LIGHT/TNFSF14 Levels in Carotid Plaques Are Associated With Symptomatic Cerebrovascular Disease
Article information
Abstract
Background and Purpose
Plaque rupture is the underlying cause of most cardiovascular events, such as stroke and myocardial infarction. The co-stimulatory molecule LIGHT (tumor necrosis factor superfamily member 14, TNFSF14) has been detected in foam cell-rich regions of atherosclerotic plaques, but whether it has a role in plaque stability is not known. This study investigates the association between intraplaque LIGHT levels and plaque vulnerability.
Methods
LIGHT levels were measured in homogenates of carotid endarterectomy samples by proximity extension assay (n=202) and through bulk RNA sequencing and spatial transcriptomics (Visium) of plaques from patients included in the Carotid Plaque Imaging Project. Homogenates were further examined by multiplex analyses and enzyme-linked immunosorbent assay, and plaque sections by immunohistochemistry.
Results
Plaque levels of LIGHT were associated with occurrence of preoperative cerebrovascular symptoms, including stroke. LIGHT levels correlated with a histological plaque vulnerability index, necrotic core size, and inflammatory cytokine levels. Additionally, expression of extracellular matrix turnover machinery components, including the collagen cross-linking proteoglycan fibromodulin and matrix metalloproteinases 1, 2, 9, and 10, was associated with plaque LIGHT levels.
Conclusion
Expression of LIGHT in atherosclerotic plaques not only correlates with markers of plaque destabilization, but is also significantly elevated in plaques from symptomatic compared to those from asymptomatic patients. These results associate LIGHT content with a rupture-prone plaque phenotype, potentially upregulated as part of a reparative response, warranting further studies.
Introduction
Thrombus formation following atherosclerotic plaque rupture, resulting in obstruction of the arterial lumen, is the most common underlying cause of cardiovascular events, such as stroke and myocardial infarction [1]. Vulnerable or rupture-prone plaques are often rich in lipids and inflammatory cells, and covered by a thin fibrous cap.
Co-stimulatory molecules, part of the immune checkpoint family, have been revealed as essential drivers of atherosclerotic plaque progression and are thus emerging as strong drug target candidates for stabilization of vulnerable plaques [2-4]. Tumor necrosis factor superfamily member 14 (TNFSF14), also known as LIGHT and CD258, is expressed mainly by lymphocytes and antigen-presenting cells [5,6]. LIGHT has been implicated as a driver of disease progression in several inflammatory diseases, often featuring fibrotic and/or autoimmune elements, including rheumatoid arthritis [7], asthma [8], atopic dermatitis [9], and systemic sclerosis [10]. In atherosclerotic mouse models, LIGHT deficiency has been reported to both aggravate and ameliorate atherosclerosis through differing effects on inflammation and immune cell homeostasis [11,12]. Little is known about the potential involvement of LIGHT signaling in human cardiovascular disease (CVD), and further insight is vital for future evaluation of LIGHT as a potential therapeutic target.
While LIGHT is reportedly expressed in foam cell-rich regions of human atherosclerotic plaques [13,14], it is not known whether LIGHT has a role in promoting a high-risk plaque phenotype. To address whether LIGHT levels are associated with features of atherosclerotic plaque vulnerability, we have analyzed LIGHT expression in relation to plaque stability markers in human carotid endarterectomy plaques.
Methods
Detailed methods are provided in Supplementary Methods.
Study cohort
The study was approved by the Swedish Ethical Review Authority (472/2005; 2014/904; 60/2008, 2023-05910-01) and was carried out in accordance with the principles of the 1975 Declaration of Helsinki. All included study subjects gave written informed consent. Sharing of the study datasets containing pseudonymized participant data is subject to limitations specific to the ethical permit and the General Data Protection Regulation (GDPR).
Carotid plaques (n=202) and plasma samples (n=558) of the Carotid Plaque Imaging Project (CPIP) cohort were obtained from patients undergoing endarterectomies at the Vascular Department of Skåne University Hospital (Malmö, Sweden) between 2005 and 2010. Clinical characteristics of the included subjects are described in Supplementary Tables 1 and 2. Blood samples were collected on the day before surgery, and excised plaques were immediately snap-frozen in liquid nitrogen. Plaque homogenates were prepared as previously described [15].
