 4 (IGFBP4; EOPE: FDR = 0.0030 and LOPE:
   FDR = 0.00396) and inter-alpha-trypsin inhibitor heavy chain
   H2-4 (ITIH2-4), were significantly altered in abundance in both
   phenotypes. Through validation we confirmed that ITIH2 was perturbed
   only in LOPE (p = 0.005) whereas ITIH3 and ITIH4 were perturbed in both
   phenotypes (p < 0.05). Overall, lipid metabolism/transport proteins
   associated with atherosclerosis were highly abundant in LOPE, however,
   ECM proteins had a more pronounced role in EOPE. The complement cascade
   and binding and uptake of ligands by scavenger receptors, pathways,
   were associated with both EOPE and LOPE.

   Subject terms: Biochemistry, Computational biology and bioinformatics

Introduction

   Hypertensive disorders in pregnancy are the leading cause of morbidity
   and mortality, affecting up to 10% of all^[44]1 pregnancies and include
   chronic hypertension, gestational hypertension, eclampsia and
   pre-eclampsia; all of which increase the risk of complications in both
   mothers and babies during gestation. Preeclampsia, in particular, is
   characterised by the new-onset of gestational hypertension in the
   presence of proteinuria or other organ damage often involving kidneys
   or liver. It affects 5–7% of pregnancies and causes approximately
   76,000 maternal deaths and 500,000 foetal deaths^[45]2.

   The classification of preeclampsia has been evolving over the last
   decade. In 2013, the American College of Obstetricians and
   Gynaecologists (ACOG) and International Society for the Study of
   Hypertension in Pregnancy (ISSHP) incorporated other symptoms/features
   including liver dysfunction, thrombocytopenia, cerebrovascular events
   or foetal growth restriction (FGR), in the absence of proteinuria, to
   diagnose preeclampsia^[46]1–[47]3. Preeclampsia is a multifactorial and
   heterogeneous disorder, stratified depending on the time of onset into:
   (i) early-onset preeclampsia (EOPE) manifested before 34 weeks of
   gestation, and (ii) late-onset preeclampsia (LOPE) manifested from
   34 weeks of gestation. Although EOPE and LOPE share the same clinical
   features, these two phenotypes of preeclampsia lead to different
   outcomes. EOPE is commonly associated with foetal growth restriction
   (FGR) and abnormal uterine artery Doppler, often leading to preterm
   birth and a higher risk of post-pregnancy morbidities^[48]4,[49]5. On
   the other hand, LOPE appears to be a less severe disorder, often
   displaying a normal or slightly increased uterine resistance index and
   a lower rate of FGR^[50]5,[51]6. There is no distinct delineation
   between EOPE and LOPE, with most patients with preeclampsia presenting
   elements of both pathologies, proposing a clinical spectrum.

   The lack of untargeted discovery studies involving multiomics analyses
   impeded understanding of the molecular differences between these two
   phenotypes of preeclampsia. In a systematic review summarising
   quantitative proteomics-based studies using human samples from women
   affected by preeclampsia, the vast majority of studies included samples
   from one of the preeclampsia phenotypes only or grouped both phenotypes
   together, lacking comparison between EOPE and LOPE^[52]7. In this
   study, we conducted an unbiased, label-free quantitative proteomics
   analysis using non-depleted plasma samples collected from patients with
   EOPE (n = 17) and LOPE (n = 11), compared with age- and BMI-matched
   normotensive controls (n = 18). We also validated a subset of 17
   proteins, using targeted proteomics of plasma samples from a similar
   groups of patients. The use of plasma samples should better reflect the
   pathophysiology of EOPE and LOPE as systemic conditions affected by
   widespread endothelial dysfunction^[53]8, and identify potential
   therapeutic targets for future personalised treatment development.

