Who Gains? Genetic and Neurophysiological Correlates of BMI Gain Upon College Entry in Women

Abstract

The present investigation examined P3 event-related electroencephalographic potentials and a short and selected list of addiction-related candidate gene single nucleotide polymorphisms (SNPs) within 84 female students, aged 18–20 yrs. The students were assigned to groups defined by the presence versus absence of a positive body mass index (BMI) change from the pre-college physical exam to the current day. Analyses revealed significantly greater P3 latencies and reduced P3 amplitudes during a response inhibition task among students who exhibited a BMI gain. BMI gain was also significantly associated with a ANKK1 SNP previously implicated in substance dependence risk. In logistic regression analyses, P3 latencies at the frontal electrode and this ANKK1 genotype correctly classified 71.1% of the students into the BMI groups. The present findings suggest that heritable indicators of impaired response inhibition can differentiate students who may be on a path toward an overweight or obese body mass.

INTRODUCTION

The “Freshman 15” refers to a legend in popular culture about the magnitude of weight gain to be expected during the freshman year of college. Of course, legends are often inaccurate anecdotes and rarely generalize. In fact, many investigations of the legend have found that 15 pounds overestimates the mean and underestimates variability around it (Vella-Zarb and Elgar, 2009). Yet, it does appear that college freshmen gain more weight than the national average for U.S. citizens of this age group (Holm-Denoma et al., 2008).

In an attempt to better estimate the actual change in body weight during the early college years, many investigators have simply measured it over time and reported the change. Other investigators have conducted more elaborate analyses identifying different trajectories of weight gain associated with sex (Cain et al., 2008), race/ethnicity (Webb, 2012), type of residence (Kapinos and Yakusheva, 2011), peer groups (Yakusheva et al., 2011), and other demographic or environmental variables. Notable in its absence from the literature has been a careful consideration of the role of dispositional factors, such as genetics, personality, and cognitive control, in moderating or amplifying weight gain. Indeed, among the most interesting recent discoveries in behavior genetics has been the demonstration that the environment can be shaped by genetic factors—for example, the choice of peers is powerfully influenced by additive genetic effects (Kendler et al., 2007). The results of other recent behavior genetic studies (Dick et al., 2006) suggest that the genetic contribution to behavior dyscontrol becomes more apparent during late adolescence when the subject is becoming independent of the original home environment.

A major goal of the present study was to assess the contribution of specific candidate genes to weight change upon college entry. To meet this goal, we could have examined candidates chosen from the literature on obesity. In that literature, the best candidate (Fawcett and Barroso, 2010) is the FTO (fat mass and obesity associated) gene which may promote risk by a mechanism that involves epigenetic processes (Jia et al., 2011) or the regulation of distant genes, such as IRX3 (Smemo et al., 2014). However, our interest was not in obesity per se but in weight gain. Also, our interest was not in the area of genetic contributors to endocrine and hormonal dysfunction but to aberrant cognitive processes that may promote weight gain by other means—for example, by disrupting cognitive control (Kamijo et al., 2012a; Loeber et al., 2013; Pauli-Pott et al., 2010; Wirt et al., 2014) and monitoring (Cournot et al., 2006; Gunstad et al., 2007) processes potentially relevant to food choice, eating frequency, and portion size estimation. Accordingly, we focused on a different set of candidates.

