Converging evidence for an impact of a functional NOS gene variation on anxiety-related processes

Abstract

Being a complex phenotype with substantial heritability, anxiety and related phenotypes are characterized by a complex polygenic basis. Thereby, one candidate pathway is neuronal nitric oxide (NO) signaling, and accordingly, rodent studies have identified NO synthase (NOS-I), encoded by NOS1, as a strong molecular candidate for modulating anxiety and hippocampus-dependent learning processes. Using a multi-dimensional and -methodological replication approach, we investigated the impact of a functional promoter polymorphism (NOS1-ex1f-VNTR) on human anxiety-related phenotypes in a total of 1019 healthy controls in five different studies. Homozygous carriers of the NOS1-ex1f short-allele displayed enhanced trait anxiety, worrying and depression scores. Furthermore, short-allele carriers were characterized by increased anxious apprehension during contextual fear conditioning. While autonomous measures (fear-potentiated startle) provided only suggestive evidence for a modulatory role of NOS1-ex1f-VNTR on (contextual) fear conditioning processes, neural activation at the amygdala/anterior hippocampus junction was significantly increased in short-allele carriers during context conditioning. Notably, this could not be attributed to morphological differences. In accordance with data from a plethora of rodent studies, we here provide converging evidence from behavioral, subjective, psychophysiological and neuroimaging studies in large human cohorts that NOS-I plays an important role in anxious apprehension but provide only limited evidence for a role in (contextual) fear conditioning.

Introduction

Anxiety disorders are highly prevalent and costly psychiatric disorders (Gustavsson et al., 2011) which can be studied in the laboratory by using fear conditioning paradigms (Lissek et al., 2005; Mineka and Zinbarg, 2006; Duits et al., 2015). Thereby, processes involved in phasic fear and sustained anxiety (Davis et al., 2010), which represent different features of anxiety disorders (Grillon et al., 2006), can be experimentally investigated through cued and contextual conditioning paradigms, respectively. Phasic fear responses are elicited by aversive events [unconditioned stimulus (US)] which can be predicted by a discrete cue (i.e. cued conditioning). Unavailability of such discrete predictors [conditioned stimuli, (CSs)] results in unpredictability of danger which in turn induces longer-lasting sustained anxiety responses to the global situation (i.e. context conditioning) (Fanselow, 1994; Davis et al., 2010).

The related, albeit distinct neural systems underlying cued (phasic) fear and contextual (sustained) anxiety responses have successfully been delineated in both rodents (Davis et al., 2010) and humans (Marschner et al., 2008; Alvarez et al., 2011; Andreatta et al., 2015), which underscores the translational validity of this experimental approach. Thereby, the amygdala and insula have been implicated in cued fear, while contextual anxiety involves the amygdala, the bed nucleus of the stria terminalis and most importantly the hippocampus (Marschner et al., 2008; Lang et al., 2009; Davis et al., 2010; Alvarez et al., 2011; Pohlack et al., 2012; Winkelmann et al., 2015).

These well-described neural underpinnings of fear learning provide a starting point to dissect the molecular pathways underlying individual differences in fear acquisition, an approach that is currently at the forefront of research (Graham et al., 2014). In this respect, laboratory models of fear conditioning, which can be directly translated across species, are optimally suited because of substantial heritability of both experimental fear conditioning (Merrill et al., 1999; Hettema et al., 2003; Lonsdorf and Kalisch, 2011) and clinical anxiety disorders (Gordon and Hen, 2004; Leonardo and Hen, 2006). In addition, fear conditioning represents a simple behavioral paradigm that elicits robust behavioral responses with sufficient inter-individual variability and adequate re-test reliability (Zeidan et al., 2012) that can be easily measured and quantified (Lonsdorf and Kalisch, 2011; Lonsdorf and Baas, 2015). According studies, however, have mainly focused on phasic fear responses (Lonsdorf et al., 2009; Raczka et al., 2010; Lonsdorf and Kalisch, 2011; Lonsdorf et al., 2014a), while sustained anxiety responses following context conditioning (Pohlack et al., 2011; Glotzbach-Schoon et al., 2013; Baas and Heitland 2015; Mühlberger et al., 2014) were only considered more recently. A particularly strong candidate for studying biologically based individual differences in hippocampus-dependent context conditioning is the ubiquitous signaling molecule nitric oxide (NO). In rodents, the NO synthesizing enzyme, NO synthase 1 (NOS-I, encoded by the Nos1 gene) has convincingly been implicated in anxiety-like behavior, synaptic plasticity, hippocampal long-term potentiation and memory formation (Schuman and Madison, 1991). Most importantly, the hippocampus features high expression levels of NOS-I (Schuman and Madison, 1991), and Nos1 knockout (Kelley et al., 2009) as well as pharmacological manipulations (Kelley et al., 2010) have a marked effect on freezing in rodents, particularly in context but not cued fear conditioning. Consequently, there is strong a priori evidence and biological plausibility for the involvement of NO signaling in hippocampus-dependent context conditioning and anxiety-related processes.