Results

Patient characteristics

   The clinical characteristics of the participants used for untargeted
   and targeted proteomics are presented in Tables [54]1 and [55]2,
   respectively. The distribution of age and BMI were similar across EOPE,
   LOPE and healthy pregnancy groups, whereas gestational age (GA) at
   delivery was significantly lower in EOPE (31.3 ± 2.5 weeks) and LOPE
   (37.0 ± 1.9 weeks), compared to the healthy pregnancy controls
   (39.4 ± 0.9 weeks, p < 0.05; Table [56]1), in line with frequent early
   delivery of the baby in EOPE compared to LOPE or healthy
   controls^[57]5. Maternal blood pressure was higher in EOPE (systolic
   blood pressure (SBP):156.4 ± 26.5 or diastolic BP (DBP):103.1±10.7) and
   LOPE (SBP:144.4 ± 16.8 or DBP:96.9±11.6) compared to healthy
   pregnancies (SBP113.9 ± 8.5 or DBP:72.2±8.1), (p < 0.05). Heart rate
   was increased in women with EOPE (94.6 ± 29.8) compared to healthy
   controls (74.4 ± 6.3, p < 0.05; Table [58]1). Differences were similar
   in the validation groups (except there was no difference in the heart
   rate between EOPE and healthy control, p = 0.05), which included the
   same healthy controls (n = 18) and slightly different EOPE (n = 14) and
   LOPE (n = 14), groups (Table [59]2).

Table 1.

   Summary of patient characteristic (discovery/untargeted proteomics
   set).
   Characteristics EOPE (n = 17) LOPE (n = 11) Healthy pregnancy (n = 18)
   Age (y) 34.0 ± 7.1 33.0 ± 4.0 31.1 ± 3.8
   BMI (kg/m^2) 27.6 ± 4.8 28.3 ± 3.4 24.5 ± 5.5
   GA at delivery (wk) 31.3 ± 2.5 ^¶& 37.0 ± 1.9 ^§& 39.4 ± 0.9 ^¶§
   Number of pregnancies (no.) 1.6 ± 0.9 2 ± 0.9 1.7 ± 0.8
   Systolic blood pressure (mm Hg) 156.4 ± 26.5 ^¶ 144.4 ± 16.8 ^§
   113.9 ± 8.5 ^¶§
   Diastolic blood pressure (mm Hg) 103.1 ± 10.7 ^¶ 96.9 ± 11.6 ^§
   72.2 ± 8.1 ^¶§
   Heart rate (bpm) 94.6 ± 29.8 ^¶ 82.8 ± 16.5 74.4 ± 6.3 ^¶
   Medications
   Methyldopa (no. [%]) 17 [100%] 11 [100%] 0 [0%]
   Amlodipine (no. [%]) 5 [29.4%] 2 [18.2%] 0 [0%]
   Dexamethasone (no. [%]) 11 [64.7%] 1 [9.1%] 0 [0%]
   Nadroparin calcium (no. [%]) 3 [17.6%] 1 [9.1%] 0 [0%]
   Diazepam (no. [%]) 3 [17.6%] 8 [72.7%] 0 [0%]
   Magnesium sulphate (no. [%]) 4 [23.5%] 1 [9.1%] 0 [0%]
   Nifedipine (no. [%]) 0 [0%] 1 [9.1%] 0 [0%]
   Low-sodium diet (no. [%]) 11 [64.7%] 8 [72.7%] 0 [0%]
   [60]Open in a new tab

   BMI, body mass index; bpm, beats per minute; EOPE, early-onset
   pre-eclampsia; GA, gestational age; LOPE, late-onset pre-eclampsia.

   ^¶, P < 0.05 of a characteristic between EOPE and healthy pregnancy
   groups.

   ^§, P < 0.05 of a characteristic between LOPE and healthy pregnancy
   groups.

   ^&, P < 0.05 of a characteristic between EOPE and LOPE.

Table 2.