The three genes of interest here—ANKK1, GABRA2, and CHRM2—were chosen because all have been previously associated with elevated body mass index (Bauer et al., 2012; Chan et al., 2014; Davis et al., 2009; Laramie et al., 2009) regardless of the presence/absence of obesity. In addition, and importantly, these genes have all also been associated with behavioral, cognitive, or neural differences potentially relevant to behavioral dyscontrol. For example, GABRA2 polymorphisms have been associated with elevated impulsiveness (Bauer et al., 2012; Villafuerte et al., 2013), increased stimulation or “high” after consuming alcohol (Arias et al., 2014), an increased number of rule breaking (Trucco et al., 2014) and Conduct Disorder behaviors (Dick et al., 2006; Dick et al., 2009), and excessive high frequency electroencephalographic activity (Edenberg et al., 2004). CHRM2 has likewise been associated with risk-taking and externalizing features (Dick et al., 2008). In addition, it has been associated with depression (Wang et al., 2004), dampened electroencephalographic P300 or P300-like orienting responses to novel stimuli (Jones et al., 2006; Jones et al., 2004), and abnormal maturation of these P300 responses during adolescence (Hill et al., 2013). Finally, the Taq1a polymorphism of ANKK1 has been associated with novelty seeking and harm avoidance (Antolin et al., 2009), prolonged P300 latencies (Noble et al., 1994), abnormal P300 maturation (Berman et al., 2006), and impaired motor control (Wiener et al., 2011). We hypothesized that GABRA2, CHRM2, and ANKK1 polymorphisms would similarly predict college weight gain.

A second major goal was to examine the association of weight change with self-rated impulsivity and, in parallel, with an objective indicator of selective attention (Polich, 2007; Tascilar et al., 2011) and response inhibition (Bauer et al., 2010b; Ruchsow et al., 2008; Smith et al., 2013). The objective indicator was a highly reliable (Brunner et al., 2013), positive-going electroencephalographic response elicited by a stimulus requiring inhibition of a prepotent manual response—a so-called no-go P3 (Kamarajan et al., 2005; Smith et al., 2013). It was evaluated during a simple continuous performance test that has repeatedly been used to study disorders characterized by impulsivity and/or inattention, including Attention Deficit Disorder (Baehne et al., 2009; Woltering et al., 2013) and Bipolar Disorder (Chun et al., 2013; Wright et al., 2014), as well as childhood obesity (Kamijo et al., 2012b). It was hypothesized that the frontally-generated, no-go P3 response elicited by the task would be delayed in latency and reduced in amplitude among students who exhibited an increase in body weight. The amplitude and latency of the preceding negative peak was also analyzed. This analysis provided a test of the contribution of an underlying N2 potential that could confound the ensuing P3.

A minor goal was to improve upon the study design of Holm-Denoma and colleagues (Holm-Denoma et al., 2008) who used the later high school years, versus the initial months of college, as the basis for measuring change. The improvement implemented here was requesting height and weight data during the high school period from the students’ primary care physicians. Unlike Holm-Denoma and colleagues, we did not rely on self-reported weight and height, which have questionable validity in college students (Larsen et al., 2008; Wen and Kowaleski-Jones, 2012).

METHODS

Participants and Procedures

Over a 1-year period, one-hundred-and-four female students, aged 18–20 years, were recruited from area colleges and universities. Only freshmen and sophomores were included. They were contacted through a variety of methods, including newspaper, radio, and poster advertisements. Interested parties were asked to call the study office for information and eligibility screening. Those who appeared eligible were invited to visit the Health Center on a subsequent day for further screening and evaluations. At the time of this in-person visit, the subject reviewed and signed an informed consent agreement approved by the Health Center’s Institutional Review Board.

During the visit, height and weight were measured and converted to body mass index. Also, a more precise measure of adiposity, triceps skinfold thickness, was measured with calipers (Lange calipers, QuickMedical, Issaquah, WA) using a standard protocol. The students were then asked to complete questionnaires assessing alcohol [AUDIT; (Saunders and Lee, 2000)] and drug [DAST; (McCabe et al., 2006)] use. They provided medical history information via self-report and signed a medical records release to request historical height and weight data from their primary care physicians. The Eating Disorder Diagnostic Scale [EDDS; (Stice et al., 2004)] was used to collect information about eating disorder symptoms and diagnoses. In addition, self-ratings of impulsivity in the attention, motor, and non-planning categories were obtained from the BIS-11 (Patton et al., 1995) scale. A saliva sample was collected for DNA analysis.

