Association of Neurexin 3 polymorphisms with smoking behavior

Authors


R. Rabionet, Genes and Disease Program, Center for Genomic Regulation (CRG-UPF), C/Dr. Aiguader 88, 08003 Barcelona, Spain. E-mail: kelly.rabionet@crg.eu

Abstract

The Neurexin 3 gene (NRXN3) has been associated with dependence on various addictive substances, as well as with the degree of smoking in schizophrenic patients and impulsivity among tobacco abusers. To further evaluate the role of NRXN3 in nicotine addiction, we analyzed single nucleotide polymorphisms (SNPs) and a copy number variant (CNV) within the NRXN3 genomic region. An initial study was carried out on 157 smokers and 595 controls, all of Spanish Caucasian origin. Nicotine dependence was assessed using the Fagerström index and the number of cigarettes smoked per day. The 45 NRXN3 SNPs genotyped included all the SNPs previously associated with disease, and a previously described deletion within NRXN3. This analysis was replicated in 276 additional independent smokers and 568 controls. Case–control association analyses were performed at the allele, genotype and haplotype levels. Allelic and genotypic association tests showed that three NRXN3 SNPs were associated with a lower risk of being a smoker. The haplotype analysis showed that one block of 16 Kb, consisting of two of the significant SNPs (rs221473 and rs221497), was also associated with lower risk of being a smoker in both the discovery and the replication cohorts, reaching a higher level of significance when the whole sample was considered [odds ratio = 0.57 (0.42–0.77), permuted P = 0.0075]. By contrast, the NRXN3 CNV was not associated with smoking behavior. Taken together, our results confirm a role for NRXN3 in susceptibility to smoking behavior, and strongly implicate this gene in genetic vulnerability to addictive behaviors.

According to the 2011 World Health Organization report on the global tobacco epidemic, tobacco kills nearly 6 million people each year, causing hundreds of billions of dollars of economic damage worldwide (WHO 2011). In fact, smoking is the single greatest contributor to preventable ill health and premature death (reviewed in Bierut 2011). The development of nicotine addiction is influenced by environmental and genetic factors, and while environmental factors have a stronger influence on initiation, genetic factors play a more significant role in the transition from regular use to addiction (Bierut 2011; Vink et al. 2005).

A genetic component in the development of nicotine addiction has been showed in twin and family studies, showing an estimated heritability of 30–72% (Agrawal et al. 2008). Twin studies also showed a hereditary component for a range of diverse smoking-related phenotypes, including age at initiation, intensity and cessation (Heath et al. 2002). Similarly, familial studies showed that siblings of habitual-smoking probands are at greater risk of becoming habitual smokers (Bierut et al. 1998).

At the molecular level, efforts to untangle the genetic component of nicotine addiction have focused on both the study of candidate genes and on hypothesis-free whole genome analyses. To date, the most robust genetic finding is the link between nicotine addiction and two gene clusters of nicotine receptors: α3, α5 and β4 (CHRNA3, CHRNA5 and CHRNB4) on chromosome 15; and α6 and β3 (CHRNA6 and CHRNB3) on chromosome 8 (Tagconsortium 2010; Thorgeirsson et al. 2008). In spite of the reported strong associations, the contribution of nicotinic receptor genes to susceptibility to nicotine addiction only partially explains the heritability of this condition, suggesting that other genes contribute in an additive or epistatic manner to the development of nicotine addiction.

Nicotine dependence has also been associated with the Neurexin genes (Bierut et al. 2007; Nussbaum et al. 2008). Neurexins are presynaptic cell adhesion proteins that are essential for the development and function of GABAergic and glutamatergic synapses (Craig et al. 2006; Sudhof 2008). These synapses participate in key circuits influencing addictive behaviors (Lein et al. 2007; Ullrich et al. 1995). Two members of the neurexin gene family have been associated with nicotine dependence, Neurexin 1 and Neurexin 3 (Bierut et al. 2007; Nussbaum et al. 2008). Moreover, Neurexin 3 has also been associated with the degree of smoking in schizophrenic patients (Novak et al. 2009), impulsivity and substance abuse (Stoltenberg et al. 2011), illegal substance abuse (Lachman et al. 2007) and alcohol dependence (Hishimoto et al. 2007).