In humans, biomarkers or safe drugs targeting NOS-I are not available (Freudenberg et al., 2015) and consequently indirect measures probing NO signaling in the brain have to be considered, such as functional variation in the human NOS1 gene. NOS1 features 12 alternative untranslated first exons (1a-1l) driven by distinct promoters (Reif et al., 2009; Freudenberg et al., 2015), whereof exon 1f is preferentially expressed in the hippocampus and basal ganglia (Bros et al., 2006; Reif et al., 2006). The corresponding promoter region harbors a functional polymorphism (NOS1 ex1f-VNTR) which affects gene expression levels (long > short alleles) in reporter gene systems using human cells and human post-mortem studies (Reif et al., 2006, 2009; Rife et al., 2009; Weber et al., 2015).

We here present converging and comprehensive evidence for a link between the NOS1 ex1f-VNTR s-allele and anxious apprehension using a multi-methodological approach employing both a posteriori and a priori genotyping strategy in a total of 1019 healthy individuals, while there is limited direct evidence for a modulatory role of the NOS1 ex1f-VNTR in (context) conditioning in humans.

Materials and methods

Subjects

All subjects provided personality and life events questionnaires within the framework of a collaborative research center (SFB TRR-58, see Supplementary Data). Study 1 is based on data from this screening sample, while participants were re-invited for studies 2, 4 and 5 and participants from study 3 were re-invited from a separate screening sample. All subjects provided written informed consent, and protocols were approved by the General Medical Council of the State of Hamburg (studies 1, 2, 4 and 5) or the Medical Faculty of the University of Würzburg (study 3). See Table 1 for sample descriptives.

Table 1.

Detailed sample descriptions for studies 1–5

Study	Method	Site	N Total	N NOS1-ex1f-VNTR			HWE (P value)	Sexa (m/f)	Mean age (SEM)
				S/S	S/L	L/L
1	Questionnaires	HH	946	200	464	284	0.68	591/357	25.6 (0.17)
2b	Behavioral	HH	92	21	38	33	0.13	27/65	25.8 (0.46)
3	VR-behavioral	WU	73	20	34	19	0.56	29/44	24.1 (0.41)
4b,c	fMRI	HH	49	25 (11/14)		24	NA	24/25	25.0 (0.45)
5b	sMRI	HH	352	139 (59/180)		113	0.37	254/98	25.8 (0.25)

Note. S, short allele; L, long allele; VR, virtual reality; fMRI, functional magnetic resonance imaging; HWE, Hardy–Weinberg equilibrium; NA, not applicable as the sample was stratified for NOS1 ex1f-VNTR; HH, Hamburg; WU, Würzburg.

^aSex was equally distributed across all three genotypes for all studies (χ² test, all P > 0.158).

^bParticipants were re-invited from study 1.

^cThe sample size of N > 20 was chosen to ensure adequate power based on Thirion et al., (2007).