Students who reported medical issues that would complicate body weight or evoked electroencephalographic responses were excluded from the analysis. More specifically, the exclusions were past year pregnancy, psychosis, substance dependence, HIV, thyroid disease, cardiovascular disease, hypercholesterolemia, uncorrected visual deficits, seizure disorder, heart disease, or head injury. To ensure consistency of outside influence, students were required to have full-time status, live on campus, and participate in the school’s food service meal plan. Ineligible volunteers (n=6) were paid $30 for time and effort and dismissed. The students who completed the evaluation were paid $150 each.

To assess N2 and P3 responses during the continuous performance task, we applied electrodes in a high density, 64-channel montage using an electrode cap with a linked-ear reference. Inter-electrode impedances were kept below 10 Kohms. Electrodes placed above and below the right eye were used for eye movement and eyeblink detection. All signals were amplified and processed with a Synamps² system (Compumedics Neuroscan, Charlotte, NC) running SCAN 4.4 software.

The continuous performance task (Nichols and Waschbusch, 2004) was comprised of 50 presentations of a rare stimulus requiring response inhibition (non-target), “0”, intermingled with 200 presentations of a frequent target stimulus, “1”, of the same 200 ms duration. The stimuli subtended a visual angle of 2.86 degrees. They were presented in a white font every 1.3 seconds.

Students were instructed to press a response key with the right hand immediately upon seeing the “1” stimulus. The electroencephalogram was digitized at a rate of 500 Hz, filtered with a bandpass of 0.01–10 Hz (24 dB/octave roll-off), and sampled over a period of 100 ms preceding to 700 ms following this event.

Data Reduction and Analysis

P300 and Task Performance Data

Using software resident to the Synamp² system, evoked response waveforms were created from eye movement- and baseline-corrected epochs [see (Bauer et al., 2010b) for a description of data editing methods] with no voltage deviations greater than 50 microvolts. Time-point averaged responses were formed from the same number (n=25) of occurrences of the rare non-target stimulus across students. The amplitudes and latencies of P3 peaks were measured between 225–660 ms post-stimulus onset (Figure 1). To estimate P3 activity over the likely frontal generator, as well as over motor and parietal areas, we measured and analyzed P3 at Fz, Cz, and Pz electrode sites. N2 peaks were identified within a latency range of 150–220 ms and similarly measured. The number of trials with false alarm responses to the rare non-target and the number of correct responses as well as the average reaction time following the frequent target were also calculated.

Figure 1.

Group-averaged event related potential waveforms sorted by group. The 100 ms pre-stimulus onset baseline is indicated by the horizontal time marker. P3 is the prominent upward deflection in the middle of the waveform.

DNA data

Genomic DNA was purified in batch from frozen saliva samples by our Clinical Research Center’s core lab. DNA samples were placed in 96-well plates and genotyped using PCR based TaqMan 5′-nuclease allelic discrimination assay methods. To minimize population stratification artifact which can complicate the interpretation of allele frequencies, we adopted the typical approach and limited the analysis to subjects of European-American origin. SNPs were assayed within CHRM2 (rs12673281 and rs324650) and GABRA2 (rs279871).

In addition, the Taq1A polymorphism (rs1800497) within the ANKK1 region was assayed. The reader should note that Taq1A was once believed to be a SNP with the D2 Dopamine Receptor gene, DRD2. However, fine mapping has shown that this belief was incorrect (Neville et al., 2004). Taq1A is instead located 10kb downstream from DRD2 in an adjacent gene, ankyrin repeat and kinase domain containing 1 (ANKK1). We also assayed a second SNP (rs17115439) within ANKK1 that is further away from the DRD2 gene than Taq1A and has been shown to be more reliably associated than Taq1A with behavior dyscontrol (Dick et al., 2007).

Before the analysis, the number of genotypes at each of these 5 SNPs was collapsed from 3 to 2 levels (major allele homozygotes and minor allele carriers). Collapsing minor allele homozygotes and heterozygotes into a single group is a common strategy for maximizing power and minimizing the adverse statistical impact of a small number of minor allele homozygotes (i.e., a low cell frequency).