   Summary of patients’ characteristics (validation/targeted proteomics
   set).
   Characteristics EOPE (n = 14) LOPE (n = 14) Healthy pregnancy (n = 18)
   Age (y) 33.2 ± 7.6 32.6 ± 3.7 31.1 ± 3.8
   BMI (kg/m^2) 28.0 ± 5.0 27.5 ± 3.7 24.5 ± 5.5
   GA at delivery (wk) 31.9 ± 2.1^¶& 36.8 ± 1.9 ^§& 39.4 ± 0.9^¶§
   Number of pregnancies (no.) 1.8 ± 0.9 2 ± 0.9 1.7 ± 0.7
   Systolic blood pressure (mm Hg) 157 ± 28.9^¶ 142 ± 15.7^§
   113.9 ± 8.5^¶§
   Diastolic blood pressure (mm Hg) 103 ± 10.9^¶ 93.5 ± 12.5^§
   72.2 ± 8.1^¶§
   Heart rate (bpm) 86.1 ± 17.6 84.6 ± 15.7 74.4 ± 6.3
   Medications
   Methyldopa (no. [%]) 14 [100%] 13 [92%] 0 [0%]
   Amlodipine (no. [%]) 5 [35.7%] 2 [14.3%] 0 [0%]
   Dexamethasone (no. [%]) 9 [64.3%] 1 [7.1%] 0 [0%]
   Nadroparin calcium (no. [%]) 3 [21.4%] 2 [14.3%] 0 [0%]
   Diazepam (no. [%]) 3 [21.4%] 10 [71.4%] 0 [0%]
   Magnesium sulphate (no. [%]) 3 [21.4%] 1 [9.1%] 0 [0%]
   Nifedipine (no. [%]) 1 [7.1%] 1 [7.1%] 0 [0%]
   Low-sodium diet (no. [%]) 9 [64.%] 8 [57.1%] 0 [0%]
   [61]Open in a new tab

   BMI, body mass index; bpm, beats per minute; EOPE, early-onset
   pre-eclampsia; GA, gestational age; LOPE, late-onset pre-eclampsia.

   ^¶, P < 0.05 of a characteristic between EOPE and healthy pregnancy
   groups.

   ^§, P < 0.05 of a characteristic between LOPE and healthy pregnancy
   groups.

   ^&, P < 0.05 of a characteristic between EOPE and LOPE.

Differentially abundant plasma proteins can differentiate between the
different phenotypes of preeclampsia and healthy controls

   Label-free proteomic analysis of non-depleted plasma samples was
   conducted by measuring the relative abundance of tryptic peptides using
   data-dependent acquisition (DDA) mass spectrometry. The generated
   outputs contained 370 proteins detected across all samples with minimal
   percentage (< 15%) of missing values (Supplementary data [62]1).

   Initially, the clustering of groups was assessed through examination of
   the principal component analysis (PCA) plot (Fig. [63]1a) and heatmap
   showing the unsupervised hierarchical clustering of the individual
   samples (Fig. [64]1b). Both PCA and heatmap revealed that the proteomes
   are heterogeneous across all the samples. Following PCA, the
   differential expression (DE) analysis was performed by three separate
   comparisons; (i) EOPE versus healthy pregnancy, (ii) LOPE versus
   healthy pregnancy and (iii) EOPE versus LOPE, while adjusting for GA at
   delivery. DE proteins were defined as FDR < 0.05. In total, there were
   26 and 20 differentially abundant proteins in EOPE versus healthy
   pregnancy and LOPE versus healthy pregnancy, respectively, and one
   protein was differentially abundant between EOPE and LOPE (Table [65]3;
   Figs. [66]1c–e and [67]2a; Supplementary data [68]2).

Figure 1.

   [69]Figure 1
   [70]Open in a new tab

   Overviews of differential expression analysis. (a) Principal component
   analysis (PCA) plot of proteomic data in EOPE, LOPE and healthy
   pregnancy groups. (b) Multigroup heatmap with hierarchical clustering
   dendrogram of proteomic data levels across EOPE, LOPE and healthy
   pregnancy groups. Volcano plots of proteomic data in (c) EOPE versus
   healthy pregnancy, (d) LOPE versus healthy pregnancy, and e EOPE versus
   LOPE. The differential expression (DE) analysis was performed by
   fitting a linear regression model adjusted for GA at delivery. DE
   proteins were defined as Benjamini–Hochberg adjusted P value < 0.05, as
   indicated by the proteins above the cut-off lines.

   Source data are provided as a Source Data file.

Table 3.