Genotyping and genotype grouping

Genomic DNA was prepared from whole blood and samples were genotyped for NOS1 ex1f-VNTR (Table 1) as described previously (Reif et al., 2009) or other methods (details available on request). Non-concordant genotypes (N = 8), where available from both methods, were excluded. Alleles were dichotomized in short (S) and long (L) alleles (Reif et al., 2006, 2009). While in all published cases the S-allele was associated with worse outcome (irrespective of phenotype), the genetic model varied in previous studies with evidence for both dominant (Kurrikoff et al., 2012) recessive (Reif et al., 2009, 2011; Hoogman et al., 2011; Kurrikoff et al., 2012), as well as allelic (Weber et al., 2015) effects. Anxiety-related phenotypes generally seem to reflect a dominant model, while impulsive traits tend to be linked to a recessive model. As published studies are by far too sparse to draw firm conclusions, both a recessive as well as a dominant model were applied to studies 1–3, in both cases considering the S-allele being the risk allele.

Fear- and anxiety-related traits

Six anxiety-related questionnaires available for the screening sample (study 1) were chosen a priori [German versions of trait anxiety (STAI-T, Spielberger et al., 1970), trait worrying (PSWQ, Meyer et al., 1990), social anxiety (SPAI, Turner et al., 1989), anxiety sensitivity (ASI, Reiss et al., 1986), behavioral inhibition (BIS, Strobel et al., 2001) and negative affect (PANAS-N, Watson et al., 1988)]. A multivariate analysis of variance including genotype as between-subject factor was performed.

To explore the specificity and trans-diagnostic impact of NOS1 ex1f-VNTR beyond anxiety-related traits, an exploratory analysis was tested for an association of genotype with self-reported depression (ADS-K, Hautzinger and Bailer, 1993).

Experimental design—contextual fear conditioning

Studies 2 and 4 employed a well-established (Haaker et al., 2013; Lonsdorf et al., 2014b) combined cue and context fear conditioning paradigm (Figure 1A). In brief (see Supplementary Data for details), subjects passively viewed a sequence of three different rooms (duration study2/study4: 60 s/45 s each) as context CSs (CXT). The rooms were assigned counterbalanced to a cue conditioning (predictable), a context conditioning (unpredictable) or a safe condition. During fixed time windows, three different symbols that served as cue CSs (Cue, 5 s) were superimposed on the, respectively, assigned room (Figure 1A). A black screen with a white fixation cross served as inter-trial interval (ITI) for 6–8 s (mean 7 s). An individually adjusted electro-tactile stimulus served as US. To induce cue conditioning in the predictable condition, the US was delivered coinciding with cue offset as a discrete predictor (PCue) for the US within the context (PCXT). For context conditioning, in the unpredictable condition, the US was administered on average two times (range 1–3) within fixed time windows during the context (UCXT). No US was delivered while the cue (UCue) was present (with two exceptions to avoid perception of the UCue as a safety signal), making the context itself the best US predictor. In the safe condition, no US was administered. Sequences of context and cue presentation were divided into three blocks, each followed by ratings of fear/tension/stress on a 100-point visual analog scale for each CXT and Cue.

Fig. 1.

Experimental timeline and trial structure for the combined cue and context conditioning paradigm in (A) studies 2 (behavioral) and 4 (fMRI) and (B) study 3 (VR-behavioral). All context and cue stimuli were counterbalanced between subjects and across all conditions. Note that experimental stimuli are taken from studies 4 (A) and 3 (B) for illustrative purposes. Bolt denotes US. UCXT: unpredictable context; PCXT: predictable context; SCXT: safe context; ITI: inter-trial-interval. In this example, the rubber serves as unpredictable cue (UCue), the highlighter as predictable cue (PCue) and the USB-stick as safe cue (SCue).

In study 3, data from a virtual reality (VR) contextual fear conditioning paradigm were reanalyzed (Glotzbach-Schoon et al., 2013) (VR-behavioral, Figure 1B). Briefly (see Supplementary Data for details), in a counterbalanced, pseudo-randomized sequence, subjects were passively moved through two different virtual office rooms. One room served as unpredictable context (UCXT) where one to three electro-tactile USs were unpredictably delivered. Another room served as safe context (SCXT) where no USs were delivered (Figure 1B). Additionally, a virtual corridor served as ITI. Ratings including evaluations of perceived anxiety, valence, arousal and US expectancy were provided on a 100-point visual analog scale after both acquisition phases for the UCXT and SCXT.