Indications for surgery were as previously described [16]. Briefly, cerebrovascular symptoms (i.e., amaurosis fugax, transient ischemic attack, or stroke) were evaluated by a neurologist, and the degree of carotid artery stenosis was assessed by duplex-ultrasound based on flow velocities as previously described [17]. Patients exhibiting stenosis >70% associated with ipsilateral symptoms were categorized as symptomatic, and patients exhibiting carotid stenosis >80% without ipsilateral symptoms (during the 6 months leading up to endarterectomy) were characterized as asymptomatic.
Plaque homogenate and plasma analyses
Levels of soluble LIGHT were analyzed in plaque homogenates and plasma through the proximity extension assay (PEA) technique using the Proseek Multiplex CVD96x96 reagents kit (Olink Bioscience, Uppsala, Sweden) as described previously [18] with selected studies overviewed in Supplementary Table 3.
Histology
A segment (2-mm thick) was cut from the most stenotic region of each plaque (before homogenization), embedded in optimal cutting medium (Sakura Finetek Europe BV, Japan) and sectioned (8 μm) using a cryostat for use in histological analyses (Oil Red O, Russell-Movat pentachrome, and Von Kossa stains, as well as alpha smooth muscle actin [α-SMA], a disintegrin and metalloproteinase with thrombospondin type 1 repeats (ADAMTS7), cartilage oligomeric protein [COMP], CD163, CD68, fibromodulin, glycophorin A, lumican, and cleaved type I and II collagen [3/4 fragment] immunohistochemical stains).
Necrotic plaque regions are defined as acellular and non-fibrotic plaque areas, visualized by Russell-Movat’s pentachrome stain. Stained slides were scanned and digitized using an Aperio ScanScope digital slide scanner (Aperio Technologies Inc., Vista, CA, USA). Quantifications were performed on blinded samples, using BioPix iQ (version 2.3.1) imaging software (Biopix AB, Gothenburg, Sweden). Vulnerability index was calculated as described previously [19]:
RNA sequencing
One-millimetre cross-sectional segments taken from the most stenotic part of each plaque were used for RNA isolation (n=78), which was performed as described previously [20]. Briefly, preparation of RNA cleared of ribosomal RNA was performed through Trizol extraction (Ribo-Zero™ Magnetic Kit [Epicentre, Vilnus, Lithuania]). RNA sequencing was performed using the Illumina HiSeq2000 and NextSeq 500/550 platforms, following established procedures [21,22]. Transcript-level quantification was done using Salmon [23], utilizing GENCODE’s transcriptome release for mapping [24]. Gene counts were then summarized using tximport [25] and were subsequently standardized across all samples using the trimmed mean of M-values (TMM) method via edgeR [26]. This normalization resulted in gene expressions represented as log2-transformed counts per million (CPM) after applying voom transformation. Moreover, any potential differences introduced by the sequencing platforms were tackled through applying an empirical Bayes method [27] to minimize potential batch effects. Selected studies are overviewed in Supplementary Table 3.
Spatial transcriptomics
To confirm the spatial distribution of TNFSF14 within plaque tissue sections, we utilized the spatial gene expression dataset of human carotid plaque tissues described previously by our group [20,28]. Briefly, linked .cloupe files corresponding to three human carotid plaque specimens were imported into Loupe Browser v8.0 (10X Genomics, Pleasanton, CA, USA). These files were subsequently used to explore and visualize TNFSF14-positive spots overlaid on the plaque tissue sections, enabling their detailed spatial analysis.
Statistics
The variables of interest were shown as median with interquartile range (IQR). For analysis of plasma, plaque histology sections, and homogenates, the Mann-Whitney U test was used to compare groups and Spearman’s rank correlation was used for continuous variables. The chi-square test was used for comparisons between categorical variables. A P-value of <0.05 was considered statistically significant. Adjustments for multiple comparisons were made when appropriate according to the Benjamini-Hochberg procedure (false-discovery rate) and the Holm-Šídák test. Statistical analysis was performed using SPSS 28.0.1.1 (IBM Corp., Armonk, NY, USA) and GraphPad Prism (version 9.0.0 for Windows, GraphPad Software, San Diego, CA, USA; www.graphpad.com). All these causal analyses were implemented in R software, version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria).