   DE plasma proteins of preeclampsia.
   Protein logFC FDR Validation results logFC (p value)
   EOPE versus healthy pregnancy
   IGLV3-21  − 0.9940833 0.002

    − 1.10703

   (p = 0.02)
   FN1 1.07026906 0.003

   0.749238

   (p = 0.007)
   CFD 1.33165459 0.003 n/a
   IGFBP4 4.2479972 0.003 n/a
   PAPPA2 10.6575471 0.005 n/a
   CPN2 0.44690077 0.007

   0.182445

   (p = 0.008)
   HRG 0.81057059 0.008

   0.35572

   (p = 0.007)
   C1R 0.33304499 0.008 n/s
   ITIH3 0.67823875 0.012

   0.429546

   (p = 0.02)
   CST3 0.73940906 0.018 n/a
   HGFA 2.72402884 0.018 n/s
   HPX/HEMO  − 0.2855547 0.023

    − 0.18325

   (p = 0.007)
   H2BC14 9.86191588 0.023 n/a
   H2BC12L 9.86191588 0.023 n/a
   TF  − 0.3914363 0.023  − 0.3227 (p = 0.001)
   SVEP1  − 2.648795 0.023 n/a
   A2M 0.34482872 0.028 n/a
   SERPINA5 1.41940662 0.031 n/a
   B2M 0.70236879 0.031 n/a
   AGT  − 0.3379023 0.04 n/a
   AMBP 0.46608993 0.04 n/a
   TMSB4X  − 10.989325 0.04 n/a
   ENPP2 2.17783194 0.046 n/a
   APOE 0.44371805 0.046 n/s
   APOC4-APOC2 0.56440844 0.046 n/s
   IGKV1-9  − 0.8285979 0.048 n/a
   IGHV1OR15-1 n/s

    − 0.38218

   (p = 0.004)
   ITIH4 n/s

   0.294583

   (p = 0.02)
   LOPE versus healthy pregnancy
   IGLV3-21  − 1.0237716 0.004

    − 2.207995088

   (p < 0.0001)
   IGHV1OR15-1  − 4.0259103 0.004

    − 0.24114122

   (p = 0.02)
   CFD 1.49090423 0.004 n/a
   FN1 1.13370152 0.004 n/s
   HPX/HEMO  − 0.4013146 0.004

    − 0.134143098

   (p = 0.02)
   IGFBP4 4.91273421 0.004 n/a
   APOC4-APOC2 0.89848759 0.004

   0.285337364

   (p = 0.02)
   APOE 0.70579726 0.004

   0.484743909

   (p = 0.003)
   HRG 0.95744956 0.004

    − 0.315575546

   (p = 0.005)
   CPN2 0.47425523 0.011

   0.317003488

   (p < 0.0001)
   C1R 0.34745024 0.017

   0.475339172

   (p = 0.002)
   AOC1  − 2.6911841 0.021 n/a
   VWF 0.9703592 0.024 n/s
   PDHX 1.66742853 0.024 n/a
   APOC3 0.66060763 0.03

   0.937780283

   (p < 0.0001)
   MASP2 0.82804763 0.03 n/s
   AMBP 0.52819589 0.0498 n/a
   IGHV6-1  − 1.1543323 0.0498 n/a
   TF  − 0.385595866 (p < 0.0001)
   amiD 0.77746779 0.03 n/a
   ITIH2 0.37228745 0.03

   0.290554799

   (p = 0.005)
   ITIH3 n/s

   0.434991009

   (p = 0.008)
   ITIH4 n/s

   0.349781944

   (p = 0.005)
   EOPE versus LOPE
   IGHV1OR15-1 3.83674304 0.005 n/s
   APOC3 n/s  − 0.93954084 (p < 0.0001)
   HRG n/s 0.671295765 (p < 0.0001)
   [71]Open in a new tab

   n/a—not available.

   n/s—non-significant.