Analysis Plan

Adequate data were provided by the primary care physicians of 84 of the 104 students. These 84 students were included in the analysis. To compute the change in body mass index (BMI) from the pre-college to the college years, we converted all available heights and weights to BMIs. The pre-college BMI most proximal in time to college entry (between 14 and 18 years of age) was then subtracted from the current BMI measured in the laboratory. This quantity was divided by the number of intervening months [mean interval (sd)=34.9(8.8) months] to correct for variability in the interval. A positive change/time ratio assigned a student to the positive BMI gain, BMI ↑, group (n=57). A negative ratio or a zero result placed a student in the other, BMI ↓→, group (n=27). Differences between these groups in background characteristics, genetics, N2 and P3 event related potentials, and task performance were analyzed. In all of the ANOVAs, the number of months intervening between the pre-college and college BMI assessment was entered as a covariate.

Some readers may argue with our decision to apply a cut-point and assign subjects to groups when BMI change could have been analyzed as a continuous variable. We argue that there is merit in both approaches. Yet, in public health generally and obesity risk prevention specifically, the important questions about intervention must be grounded in simple yes/no decisions that rely on explicit or implicit cutpoints. In preventing or minimizing college weight gain, the important question is “who is at risk?”. For this reason, we think that there is greater value in focusing on BMI change as a dichotomous variable although its underlying distribution is obviously continuous. To address those who disagree with this argument, we conducted additional analyses on the P3 data in which BMI change per month was entered as a continuous criterion variable, with the time interval between assessments as a covariate.

RESULTS

Background Characteristics

Table 1 shows group means and standard deviations for age, prior BMI, current BMI, interval duration, current triceps skinfold thickness (average of right and left), alcohol and drug problems from the AUDIT and DAST, binge eating severity score and eating disorder frequencies from the EDDS, and BIS-11 impulsivity scale scores. ANOVAs and χ² tests comparing the groups revealed no significant differences on most variables. The only variable differentiating the groups was current triceps skinfold thickness [F(1,82)=4.6, p<0.05] which was significantly greater in the BMI ↑ group.

Table 1.

Background characteristics sorted by group.

	BMI ↓→ N=27	BMI ↑ N=57
Age in yrs (SD)	19.7(1.19)	19.3(1.21)
Previous Body Mass Index (kg/m2)	24.7(4.42)	22.9(5.37)
Current Body Mass Index (kg/m2)	23.3(3.16)	24.9(5.90)
Months between BMI assessments	33.9(11.1)	35.4(7.7)
Current Triceps Skinfold Thickness (mm)	13.1(2.72)	15.2(3.69)*
BIS-11 Attentional Impulsiveness	16.0(3.65)	16.9(4.17)
BIS-11 Motor Impulsiveness	22.0(4.22)	23.1(4.31)
BIS-11 Non-planning Impulsiveness	23.3(4.47)	23.6(4.90)
DAST-10 Drug Problems	1.3(1.54)	1.4(1.72)
AUDIT Alcohol Problems	7.7(5.62)	7.6(4.97)
EDDS Binge Eating Severity Score	1.7(1.98)	1.5(1.90)
EDDS Binge Eating Disorder (frequency/%)	1/3.7	9/15.7
EDDS Bulimia Nervosa (frequency/%)	3/11.1	5/8.7

^*p<0.05.

Genetics

The frequencies of the 5 genotypes were all consistent with Hardy-Weinberg expectations: ANKK1 SNP at rs17115439 (χ²=0.71, p=n.s.); ANKK1 at rs1800497 (χ²=1.76, p=n.s.); CHRM2 at rs12673281 (χ²=0.01, p=n.s.); CHRM2 at rs324650 (χ²=0.24, p=n.s.); GABRA2 at rs279871 (χ²=1.61, p=n.s.). Simple χ² tests were used to examine the associations between the five genotypes and BMI change. Only the distal ANKK1 SNP was related to group membership (χ²=5.94, df=1, p=0.01) at a Bonferroni-adjusted significance criterion. In the BMI ↓→ group, the genotype frequencies were 15 (major allele homozygote: cc) and 12 (minor allele carrier: ct or tt). In the BMI ↑ group, the frequencies were 16 (cc) and 41 (ct or tt). The latter group demonstrated a greater prevalence of the rare allele.