Results
Carotid plaque LIGHT levels are associated with cerebrovascular symptoms
LIGHT levels in carotid endarterectomy plaques from patients of the CPIP cohort were higher in plaques from patients with preoperative cerebrovascular symptoms compared to asymptomatic patients (P<0.001) (Figure 1A and Supplementary Table 4). The association remained significant when adjusted for age and sex (odds ratio [OR] 1.088, 95% confidence interval [CI] 1.015–1.165, P=0.017). Significant associations between LIGHT levels and other clinical parameters were not observed (Supplementary Tables 4 and 5).
LIGHT expression in human carotid endarterectomy plaques. LIGHT expression in human carotid endarterectomy plaques. (A) Protein levels measured by proximity extension assay (n=202; 95 asymptomatic and 107 symptomatic patients). (B) mRNA expression (n=78; 27 asymptomatic and 51 symptomatic patients). In both analyses, LIGHT levels were significantly higher in symptomatic than in asymptomatic plaques. Mann–Whitney U test results are visualized with violin plots, with dotted lines denoting the median and quartiles. log2CPM, log2-transformed counts per million.
Accordingly, gene expression of TNFSF14 analyzed by RNA sequencing was also higher in plaques from patients with preoperative cerebrovascular symptoms than in those without symptoms (log2CPM, 3.55 [IQR 2.96–4.26] vs. 2.95 [IQR 2.15–3.56]) (Figure 1B).
In the open-access single-cell RNA-sequencing dataset created by Slenders et al. [29] (retrieved through PlaqView [30]), the most pronounced gene expression of TNFSF14 in human carotid plaques was found in T-cells, natural killer (NK)-cells, and pericytes; while moderate expression was seen among erythrocytes, macrophages, and epithelial cells; and only weak expression among endothelial cells, smooth muscle cells (SMCs), and B-cells. In plaques of the CPIP cohort, spatial transcriptomics showed TNFSF14 expression to be present both in the non-necrotic parts of the plaque core, as well as in the fibrous cap (Figure 2).
Representative image showing localization of TNFSF14 gene expression in the core region of a carotid plaque. Spatial transcriptomics dots indicate loci of TNFSF14 expression, with the color scale representing lognormalized expression levels (stained with hematoxylin and eosin. Scale bar=2.5 mm).
LIGHT levels are associated with a pro-inflammatory plaque environment
To explore whether the association between LIGHT levels and preoperative cerebrovascular symptoms was reflected by a plaque milieu characteristic of a vulnerable phenotype, we assessed how LIGHT levels associated with pro-inflammatory plaque markers. Plaque LIGHT levels correlated with the composite histological plaque vulnerability index [19] (rho=0.552, P=3.0×10-14) (Table 1), a ratio between the destabilizing components—macrophages (marker CD68), intraplaque hemorrhage (glycophorin A), and lipids (Oil Red O)—and stabilizing components—SMCs (α-SMA) and collagen (Movat’s pentachrome). LIGHT levels also correlated with each of the individual destabilizing parameters (in the case of lipids, also confirmed by enzyme-linked immunosorbent assay quantification of oxidized low-density lipoprotein [oxLDL] in plaque homogenate) and showed a negative correlation with the stabilizing parameter α-SMA, but not with collagen (Table 1). Additionally, LIGHT levels were negatively correlated with the presence of calcium (Table 1).
Correlation of plaque LIGHT levels with plaque vulnerability features and cytokine content
Furthermore, plaque LIGHT expression showed a positive correlation with additional histological parameters associated with a pro-inflammatory plaque environment, namely necrotic core size and the scavenger receptor CD163. LIGHT levels also correlated with plaque content of active Caspase-3, as well as with interleukin (IL)-1β, IL-6, chemokine (C-C motif) ligand (CCL)-2 (MCP-1), CCL4 (MIP-1β), CCL5 (RANTES), and platelet-derived growth factor (PDGF). LIGHT levels correlated negatively with CCL11 (eotaxin) (Table 1).
Finally, plaque LIGHT levels correlated with several components belonging to the extracellular matrix (ECM) turnover machinery, such as histologically detected cleaved collagen, the collagen cross-linking small leucine-rich proteoglycans fibromodulin and lumican, the matricellular protein COMP, and the remodeling regulator ADAMTS7 (Table 2). In line with these observations, LIGHT levels correlated with matrix metalloproteinases (MMPs) 1, 2, 9, and 10, and tissue inhibitor of metalloproteinase (TIMP)-1, though inversely with transforming growth factor (TGF)-β2, measured in plaque homogenate (Tables 2 and 3).