In addition to single nucleotide polymorphisms (SNPs), structural variants involving neurexin genes have been described in autism spectrum disorders (Ching et al. 2010; Gauthier et al. 2011; Kim et al. 2008; Vaags et al. 2012; Wisniowiecka-Kowalnik et al. 2010) and schizophrenia patients (Kirov et al. 2008). These copy number variants (CNVs) are rare, large rearrangements. A common CNV in Neurexin 3 (NRXN3) has also been described (Conrad et al. 2010). The latter CNV, termed NRXN3_del henceforth, involves a common insertion/deletion polymorphism spanning 8.9 kb of the second intron of the NRXN3 β isoform, and its association with disease/psychiatric phenotypes has not been investigated. As tobacco addiction is a common mild phenotype, NRXN3_del may be involved in susceptibility to smoking behavior.

In the present study we investigated the role of NRXN3 in nicotine addiction. We performed a comprehensive genetic study in Spanish smokers and control subjects, which included analysis of tagSNPs and candidate SNPs, and the common CNV located within the NRXN3 gene.

Materials and methods

Samples

All participants in the smoking group were individuals interested in receiving treatment to give up smoking. The original cohort (TAB_Discovery) included 157 smokers recruited and evaluated at the Tobacco Unit of the Addictive Behavior Unit at the Hospital Universitari Vall d’ Hebron (Barcelona, Catalonia, Spain). These subjects were evaluated with the Addiction Research Unit questionnaire (ARU) (West & Russell 1985), the Beck Depression Inventory (BDI) (Beck et al. 1961) and the State-Trait Anxiety Inventory (STAI) (Spielberger et al., 1970). Subjects smoking regularly that requested for treatment to quit smoking were considered as smokers. The replication sample (TAB_replication) consisted of 276 smokers, recruited at different primary care centers in Catalonia, Spain (ClinicalTrials.gov identifier: NCT00125905). These subjects were evaluated with the Richmond test, to assess their motivation to quit smoking. In these sample sets, subjects smoking regularly and willing to quit smoking were considered as smokers. Nicotine dependence was assessed in both samples using the Fagerström index (Heatherton et al. 1991) and through the number of cigarettes smoked per day.

The control sample consisted of 1163 healthy blood donors recruited from the Blood bank (Banc de Sang i de Teixits) at the Hospital Universitari Vall d’Hebron. Blood donors were asked to fill in a questionnaire including the question ‘have you ever smoked’; subjects providing a negative answer were considered non-smokers. This control sample was further divided into two random cohorts using the ‘Research Randomizer’ v.3.0 software (http://www.randomizer.org/): CVH1 (n = 595) and CVH2 (n = 568). CVH1 and CVH2 were used as controls for the discovery and replication cohorts, respectively. No significant differences in sex or age were detected between the groups, and the demographic and clinical attributes associated with the samples are summarized in Table 1.

Table 1.  Summary of the demographic and clinical data from the cohorts included in the study
VariableControlsCases
CVH1CVH2TAB_ DiscoveryTAB_ Replication
 (N = 595)(N = 568)(N = 157)(N = ‘276)
Gender, n (%)
 Male216 (36.6)180 (31.7)82 (53.6)118 (43.2)
 Female375 (63.4)388 (68.3)69 (46.4)155 (56.8)
Mean age, yrs (±SD)53.8 (16.8)54.1 (17.5)46.6 (7.7)45.5 (13.5)
Cigarettes/day (±SD)21.9 (9.9)19.8 (11.4)
Fagerström score (±SD)5.12 (2.42)4.16 (2.58)

All participants, both cases and controls, were of Spanish Caucasian origin and they provided informed consent before enrollment. The project was approved by the ethics committees at all the participating recruitment centers.