Psychophysiological measures

Phasic skin conductance responses to cue and context onsets (SCRs, studies 2 and 4) and tonic skin conductance levels during contexts (SCLs, study 3) as well as fear potentiated startle responses (FPS, studies 2 and 3) were employed. SCL was not acquired in studies 2 and 4 due to the intermittent cue presentations during contexts. Details on data acquisition, post-processing and response definition as employed in studies 2–4 (Glotzbach-Schoon et al., 2013; Haaker et al., 2013; Lonsdorf et al., 2014b) have been reported previously for SCR, SCL and FPS (see Supplementary Data for details).

Data analyses—ratings and psychophysiology

Ratings were analyzed by an analysis of variance with context (study 2/4: 3; study 3: 2) as the within-subject variable and genotype as the between-subject variable for mean ratings across all rating blocks. For SCR, SCL and FPS, repeated-measures analyses of variance only the late acquisition phase, which is reflective of the end-point of learning, was considered. In addition, the ITI responses were included for FPS to allow for analysis of generalized startle potentiation to all CSs as compared to baseline. An α level of P ≤ 0.05 was considered significant and Greenhouse–Geisser correction was applied if necessary. Partial Eta² (referred to as η²) is provided as a measure of effect size.

Data analyses—functional and structural MRI

Functional magnetic resonance imaging (fMRI) and structural magnetic resonance imaging (sMRI) data were acquired on a 3T MRI scanner (MAGNETOM Trio, Siemens, Erlangen, Germany). Preprocessing and analyses were performed with SPM8 (http://www.fil.ion.ucl.ac.uk/spm/) running on MATLAB2013a (The MathWorks, Natick, MA). In brief, fMRI preprocessing included realignment, unwarping, co-registration and normalization (using DARTEL, Ashburner, 2007). sMRI, for voxel-based morphometry analyses, was processed using the VBM8 toolbox (www.http://dbm.neuro.uni-jena.de/vbm/). Multiple comparisons were controlled for by using family-wise error correction. Hippocampus and amygdala were chosen a priori as regions of interest due to their implication in context conditioning (Marschner et al., 2008; Lang et al., 2009; Alvarez et al., 2011) and used for small volume correction (SVC) based on anatomically defined masks [http://www.cma.mgh.harvard.edu/; probability threshold 0.7 (Desikan et al., 2006)]. For additional exploratory whole-brain analyses, an uncorrected (uc) threshold of P < 0.001 and a cluster size of k ≥ 15 were used. See Supplementary Data for details.

Results

Study 1: NOS1 ex1f-VNTR is associated with anxiety-related traits

For anxiety-related traits, a S-allele recessive model yielded a significant impact of NOS1 ex1f-VNTR [F(6,939) = 2.29, P = 0.033, η²= 0.014] with significant contributions of worrying and trait anxiety (Figure 2) and trend-wise effects on anxiety sensitivity and social anxiety (see Supplementary Table S1 and Supplementary Data and Supplementary Data for details). In contrast, a dominant model did not yield a significant impact of NOS1 ex1f-VNTR genotype [F(6,939) =1.691, P = 0.120, η²= 0.011, for an illustration of all three genotypes see Supplementary Figure S3].

Fig. 2.

Self-reported worrying (PSWQ) and trait anxiety (STAI-T) for S-homozygotes (S/S) and L-carriers (S/L, L/L). Asterisks indicate statistical significance with **P < 0.01. Error bars represent SEM.

In addition, a trans-diagnostic impact (Insel et al., 2010) of NOS1 ex1f-VNTR beyond anxiety-related traits was shown by higher depression scores in a recessive model [F(1,946) = 6.41, P = 0.012, η²= 0.007, Supplementary Table S2 and Supplementary Data] but not in a dominant model [F(1,946) = 0.580, P = 0.446, η²= 0.001, for an illustration of all three genotypes see Supplementary Figure S5] even though high correlations between the measures have to be noted (ADS-K/STAI-T: r = 0.653, P < 0.001; ADS-K/PSWQ: r = 0.475, P < 0.001).