Correlation of plaque LIGHT levels with features related to plaque matrix turnover
Spearman correlations between LIGHT (A.U.) measured in plaque homogenate and TGF-β
Though associated with preoperative cerebrovascular symptoms, plaque levels of LIGHT were not associated with future cardiovascular events or cardiovascular mortality (after median follow-up periods of 38 [IQR 15–62] and 44 [IQR 21–66] months, respectively).
Circulating LIGHT and CVD
To further explore causal associations for LIGHT on CVD, Mendelian randomizations were conducted using large genome-wide association studies (GWAS) of circulating LIGHT, AS, AIS, LAS, CAD, and MI (Supplementary Table 6). We found that circulating LIGHT was causally associated with an increased risk of LAS (OR=1.379, 95% CI=1.018–1.868, P=0.038) using GWAS data of circulating LIGHT from the deCODE project. However, this association was not consistently observed when using GWAS data from other consortia. No causal effects of circulating LIGHT were identified for other cardiovascular outcomes (Supplementary Figure 1 and Supplementary Table 7). In line with these results, circulating LIGHT levels in the CPIP cohort were neither associated with preoperative symptoms (P=0.230), nor with future cardiovascular events or survival after adjusting for possible confounders (Supplementary Tables 8 and 9).
Discussion
Here, we show that LIGHT levels are higher in carotid endarterectomy plaques from symptomatic, compared to asymptomatic patients, and associated with markers of an inflammatory plaque environment with active ECM turnover (Figure 3).
LIGHT/TNFSF14 levels in human carotid plaques are associated with symptomatic cerebrovascular disease. Expression of LIGHT is elevated in plaques from symptomatic, compared to asymptomatic, patients and correlates with markers of plaque destabilization. Created in part with BioRender (Shami A, 2025; https://BioRender.com/p35q802).
Much of the current interest in the potential therapeutic value of interfering with co-stimulatory signaling pathways to stabilize and/or combat progression of atherosclerosis has been generated through reports by us and others identifying co-stimulatory molecules as having essential roles in atherosclerotic CVD [31]. While in experimental models, co-stimulatory molecule triggering has most often been described as atherogenic, in the case of LIGHT, reports from murine atherosclerosis models have yielded contrasting results. Though inactivation of the LIGHT/lymphotoxin beta receptor (LTβR) axis ameliorates inflammation and macrophage proliferation associated with atherosclerosis burden in the presence of insulin resistance and metabolic syndrome [12], LIGHT deficiency has also been found to aggravate atherosclerosis by promoting plaque apoptosis as well as Th17 (over Treg) cell populations [11]. Contrastingly, overexpression of LIGHT by T-cells was found to exacerbate hypercholesterolemia [32], which also suggests aggravated atherogenesis, though it should be noted that the latter two studies were carried out in different murine models (apolipoprotein E and low-density lipoprotein receptor deficiency, respectively). Considering these data, exploring LIGHT expression in relation to plaque vulnerability parameters in human plaques is an important next step in investigating the potential role of this molecule as a key player in atherosclerosis.
Though increased levels of circulating LIGHT have been reported after stroke [33], as well as in patients diagnosed with stable and unstable angina [34,35], in our endarterectomy cohort, we did not find any association between circulating LIGHT levels and either preoperative cerebrovascular symptoms or post-operative cardiovascular events or mortality. Surprisingly, there was instead a tendency towards negative association with cardiovascular mortality; however, it did not hold up to adjusting for possible confounders. Moreover, though associated with a rupture-prone plaque phenotype at the time of endarterectomy, plaque levels of LIGHT did not show potential as a marker for future cardiovascular survival. These results were further corroborated by Mendelian randomization using three large cohorts (SCALLOP [Systematic and Combined Analysis of Olink Proteins] [36], deCODE [37], and UK Biobank Pharma Proteomics Project [UKB-PPP] [38]), which did not demonstrate a strong causal relationship between plasma LIGHT levels and cardiovascular outcomes. Though a weak causal effect on LAS was identified using the deCODE cohort, it could not be validated in the other two examined datasets, suggesting limited generalizability. These results also raise the possibility that LIGHT-upregulation in vulnerable plaques may be induced as part of a process not actively driving plaque destabilization but rather serve a reparative function. Moreover, the role of LIGHT in plaques may be linked to the activity of other biological factors, such as MMP-9 degradation of the cap or low plaque content of reparative elements such as TGF-β2, which is consistent with our observed association of plaque LIGHT content with higher MMP-9 and lower TGF-β2 levels.