   AGT, angiotensinogen; AMBP, protein AMBP; amiD,
   N-acetylmuramoyl-L-alanine amidase; AOC1, amiloride-sensitive amine
   oxidase [copper-containing]; APOC3, apolipoprotein C-III; APOC4-APOC2,
   APOC4-APOC2 readthrough (NMD candidate); APOE, apolipoprotein E; A2M,
   alpha-2-macroglobulin; B2M, beta-2-microglobulin; CFD, complement
   factor D; CPN2, carboxypeptidase N subunit 2; CST3, cystatin-C; C1R,
   complement C1r subcomponent; ENPP2, ectonucleotide
   pyrophosphatase/phosphodiesterase family member 2; FN1, fibronectin;
   HGFA, hepatocyte growth factor activator; HPX, hemopexin; HRG,
   histidine-rich glycoprotein; H2BC12L, Histone H2B type F-S; H2BC14,
   histone H2B type 1-M; IGFBP4, insulin-like growth factor-binding
   protein 4; IGHV1OR15-1, immunoglobulin heavy variable 1/OR15-1
   (non-functional) (Fragment); IGHV6-1, immunoglobulin heavy variable
   6–1; IGLV3-21, immunoglobulin lambda variable 3–21; IGKV1-9,
   immunoglobulin kappa variable 1–9; ITIH2, inter-alpha-trypsin inhibitor
   heavy chain H2; ITIH3, inter-alpha-trypsin inhibitor heavy chain H3;
   ITIH4, inter-alpha-trypsin inhibitor heavy chain H4; MASP2,
   mannan-binding lectin serine protease 2; PAPPA2, pappalysin-2; PDHX,
   pyruvate dehydrogenase protein X component mitochondrial; SERPINA5,
   plasma serine protease inhibitor; SVEP1, sushi von Willebrand factor
   type A EGF and pentraxin domain-containing protein 1; TF,
   serotransferrin; TMSB4X, thymosin beta-4; VWF, von Willebrand factor.

Figure 2.

   [72]Figure 2
   [73]Open in a new tab

   Highlighted proteins in differential expression analysis. (a) Triple
   Venn diagram summarising the differentially expressed proteins unique
   and overlapped between EOPE versus healthy pregnancy, LOPE versus
   healthy pregnancy and EOPE versus LOPE. (b) Three-dimensional plot of
   fold changes (log[2]-transformed) of all identified proteins, with
   regression plane filled with colour indicator of log[2]-fold changes.

   Source data are provided as a Source Data file.

   Overall, the differences in proteome profiles between EOPE or LOPE and
   healthy pregnancy were largely associated with impaired angiogenesis
   and included perturbed abundance of inter-alpha-trypsin inhibitor heavy
   chain H2-4 (ITIH3 in EOPE and ITIH2/4 in LOPE)^[74]8, insulin-like
   growth factor-binding protein 4 (IGFBP4)^[75]9–[76]12 and
   histidine-rich glycoprotein (HRG)^[77]13–[78]15 (Fig. [79]2a).
   Immunoglobulin heavy variable 1/OR15-1 (IGHV1OR15-1) was the only
   protein differentially abundant between the two subtypes of
   preeclampsia, being substantially higher in abundance in EOPE compared
   to LOPE (FC = 14.29, FDR = 0.005) (Fig. [80]2b). Through our targeted
   proteomics we validated the changes observed with ITIH3 in EOPE
   (logFC = 0.429, p = 0.02) and ITIH 2&4 in LOPE (logFC = 0.290,
   p = 0.005 and logFC = 0.3498, p = 0.005, respectively). ITIH4 was also
   significantly upregulated in EOPE (logFC = 0.294, p = 0.02) therefore
   making ITIH2 uniquely perturbed in LOPE (Table [81]3). Interestingly,
   between EOPE and LOPE, IGHV1OR15-1 was not significantly different in
   abundance using the targeted proteomics but rather APOC3
   (logFC = − 0.939, p < 0.0001) and HRG (logFC = 0.671, p < 0.0001). HRG
   is a well-studied biomarker of preeclampsia related to platelet
   haemostasis ^[82]16–[83]18 that was increased ~ twofold in both
   phenotypes of preeclampsia (EOPE: FDR = 0.008; LOPE: FDR = 0.004) when
   we applied untargeted proteomics. However, following validation, HRG
   showed similar trend in EOPE (p = 0.007) versus healthy controls
   whereas it was downregulated in LOPE (p = 0.005) versus healthy
   controls and as such it was differentially abundant between the two
   phenotypes.

   Furthermore, there were 11 DE proteins commonly shared between EOPE and
   LOPE. Immunoglobulin lambda variable 3–21 (IGLV3-21) was the top
   differentially abundant protein shared between EOPE (FDR = 0.002) and
   LOPE (FDR = 0.004), with an approximately 50% decrease in its
   abundance in either of phenotypes compared to healthy pregnancy
   controls. This was also confirmed as part of validation using targeted
   proteomics. Well-characterised preeclampsia biomarkers including
   fibronectin 1 (FN1)^[84]19,[85]20 and complement factor D
   (CFD)^[86]21,[87]22 were ~ twofold increased in abundance and among the
   most significant abundant proteins in both EOPE and LOPE, compared to
   control. Whilst CFD did not undergo validation, using targeted
   proteomics FN1 was only significant in EOPE (p = 0.007) and not in
   LOPE, versus healthy controls.