In sum, results from study 1 indicate a recessive effect of the NOS1 ex1f-VNTR S-allele (see also Supplementary Table S1 and Supplementary Data and Supplementary Data) on anxious apprehension and depression (see also Supplementary Table S2 and Supplementary Data) while previous work has provided evidence for S-allele dominance.

Studies 2 and 3: NOS1 ex1f-VNTR effects on anxious apprehension and contextual fear conditioning

We further explored both a recessive and a dominant model in two data sets in healthy humans for an association between NOS1 ex1f-VNTR and anxious apprehension as well as contextual fear conditioning in particular as suggested from the rodent literature (Kelley et al., 2009; Zhang et al., 2010). In line with the hypothesis of an association with anxious responding, S-allele carriers showed significantly increased subjective anxiety ratings [study 2: main effect of genotype: F(1,88) = 5.10, P = 0.026, η²= 0.055, Figure 3A, Supplementary Table S3] and significantly decreased valence ratings [study 3: main effect of genotype: F(1,71) = 4.49, P = 0.038, η²= 0.060, Figure 3B, Supplementary Table S4] for all contexts, in absence of a genotype*context interaction (see Supplementary Tables S3 and Supplementary Data). Together, this suggests generally heightened subjective anxiety and aversion during context conditioning for S-carriers in spite of no evidence for a modulatory role on discriminating dangerous and safe contexts on a subjective level.

Fig. 3.

Subjective anxiety and valence ratings and fear-potentiated startle (FPS) in behavioral studies (study 2: A,C; study 3: B,D) for S-carriers and L-homozygotes during context conditioning. UCXT: unpredictable context, PCXT: predictable context, SCXT: safe context, ITI: inter-trial-interval; data points in (B) and (D) represent mean values of blocks. Error bars represent SEM. Gray areas included data corresponding to the statistical analyses. All data-points are included for illustrative purposes.

On an autonomic level, a direct comparison of the late acquisition phase, which is reflective of the end-point of learning, revealed a main effect of genotype in FPS in study 2, [F(1,64) = 4.43, P = 0.039, η²= 0.065], mirroring results for subjective ratings, in absence of a genotype*context interaction (F < 1). In study 3, however, a significant interaction of genotype*context [F(2,142) = 3.81, P = 0.024, η²= 0.051], as indicated by significantly higher FPS during the UCXT as compared to the SCXT and the ITI in S-carriers (Figure 3D, Supplementary Table S4) was observed in absence of a main effect of genotype (F < 1). Together, FPS results from studies 2 and 3 support an effect of NOS1 ex1f-VNTR on subjective anxious apprehension as observed in study 1 but provide only suggestive evidence for a direct modulatory role on context-conditioning processes. As in most behavioral and imaging genetic studies, however, phasic (study 2) and tonic (study 3) SCRs did not show any genotype-dependent main or interaction effects (Supplementary Tables S3 and Supplementary Data).

Under the assumption of a recessive model, no significant effects on anxiety ratings or autonomic measures in both studies were observed and also no significant effects on valence ratings in study 2 (all F’s < 1; P values > 0.3, see Supplementary Figure S6 for an illustration of all three genotype groups). Only for study 3, significant context*genotype interactions were observed for expectancy [F(1,71) = 4.94, P = 0.029, η²= 0.029] and arousal ratings [F(1,71) = 6.10, P = 0.016, η²= 0.079] as well as significant genotype differences in expectancy ratings [F(1,71) = 4.49, P = 0.038, η²= 0.059].