Suggesting a protective role for LIGHT in atherosclerotic plaque progression by coordinating local T-cell immunity, Hurtado-Genovés et al. [11] found lower TNFSF14 gene expression in atherosclerotic aortic wall biopsies from CAD patients to be associated with altered levels of T-cell differentiation markers and cell death, while anti-apoptotic and cytotoxic pathways were induced by LIGHT-ligation in vitro [11]. Beyond these studies, the role of LIGHT in plaque vulnerability has only been investigated in human atherosclerosis by immunohistochemical detection in macrophage-rich plaque regions [13,14] and through an observed tendency for greater LIGHT presence in inflammatory, compared to fibrous, plaque regions (though notably measured by Western blot in only two samples) [13].
Our finding of associations between the levels of LIGHT and inflammatory mediators and lipids in human carotid plaques may further suggest that effects of LIGHT on plaque vulnerability features differ with lymphocyte or myeloid cell stimulation. Also in line with this notion are the particularly strong correlations between plaque LIGHT content and pro-inflammatory cytokines expressed by several different cell types, such as the IL-1β/IL-6 axis and CCL2 (by myeloid and non-immune cells) and CCL5 (mainly by T-cells) [39,40]. These results align with previous reports of myeloid cell LIGHT signaling promoting pro-inflammatory and atherogenic features. These studies included observed LIGHT-upregulation in blood Ly6Chigh macrophages (from ApoE-/- mice) and following stimulation with oxLDL [35], and exogenous LIGHT in oxLDL-induced THP-1-derived macrophages [34]. In addition, contrary to reports on T-cells, rather than prevent cell death and cytotoxicity, LIGHT stimulation of macrophages seems to mediate vascular endothelial growth factor (VEGF)-induced apoptosis [41]. Collectively, studies so far thus indicate LIGHT triggering in plaques to have both atheroprotective and atherogenic functions depending on cellular context and future studies will be required to draw further conclusions on the net effects of LIGHT signaling in human atherosclerotic CVD.
The occurrence of plaque rupture is dictated by differing plaque ECM quantity, structure, and components, a balance that may be shifted by the presence of co-stimulatory signaling [1,2,42]. LIGHT has previously been found to promote SMC proliferation and drive a fibrous response in keratinocytes [43,44]. Though driving plaque fibrosis may be hypothesized as another potential atheroprotective role for LIGHT, our study found LIGHT to correlate mainly with ECM synthesis and degradation markers, as well as with (mesenchymal cell mitogen) PDGF, rather than with structural ECM components, such as collagen. This suggests LIGHT signaling to be associated with active tissue repair secondary to ECM degradation, as promoted by a highly inflamed plaque microenvironment rather than direct involvement in driving a fibrous response in plaques. The negative correlation between LIGHT and TGF-β2 is also in line with this notion.
The study has certain limitations. The reported results are mainly related to associations through which conclusions about a causal role in plaque destabilization cannot be drawn. Though MR analyses were performed, we are still limited by available expression data from carotid tissues, specifically plaque tissue. Our LIGHT measurements are limited to one Swedish cohort with advanced atherosclerosis, and generalization would require similar assessments in additional cohorts representing different ethnicities and stages of the disease. Furthermore, the sex of the included subjects is skewed towards men (68%), which broadly mirrors the sex ratios for carotid endarterectomies performed at the Skåne University Hospital but nonetheless limits our ability to stratify analyses by sex. However, we found no association between LIGHT plaque content and sex, making bias less likely. Finally, the proximity extension assay used to measure LIGHT content in plaques is semi-quantitative and, thus, unable to provide absolute plaque-concentration of the analyte. The assay can also not distinguish between soluble and membrane-bound LIGHT in plaque homogenates; however, both forms can induce signaling through herpesvirus entry mediator (HVEM) and LTβR.