Genotyping of NRXN3 variants

We selected 45 SNPs within the genomic region of NRXN3 using the Haploview software (Barrett et al. 2005). These SNPs capture 42% of alleles in NRXN3α (1460 kb) and 66% in NRXN3β at r2≥ 0.8, based on the CEU HapMap genotyped SNPs (version 2, release 21). In the selection process, potential functional variants, variants located near splice sites and SNPs previously associated with disease were included as TagSNPs (Heard-Costa et al. 2009; Hishimoto et al. 2007; Novak et al. 2009). A total of 45 SNPs (41 TagSNPs and 4 singleton SNPs) and 1320 samples (CVH1, CVH2 and TAB_Discovery) were genotyped (Table 2) using VeraCode technology (Illumina, Inc., San Diego, CA, USA) at the CeGen genotyping facility (Centro Nacional de Genotipado, Genoma España, Barcelona), following the manufacturer's instructions. The assays developed for the VeraCode beads were analyzed using Illumina's BeadXpress Reader System and the data analyzed using BeadStudio v.2.0 software (Illumina, Inc., San Diego, CA, USA). As a quality control, 5% of the genotyped samples were duplicated, and no-template controls and six Hapmap trios were included(NA10840–NA12286–NA12287, NA12766–NA12775–NA12776, NA12818–NA12829–NA12830, NA 12832–NA12842–NA12843, NA12865–NA12874–NA12875, NA128 77–NA12889–NA12890).

Table 2.  SNPs tested for association
NamePositionHwpval%GenoMAFAllelesP value*
  1. Markers included in the NRXN3-del genotyping appear in bold.

  2. *Genotypic association under the log-additive model corrected by sex and age.

  3. SNPs analyzed in substance-abuse-related papers.

rs100421278250978199.60.134G:A0.1609
rs3143178510788199.90.124C:A0.4279
rs171084577852053011000.111A:T0.2536
rs2202167785693770.612199.60.385C:A0.5141
rs2202175786254030.242199.70.394A:C0.9386
rs12323794786313510.28651000.365G:A0.5657
rs65744957866902611000.396G:A0.6821
rs116272697878432411000.332G:A0.6109
rs4603484789254690.97251000.432A:C0.3402
rs2196443789469680.16081000.194A:C0.0369
rs10146997790149140.0471000.179A:G0.4977
rs11848580790385400.033299.90.392A:G0.0368
rs1424850790677560.281000.285T:A0.0002
rs221415791156210.8291000.183A:G0.1393
rs715362579119014199.70.172G:A0.4427
rs8022725791292220.80111000.359G:C0.2103
rs2543576791419810.43399.90.446G:A0.0143
rs221497791582320.38671000.112G:A0.0020
rs2214737917436811000.187A:T0.0009
NRXN3del79176041165.60G:GNA
rs12894142791792131.11E-0783.40.027A:GNA
rs12100748791837475.00E-0483.80.007G:ANA
rs22144979193929199.50.258G:A0.1448
rs101305937921181411000.321C:G0.4242
rs1030127792311920.86931000.123A:G0.5384
rs178377792477780.63151000.147A:G0.3535
rs802176779263758199.90.198G:A0.1039
rs5014481792826140.20731000.429C:A0.7918
rs8018724793185570.44321000.368C:A0.3199
rs2215840793398200.70871000.371A:G0.9289
rs10145867793399260.77311000.083G:A0.2061
rs932265793601770.40851000.159G:A0.1521
rs1159039793613450.27361000.344A:G0.8745
rs760288793766820.58891000.336A:G0.6339
rs2293839793872120.24071000.304G:A0.9259
rs8019381793903350.36271000.162G:A0.4028
rs994010794193550.67631000.235A:G0.5751
rs9323679794215430.34531000.145A:G0.6752

We also explored the possible involvement of a previously described NRXN3 CNV in susceptibility to smoking behavior. This CNV, NRXN3_del, consists of a common insertion/deletion polymorphism spanning 8.9 kb of the second intron of the NRXN3 β isoform (Conrad et al. 2010). To genotype this structural variant, we included two SNPs located within the deleted region in the Veracode assay (rs12894142 and rs12100748), and we designed a third SNP assay (NRXN3del) with each of its extension probes flanking the breakpoints of the CNV. A combination of the results for these SNPs was used to assess the genotype. In addition, NRXN3_del was genotyped by multiplex polymerase chain reactions (PCRs) with 5′ FAM modification, followed by capillary electrophoresis in a 3730XL automatic sequencer and analysis with the Gene Mapper package (Applied Biosystems, Foster City, CA, USA). In each experiment, we included both negative and positive controls, consisting of three Hapmap samples with validated genotypes for this CNV (Conrad et al. 2010): NA06991 (homozygous deleted), NA010831 (heterozygous) and NA06994 (homozygous non-deleted). The observed concordance between the two assays was 100%.