Study 4: NOS1 ex1f-VNTR effects on context conditioning-related brain activation

We further explored an impact of the NOS1 ex1f-VNTR neural activation pattern during contextual fear conditioning, which has been suggested to be more tightly linked to the effects of genetics than behavioral measures (Meyer-Lindenberg and Weinberger, 2006). For this study, we selectively invited participants based on their NOS1 ex1f-VNTR genotype (S-carriers vs L-homozygotes; S-allele dominant model) to extend these behavioral results using functional neuroimaging (study 4). Results demonstrate successful contextual fear conditioning in anxiety ratings and SCRs in the absence of group differences (see Supplementary Table S5). In the critical context conditioning contrast (UCXT > SCXT), S-carriers showed as expected higher activation in the right amygdala (P_SVC = 0.018) and the anterior part of the right hippocampus (P_SVC = 0.032, see Figure 4A and B) supporting suggestive evidence from study 3 for an impact of NOS1 ex1f-VNTR on context conditioning. Of note, exploratory grouping based on all three genotype groups (despite the a priori genotyping procedure) suggests an allele-dose effect on extracted parameter estimates from the hippocampus activation cluster (see Supplementary Figure S7). Exploratory whole-brain analyses (P < 0.001uc) additionally revealed higher activation in S-carriers in other regions ascribed to the fear network such as dorsomedial prefrontal cortex (dmPFC), orbitofrontal cortex, anterior and subgenual cingulate cortex as well as ventral striatum (see Supplementary Table S6), while no areas showed increased activations for L-homozygotes. Of note, significance of these results did not change when using a different preprocessing pipeline (i.e. including realignment parameters in the first level model instead of employing unwarp to account for task-related motion, see Supplementary Table S7 and Supplementary Data) or when excluding participants that showed movement > 1 voxel size (see Supplementary Data for details).

Fig. 4.

(A) Genotype-dependent activation differences (study 4) at the junction of the amygdala (x,y,z: 26,−2,−26, T = 3.49) and the anterior hippocampus (x,y,z: 30,−8,−24; T = 3.51) for S-carriers as compared to L-homozygotes during context conditioning (UCXT > SCXT) and (B) corresponding signal changes derived from the anterior hippocampus peak coordinates. Display threshold at P < 0.001uc. Error bars represent SEM.

Study 5: NOS1 ex1f-VNTR effects on gray matter volumes

No significant group differences were observed in the amygdala and hippocampus regions of interest at P < 0.05 FWE-corrected or at a more lenient exploratory threshold (P < 0.001uc) in a subsample of study 1 (N = 352, study 5). Exploratory whole-brain analyses at this threshold, however, revealed larger gray matter volumes in S-carriers in the right parahippocampal/fusiform area (Supplementary Figure S9), right dmPFC, left dlPFC, right precentral gyrus and bilateral occipital cortex (Supplementary Table S8), while no area displayed increased volume for L-homozygotes. Thus, significant differences in BOLD activity between S-carriers and L-homozygotes in the amygdala/hippocampus region as well as differences at a more lenient threshold in other areas ascribed to the fear network (with the exception of the dmPFC) are unlikely to arise from volume differences.

Discussion

In a translational approach, we provide converging evidence for a robust impact of NO signaling on anxious apprehension as well as preliminary evidence for a modulatory role of functional genetic variation in NOS signaling on contextual fear conditioning in humans. In our study, decreased NOS1ex1f expression, as inferred by the presence of two ex1f-VNTR short alleles, was associated with established vulnerability factors for pathological anxiety such as trait anxiety, trait worrying and depression (Merikangas et al., 2002; Verstraeten et al., 2011) as well as with generally enhanced anxious apprehension during (contextual) fear conditioning. These results complement previously reported associations of the ex1f-VNTR S-allele with state anxiety (Kurrikoff et al., 2012), lower conscientiousness and Cluster B personality disorder (Reif et al., 2009) as well as findings of a gene × environment interaction in depression (Kurrikoff et al., 2012) and associations of NO_x blood and CSF levels with depression severity (Freudenberg et al., 2015). Moreover, S-carriers showed significantly higher activation to US-predicting (UCXT) compared to safe (SCXT) contexts at the amygdala/anterior hippocampus junction in absence of morphometric differences in these areas. In addition, on an exploratory level, other areas implicated in conditioned fear and its regulation (such as the orbitofrontal cortex, subgenual ACC, dmPFC, ventral striatum, Sehlmeyer et al., 2009) displayed stronger activation in S-carriers, again in absence of volumetric differences in these regions (with an exception of the dmPFC). These results were supported by stronger FPS discrimination in S-carriers between safe and dangerous context in late acquisition, while L-homozygotes were characterized by less sustained contextual fear responses (i.e. FPS) and correspondingly reduced amygdala/anterior hippocampal activation. Together, these series of experiments provide first, albeit limited, evidence for behavioral and neurobiological differences in (contextual) threat processing in individuals differing in NO signaling in spite of robust differences in anxious apprehension. Of note, genotype group differences in amygdala/hippocampal activation were unaffected by the employment of a different prepossessing pipeline (i.e. unwarp or inclusion of realignment parameters), which highlights the robustness of our findings—in particular in light of their rather exploratory nature.