   The proteins displaying significant changes of abundance were
   highlighted in a three-dimensional plot (Fig. [88]2b; Supplementary
   data [89]3). Among the proteins with significant fold changes in EOPE
   compared to healthy controls, a series of proteins have previously been
   reported as biomarkers of preeclampsia, including serpin family A
   member 5 (SERPINA5)^[90]23, pappalysin 2 (PAPPA2)^[91]24,[92]25,
   hepatocyte growth factor activator (HGFAC)^[93]26,[94]27 and thymosin
   beta-4 (TMSB4X)^[95]28. Whilst SERPINA5, PAPPA2, and TMSB4X did not
   undergo validation using targeted proteomics, HGFAC was not
   statistically significant as part of the validation (Table [96]3).
   IGFBP4, a protein significantly increased in abundance in EOPE
   (log[2]FC = 4.25, FDR = 0.0003) and LOPE (log[2]FC = 4.91,
   FDR = 0.004), compared to controls, is known to have anti-angiogenic
   properties^[97]9–[98]12, adding further evidence to the central role of
   impaired angiogenesis in preeclampsia.

Pathogenic pathway associated with different phenotypes of preeclampsia

   Following identification of differentially abundant biomarkers for
   different phenotypes of preeclampsia, pathway enrichment analysis was
   performed for EOPE or LOPE group proteins, compared to healthy
   controls. Pathways altered in EOPE and LOPE (FDR < 0.05) were presented
   using a triple Venn diagram (Fig. [99]3a; Supplementary data [100]4).
   The pathway Venn diagram revealed a range of pathways associated with
   altered haemostasis and immune system.

Figure 3.

   [101]Figure 3
   [102]Open in a new tab

   Pathway analysis. (a) Triple Venn diagram illustrating the number of
   pathways significantly changed in EOPE versus healthy pregnancy and
   LOPE versus healthy pregnancy. (b) Illustrative example of DE protein
   perturbations in the regulation of IGF transport and uptake through
   IGFBP. IGF insulin-like growth factor, IGFBP insulin-like growth factor
   binding protein.

   Source data are provided as a Source Data file.

   A total of 13 pathways were significantly enriched, with 6 pathways
   shared between EOPE and LOPE groups (Fig. [103]3a), compared to healthy
   controls. Differentially abundant proteins are part of the biological
   pathway regulating insulin-like growth factor (IGF) transport and
   uptake through IGF binding protein (IGFBP; Fig. [104]3b), which plays a
   significant role in both EOPE (FDR = 0.01) and LOPE (FDR = 0.002).
   Given the pro-angiogenic function of IGF signalling
   pathway^[105]29,[106]30, our findings emphasised the importance of
   impaired angiogenesis in the pathophysiology of preeclampsia. In
   addition, differentially abundant proteins were part of a series of
   haemostatic pathways, particularly platelet degranulation in response
   to elevated intra-platelet Ca^2+ pathway, which were more pronounced in
   EOPE (FDR = 0.00001; Supplementary data [107]4). Proteins shown as
   differentially abundant in EOPE compared to healthy controls were more
   closely associated with acute inflammatory pathways than those
   identified in LOPE.

Signalling networks in EOPE and LOPE

   Pairwise correlation network analysis was next performed to investigate
   protein–protein interactions (PPIs) in EOPE and LOPE. Networks were
   highlighted with the most correlated nodes (Pearson correlation r > 0.7
   or < − 0.7), where the colour and the length of the edge are
   proportional to the Pearson correlation coefficient r (Fig. [108]4a,b).
   To compare the PPIs in our data with broad reference evidence, the edge
   width was coded proportionally to the PPI-confidence scores derived
   from the STRING database^[109]31. Therefore, if consistent with the
   references, a pair of nodes are presented with a thick and opaque edge