SNPs with a significant association in the discovery samples (rs1424850, rs221473 and rs221497) were subsequently genotyped using the KBiosciences PCR SNP genotyping system (KASPar®, Kbioscience, Hoddesdon, Herts, UK), which uses a competitive allele-specific PCR, following the manufacturer's instructions. Genotyping of the independent smoker cohort (276 additional smokers) was performed at the CeGen genotyping facilities, in the CNIO Node (Centro Nacional de Genotipado, Genoma España). As a quality control, 5% of the genotyped samples had their genotypes confirmed by Sanger sequencing.

Statistical analysis

Quality control and case–control association analyses for both Veracode and Kaspar assays were performed using PLINK software, version 1.07 (http://pngu.mgh.harvard.edu/purcell/plink/) (Purcell et al. 2007). SNPs with a low genotyping rate (<95%), not fulfilling the Hardy–Weinberg equilibrium (HWE: P < 0.05), or with a minimum allele frequency below 5%, as well as samples with a low genotyping rate (<95%), were excluded from the association analyses. We tested for association for two phenotypes: smoking behavior, and level of addiction based on the Fagerström score. For this, we considered smoking behavior as a dichotomic phenotype (smoker vs. non-smoker), and we used the total Fagerström score, either as a quantitative or a nominal variable: minimally (<4), moderately (4–6) and highly (7–10) dependent. All association tests included sex and age as covariates, and they were corrected for multiple testing based on the Bonferroni correction (35 markers: P < 0.0014). Linkage disequilibrium (LD) between polymorphisms and haplotype block structures were evaluated using Haploview software (version 4.1). Regions of strong LD were defined according to the confidence intervals algorithm (Gabriel et al. 2002). We studied haplotypes present at a frequency of at least 5%, and estimated the significance of the best result via a permutation procedure (2000 permutations). The SNPassoc R package was used for all analyses (Gonzalez et al. 2007).

Assessment of SNP function

We used WGAviewer (Ge et al. 2008) and Pupasuite 3.1 (http://pupasuite.bioinfo.cipf.es/) to assess the possible functional consequences of the SNPs associated with smoking behavior. The GENEVAR tool of WGAviewer was used to evaluate the effects of SNPs on gene expression, based on Hapmap genotype and expression data (Stranger et al. 2005, 2007). Pupasuite was used to predict the pathogenicity of an SNP based on the disruption of potential transcription factor binding sites, the creation or disruption of splice sites, its location in miRNA sequences or targets, or in promoter or conserved regions.

Results

Discovery cohort analysis

Ten SNPs (4 with a low genotyping rate, 1 not fulfilling HWE and 5 with a low MAF) and 20 samples (with a low genotyping rate) did not fulfill the aforementioned quality control criteria and, thus, a total of 38 SNPs were analyzed in 153 cases (TAB_Discovery) and 583 controls (CVH1) (Fig. 1 and Table 2). No Mendelian errors were detected in the HapMap trios, and 100% concordance was observed with the described genotype in all cases. Allelic and genotypic (log-additive model) association tests showed that three SNPs were significantly associated with a lower risk of being a smoker after Bonferroni correction for multiple testing [rs1424850 [P = 0.0002, odds ratio (OR) = 0.55, 95% confidence interval (CI) = 0.40–0.76], rs221497 (P = 0.0020, OR = 0.47, 95% CI = 0.28–0.79) and rs221473 (P = 0.0009, OR = 0.54, 95% CI = 0.37–0.79); Table 3]. Based on the LD pattern of the region, nine haplotype blocks were defined, and haplotype association analysis showed an association of block 3 with a decreased risk of being a smoker (OR = 0.44, 95% CI = 0.26–0.75; permuted P = 0.0025). This block spanned 16 kb and included two of the SNPs identified in the single-SNP analysis (rs221497 and rs221473; Table 4 and Fig. 2). We also tested for any association with the level of addiction, as determined by the Fagerström index, although no significant relationships were observed for any of the SNPs evaluated.

Figure 1.

SNPs location. Location of the 38 SNPs (including the CNV assay) along the α and β NRXN3 genomic region. SNPs exhibiting association with nicotine addiction appear in bold.

Table 3.  Summary of the Veracode assay results: SNPs with a statistically significant association
 TAB_Discovery vs. CVH1TAB_Replication vs. CVH2All
P value*OR (95% CI)P value*OR (95% CI)P value*OR (95% CI)
  1. *Genotypic association under the log-additive model corrected by sex and age.