The implication of the NO system in anxious apprehension as well as depression is well in line with rodent work using pharmacological or genetic manipulations. The direction of these findings, however, seems to be in the opposite direction. While our data in humans link reduced NOS1ex1f gene expression to enhanced anxiety and depression, rodent studies have revealed anxiolytic and anti-depressive effects of pharmacological NOS-I signaling inhibition (Volke et al., 2003). Moreover, our work provides first evidence from humans for sustained contextual fear conditioning in autonomic (FPS) and neural markers (amygdala/hippocampal activation), while attenuated context conditioning was observed in Nos1-knockout mice (Kelley et al., 2009), again suggesting findings in the opposite direction in humans and rodents. A closer look at recent molecular data, however, might resolve these discrepancies. While we and others could convincingly link short alleles of NOS1 ex1f-VNTR to reduced gene expression in reporter gene assays in earlier studies (Reif et al., 2006, 2009; Rife et al., 2009), no direct evidence from human post-mortem data exist. Recently, however, we could establish that short alleles of this VNTR are associated with increased expression of total NOS1 mRNA in human amygdala tissue, likely due to compensatory up-regulation of other NOS1 isoforms (Weber et al., 2015). Because no such data are available from the hippocampus to date, the overall (possibly region specific) expressional consequences of NOS1 ex1f-VNTR require more in-depth studies on homologous cell systems and, ultimately, human tissue. Consequently, it is tempting to speculate that enhanced anxious apprehension and more pronounced context conditioning in humans as observed here may in fact result from increased NOS1 expression in the amygdala/hippocampal complex, complementing rodent work.

While this represents a compelling, albeit speculative, explanation of the present findings, it has to be acknowledged that acute systemic (i.e. pharmacological) or dramatic genetic (i.e. knockout) manipulations often employed in rodent work cannot be directly compared to the consequences of naturally occurring functional polymorphisms in humans because these may (partly) be attributed to developmental and adaptive processes. In fact such divergences have also been reported for other neuro-modulatory systems. For instance, serotonin-reuptake inhibitors (SSRIs) as the first-line treatment of anxiety and depression exert anxiolytic effects, while a variant of the gene targeted by SSRIs, the serotonin transporter (5-HTT)-linked polymorphic region (5-HTTLPR) resulting in reduced 5-HTT expression is associated with enhanced rather than attenuated anxiety (Sen et al., 2004; Schinka et al., 2004; Munafò et al., 2009). While pharmacological challenges may allow for more clear-cut interpretations, subtle inter-individual differences in signaling based on naturally occurring genetic variants may, however, be more ecologically valid. Yet, this candidate gene approach has lately fallen out of favor in behavioral and psychiatric genetics primarily due to failure to replicate initially impressive findings. While this approach admittedly suffers from several limitations, it still holds great potential to generate valid, replicable and valuable insights if certain pre-requisites are met (Lonsdorf and Baas, 2015). With respect to the work presented here, the experimental design as well as phenotype and genotype selection can be considered optimal for such an endeavor. In particular, a heritable phenotype which can be easily and reliably measured was linked to a genotype of compelling molecular impact in a biologically plausible system employing a series of multi-methodological studies that included replications in independent samples. Such an elaborate large-scale approach, however, is often only feasible within the context of collaborative research efforts which have become increasingly popular recently (Reif et al., 2014; Straube et al., 2014).