Conclusions
The content of the co-stimulatory molecule LIGHT in atherosclerotic plaques not only correlates with markers of plaque destabilization, but is also significantly elevated in plaques from symptomatic patients, compared to those from asymptomatic patients, though a robust causal effect on CVD outcomes could not be shown. These results thus associate LIGHT content with a rupture-prone plaque phenotype, potentially upregulated as part of a reparative response, warranting further studies.
Supplementary materials
Supplementary materials related to this article can be found online at https://doi.org/10.5853/jos.2025.00703.
Patient demographics in the Carotid Plaque Imaging Project (CPIP) study cohort (plaque samples)
Patient demographics in the Carotid Plaque Imaging Project (CPIP) study cohort (patient groups with plasma samples in comparison with plaque samples)
A selection of previous studies and conclusions derived from the study’s OLINK and RNA sequencing datasets from carotid endarterectomy plaques of the Carotid Plaque Imaging Project (CPIP) cohort
Associations between plaque LIGHT and clinical parameters (categorical variables)
Spearman correlations between plaque LIGHT (AU) and clinical parameters (continuous variables)
Genome-wide association summary statistics used in two-sample Mendelian randomization analysis
Results from two-sample Mendelian randomization analyses
Cox regression analysis of the association between the circulating levels of LIGHT in the Carotid Plaque Imaging Project (CPIP) cohort and cardiovascular fatal and non-fatal events (adjusted for confounding factors)
Cox regression analysis of the association between the circulating levels of LIGHT in the Carotid Plaque Imaging Project (CPIP) cohort and cardiovascular fatal events (adjusted for confounding factors)
Forest plots showing causal effects of the genetic LIGHT levels on AS, AIS, LAS, CAD, and MI. The hollow points indicate a nonsignificant association. Solid points indicate significant associations (P<0.05). Results were obtained using an inverse-variance-weighted MR analysis. Data are shown as beta and are scaled to a 1-SD increase in LIGHT levels and grouped by cohorts. AS, any stroke; AIS, any ischemic stroke; LAS, large artery stroke; CAD, coronary artery disease; MI, myocardial infarction; MR, Mendelian randomizatio; SD, standard deviation.
Notes
Funding statement
This work was supported by grants from the Swedish Heart and Lung Foundation (20200403, 20230257 to IG, 20200183 and 20220198 to AS, 20220044 and 20220284 to AE, 20220293 to EB), Swedish Research Council (2019-01260, 2023-02368 to IG, 2019-01907, 2024-02761 to AE, 2023-02521 to EB), Skåne University Hospital (N/A to IG and AE); Lund University Diabetes Centre (LUDC) and Strategic Research Area EXODIAB (2009-1039 to AS, AE, EB, JS, PS, CG, IG) funded by the Swedish Foundation for Strategic Research (LUDC-IRC15-0067 to AS, AE, EB, JS, PS, CG, IG), the Swedish Heart and Lung Association (FA 2019:15 and FA 2020:34 to A.S.), The Swedish Stroke Association (N/A to A.S.), The Royal Physiographic Society of Lund (N/A to A.S.), Anders och Birgit Andersson’s research foundation (RMh2021-0025 to A.S.), Åke Wiberg’s Foundation (M21-0071 and M22-0005 to A.S.), Anna and Edwin Berger’s Foundation (F-22-0027 to A.S.), The Gyllenstierna Krapperup’s Foundation (KR2022-0055 to A.S.), Albert Påhlsson foundation (EB), and the Swedish Society for Medical Research (CG-22-0254 to AE). The LeDucq Foundation Network of Excellence: CHECKPOINT ATHERO (22CVD02 to I.G.), the Knut and Alice Wallenberg Foundation, the Medical Faculty at Lund University and Region Skåne are also acknowledged for generous financial support.
Conflicts of interest
Consulting fees to AE from Novo Nordisk, Sanofi, Amarin, and Amgen without relationship with the current study. Other authors have no disclosures.
Author contribution
Conceptualization: AS. Study design: AS, EL, IG, JN. Methodology: CG, IG. Investigation: AS, EB, IG, PS. Statistical analysis: AS, JS, MP. Writing—original draft: AS. Writing—review & editing: all authors. Funding acquisition; AE, AS, EB, IG. Approval of final manuscript: all authors.
Acknowledgments
The authors wish to thank Mihaela Nitulescu, Ana Persson, and Lena Sundius for expert technical assistance.