  2. Significant after Bonferroni correction for multiple testing.

rs14248500.00020.55 (0.40–0.76)0.590.94 (0.73–1.19)0.00450.76 (0.73–0.92)
rs2214970.00200.47 (0.28–0.79)0.0470.70 (0.49–1.00)0.00050.62 (0.46–0.82)
rs2214730.00090.54 (0.37–0.79)0.0390.75 (0.57–0.99)0.00030.67 (0.54–0.84)
Table 4.  Haplotype association
 TAB_Discovery vs. CVH1TAB_Replication vs. CVH2All
P value*OR (95% CI)P value*OR (95% CI)P value*OR (95% CI)
  1. *Genotypic association under the log-additive model corrected by sex and age.

  2. rs221497(A)-rs221473(T): freq 0.11.

  3. global P value for 2000 permutations.

Haplotype0.00310.46 (0.27–0.77)0.0370.68 (0.47–0.98)0.00030.57 (0.42–0.77)
Permuted haplotype0.0170.230.0075
Figure 2.

Linkage disequilibrium plot. LD of the 38 (35 + 3 assessing NRXN3_DEL) included in the analysis, as assessed using Haploview software. The D′ score value is provided inside the boxes.

Association analysis for NRXN3_del was performed separately from the SNP analysis. After quality control, 154 cases and 580 controls were included in the analysis, and this deleted allele was observed at a similar frequency in both the cases and controls (cases: 0.42; controls: 0.40). Hence, there was no association between the CNV polymorphism and smoking behavior (P = 0.3, genotypic association; P = 0.46, allelic association Fisher test).

Replication

To replicate these findings, the 3 SNPs exhibiting an association were genotyped in an independent cohort of 276 individuals addicted to nicotine (TAB_replication) and in an independent control set (CVH2). We found a nominal association for rs221473 (P = 0.039, OR = 0.75, 95% CI = 0.57–0.99) and rs221497 (P = 0.047, OR = 0.70, 95% CI = 0.49–1.00), and their composite haplotype. When considering both cohorts, the level of significance increased and haplotype association persisted after permutation (OR = 0.57; 95% CI = 0.42–0.77, permuted P = 0.0075; Tables 3 and 4).

Evaluation of SNP function

The three SNPs significantly associated with smoking behavior were located in an intronic region, and in silico evaluation of their functional implications using Genevar showed no correlation between the SNP genotypes and levels of expression. Evaluation of possible functional effects using Pupasuite 3.1 showed that rs221497 and rs221449 were highly conserved as compared with the mouse genome, although no potential functional consequences were showed.

Discussion

In the present study, we performed a high-density SNP association study of the link between NRXN3 and smoking behavior in two independent samples of Spanish smokers. Our findings show a significant association between two NRXN3 SNPs and smoking behavior, and a nominal association for a third SNP. To the best of our knowledge, this is the first study to assess genetic variation in the NRXN3 gene by analyzing tagSNPs, additional putative functional variants and structural variation.

NRXN3 is one of the three members of the neurexin gene family. Neurexins are presynaptic cell adhesion proteins that are essential for the development and function of synapses, and that are implicated in neurotransmitter release. NRXN3 is located on chromosome 14 and the two existing promoters drive the expression of its α and β isoforms. Together with five canonical sites of alternative splicing, these promoters give rise to over 1000 isoforms (Rowen et al. 2002; Tabuchi & Sudhof 2002). NRXN3 is expressed in key circuits that have been implicated in addictive behaviors, including glutamatergic neurons projecting from the prefrontal cortex to the striatum and GABAergic neurons (Lein et al. 2007; Ullrich et al. 1995). Hence, NRXN3 is a strong candidate for substance abuse susceptibility.