While our results provide evidence for a robust impact of NOS1 ex1f-VNTR on anxious apprehension as well as somewhat weaker evidence for a modulatory role on (context) conditioning processes, several limitations of our work should be discussed. First, while a recessive effect of the NOS1 ex1f-VNTR S-allele was observed on anxiety- and depression-related traits, a dominant effect was found with respect to behavioral studies of experimental context conditioning; similar observations are reported in previous studies where allelic, recessive as well as dominant effects were found (Reif et al., 2009, 2011; Hoogman et al., 2011; Kurrikoff et al., 2012; Weber et al., 2015). Guided by the findings from studies 2 and 3, the subsequent fMRI study applied a prospective genotyping approach (i.e. pre-selection of an equal number of S-carriers and L-homozygotes from a pool of genotyped participants) and an interpretation of the neural activation pattern is therefore limited to this a priori model, in particular in light of divergent dominance effects in study 1 and study 2/3. From our study and previous studies, we thus cannot conclude on a clear genetic model for NOS1 ex1f-VNTR, which might even be influenced by methylation or imprinting effects. Such divergences on the genetic model, however, are not uncommon, as similar findings of differential dominance effects have also been reported for other risk genes (e.g. 5-HTTLPR, Uher and McGuffin, 2008) and might also simply reflect threshold effects based on proximity of the phenotype (autonomic vs personality measures) to the biological pathway affected by the genotype. Importantly, however, in our study and other studies, in each and every case the short allele was associated with a less favorable outcome.

Second, although SCRs have been shown to constitute a reliable measure for addressing the heritability of conditioned fear in twin studies (Merrill et al., 1999; Hettema et al., 2003), no genotype-dependent effect on SCRs was observed in our study despite of differences in FPS and pronounced genotype-dependent differences in subjective ratings and neural activation. Of note, others have not observed genotype-dependent effects on SCRs either (Lonsdorf and Kalisch, 2011; Klumpers et al., 2012; Klucken et al., 2013; Lonsdorf et al., 2014a), which might reflect insufficient sensitivity of SCR to subtle inter-individual differences. Therefore, parallel FPS-fMRI acquisition would have hold great potential for an integrative interpretation of behavioral and neural effects within our fMRI study. However, technically challenging parallel FPS-fMRI acquisition has only recently become feasible (Lindner et al., 2015) and was not yet included in this study.

Finally, while we have included exploratory analyses to investigate a trans-diagnostic impact of this genetic variant (i.e. on depression, albeit highly correlated with anxiety) our analyses were restricted to the affective domain only. Given the ubiquitous actions and pleiotropic functions of NO on inter- and intracellular processing (Freudenberg et al., 2015) as well as its close interaction with other neurotransmitter systems (e.g. serotonin and dopamine, Freudenberg et al., 2015), it is, however, highly unlikely that the impact of NOS1 ex1f-VNTR genetic variation is specific to affective phenotypes. Indeed, in addition to depression and anxiety, NOS-I signaling has already been linked to schizophrenia and cognition as well as impulsivity and ADHD (Reif et al., 2006, 2009; Kopf et al., 2011; Freudenberg et al., 2015). Consequently, whether altered NO functioning based on naturally occurring inter-individual genetic markup (e.g. NOS1 ex1f-VNTR) represents a specific vulnerability factor for a limited set of psychopathological conditions or a rather general trans-diagnostic marker (Insel et al., 2010; Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013; Doherty and Owen, 2014) for affective psychopathology needs to be addressed in future studies.

Taken together, by means of a multimodal and multi-methodological approach, investigating a biologically plausible candidate gene, we have comprehensively shown an impact of NO signaling on anxious apprehension in a series of studies allowing for independent replication. With respect to an impact on (contextual) fear conditioning, our results provide only weak support for a modulatory role on the behavioral level with stronger evidence from functional neuroimaging. Having established the relevance of the NO system for anxiety-related processes in humans, future studies are warranted to elucidate the availability and feasibility of biological indicators of NO functioning (e.g. genetic) as clinically relevant biomarkers for precision medicine in psychiatry.

Funding

This work was supported by the Deutsche Forschungsgemeinschaft (DFG grants: SFB TRR58 B01, B06 and Z02; KA1623/3-1; KA1623/4-1; RE1632/5-1).