Genetic variants in NRXN3 have been associated with various substance dependence phenotypes (Bierut 2007; Hishimoto 2007; Liu 2005; Novak 2009; Stoltenberg 2011), including the degree of smoking in schizophrenic patients (Novak et al. 2009) and nicotine dependence (Bierut et al. 2007). The convergence of associations between NRXN3 and different phenotypes suggests that genetic factors contribute to developing dependence on various classes of drugs (Bierut 2011). Nevertheless, there is only a partial overlap in the SNPs analyzed in each of the studies, various SNPs have been analyzed in only one study, and the phenotypes tested in the different studies, although related, are not the same. It is possible that different SNPs in the same gene are associated with the different substance dependence phenotypes. Several of the SNPs included in this work have also been analyzed in other studies (Table 2), although most of them did not show any association. rs1004212 had been associated with heavy smoking in schizophrenic patients and has shown nominal association with alcohol problems in men (Stoltenberg 2011), and rs8019381 had been associated with alcohol dependence (Hishimoto 2007). However, we did not find these SNPs associated with smoking behavior. Several explanations can be offered for the conflicting SNPs identified in different studies (including ours). First, as mentioned above, although related, the phenotypes investigated in the different studies are not the same, thus allowing for different SNP associations. Also, previous studies were performed on American (Bierut et al. 2007; Stoltenberg et al. 2011), Canadian (Novak et al. 2009) and Australian (Bierut et al. 2007) populations, indicating that the failure to replicate these data may also be due to genetic heterogeneity among populations. Alternatively, the observed differences may reflect distinct gene × environment interactions (Ioannidis et al. 2007). Finally, it is possible that an untyped causative variant in LD with the different signals identified in the NRXN3 gene may exist.

Population admixture in our sample may represent a potential limitation to our analysis, although no significant population effects have been observed in previous studies on Spanish populations (Julia et al. 2008; Ribases et al. 2009). Moreover, all samples in the present study were recruited in the same geographical region (Catalonia) and, thus, the west-to-east trend reported previously is unlikely to influence our data (Julia et al. 2008). Indeed, population admixture was excluded from part of the control cohort in an independent study (Ribases et al. 2009) and, hence, the influence of population admixture in the present work is probably negligible.

In our two-step study, the association observed was evident in both the discovery and replication cohorts, although the level of significance was lower in the latter of the two cohorts. This is most probably due to clinical heterogeneity between these patient cohorts, as the Fagerström index and the number of cigarettes smoked per day were significantly higher in the discovery cohort. Nonetheless, global analysis of both cohorts together showed a significant protective pooled effect.

Although we detected an association with status as a smoker, we observed no association with the Fagerström index (Table S1), suggesting that although related, these two phenotypes may have distinct genetic bases. Alternatively, a lack of statistical power may explain the failure to detect association with the Fagerström index, given the smaller effective sample size once the sample is divided into the different Fagerström categories. In fact, we did observe an association in two separate smoker cohorts, albeit in small sample sets. Performing the replication study in a larger cohort from a different population would provide strong support for the present findings, and may even show an association with the different categories defined by the Fagerström index.

The two NRXN3 SNPs associated with smoking behavior are located in an intronic region and no specific effect was predicted for either SNP. Although this intronic region is highly conserved between species, suggesting that it fulfills an important role, it is possible that the SNPs identified merely tag an as-yet-unidentified functional variant. Next-generation sequencing technologies may help identify additional functional variants within the gene that could have been tagged by the associated SNPs. Targeted sequencing of NRXN3 in larger nicotine addiction cohorts will be an important next step to determine its role in substance abuse. To better understand the functional involvement of these and other SNPs in NRXN3, it will be of interest to correlate these variants with the expression of different splice variants. Other functional assays, such as testing the response to nicotine exposure of neuronal cell lines carrying the different genotypes, will also help elucidate the functional consequences of these polymorphisms.

Despite finding no association between the NRXN3_del variant and being a smoker, we have confidently identified an INDEL by genotyping an internal SNP in combination with a specifically designed assay. The concordance of this genotyping technique with more traditional multiplex PCR was 100%, suggesting that it constitutes a useful tool to genotype common CNVs with known and defined breakpoints in a high-throughput manner.

In summary, we describe new NRXN3 SNPs that are associated with susceptibility to being a smoker in agreement with previous studies implicating neurexins in the development of addiction.

Acknowledgments

This work was supported by the Spanish Ministry of Science and Innovation (SAF2008-00357), the European Commission (ENGAGE_201413), the Fundació de la Marató de TV3 (041810), and the Generalitat de Catalunya. E.D. is supported by the Spanish Ministry of Health (FIS-PFIS, FI 08/00148). R.R. received a Ramon y Cajal grant from the Spanish Ministry of Health (RyC04_Raquel Rabionet) and M.R. is supported by a Miguel de Servet contract from the “Instituto de Salud Carlos III, “Ministerio de Ciencia e Innovación”, Spain.

Ancillary