N2 and P3 Event Related Potentials and Task Performance

ANOVAs evaluating the statistical effect of group (Table 2) revealed significantly more false alarm responses [F(1,82)=4.4, p<0.04] and greater P3 latencies at Fz [F(1,82)=5.7, p<0.02] and Cz [F(1,82)=4.5, p<0.04] electrodes following the rare non-target stimulus in the BMI ↑ group (Figure 1). Also, this group demonstrated relatively smaller P3 amplitudes at Fz [F(1,82)=3.8, p<0.05] and Cz [F(1,82)=3.6, p<0.05] electrodes. Other ANOVAs evaluating the statistical effect of group on the negative, N2 peak preceding P3 revealed no significant group differences in amplitude at Fz [F(1,82)=0.03, p=0.8] or Cz [F(1,82)=0.03, p=0.8]. The effect of group on N2 latency was likewise not significant [Fz: F(1,82)=1.37, p=0.2; Cz: F(1,82)=2.01, p=0.2].

Table 2.

Task performance and P3 amplitudes and latencies sorted by group.

	BMI ↓→	BMI ↑
False Alarms	8.8(4.42)	11.7(6.38)*
Hits	184(3.1)	182(6.4)
Reaction Time in ms	241(37.4)	246(40.5)
No-Go P3
Amplitude in μV
Fz	12.8(4.13)	10.9(4.60)*
Cz	15.8(4.96)	13.5(4.93)*
Pz	14.3(4.61)	12.5(4.67)
Latency in ms
Fz	347(25.3)	364(30.4)*
Cz	349(29.8)	371(49.1)*
Pz	364(42.9)	385(52.4)

^*p<0.05

Regression analyses predicting BMI change/month as a continuous variable, versus the dichotomous variable noted above, revealed similar results. For example, with no-go P3 latency at Fz and ANKK1 genotype as predictors, the respective standardized coefficients and corresponding statistics were β=0.23, t=2.15, p=0.03 and β=0.21, t=1.96, p=0.05. The regression equation with P3 latency at Cz was also significant: β=0.22, t=2.06, p=0.04. However, separate equations including either P3 latency at Pz or P3 amplitude at each of the three electrode sites revealed no significant results.

Classification Accuracy

Logistic regression was used to examine the conjoint ability of P3 latency at the frontal electrode (the electrode site most proximal to the generator) and the ANKK1 rs17115439 genotype collapsed from 3 to 2 levels (major allele homozygote versus minor allele carriers) to predict group membership. The regression equation was statistically significant [χ²=10.9, p=0.004]. The unstandardized beta weights and other statistics for P3 latency were B(SE)=0.02 (0.011), Wald=4.93, p=0.02, and OR=1.02. For the ANKK1 SNP, the statistics were B(SE)=1.19 (0.51), Wald=5.32, p=0.02, and OR=0.30. The overall classification accuracy was 71.1%.

Other Correlates of BMI Change and P3 Latency

A few additional analyses were performed to resolve outstanding issues. The first of these analyses attempted to discern if BMI was an acceptable measure of adiposity in the present sample. To this end, simple correlations of current BMI and BMI change were computed with a superior index of adiposity--triceps skinfold thickness. The correlations were statistically significant: r=0.82 (p<0.001) and r=0.28 (p=0.03), respectively.

The next analysis tested simple associations between P3 latency at Fz and a self-report measure of poor behavior control, i.e., BIS-11 score. These correlations were significant for all three subscales. A positive correlation was detected between P3 latency and Motor (r=0.35, p=0.001), Non-planning (r=0.36, p=0.001), and Attentional (r=0.21, p=0.04) Impulsiveness.

The final set of analyses attempted to locate the neuroanatomical source of the difference between groups in P3. With Curry Version 5.0 software (Compumedics Neuroscan, Charlotte, NC), we re-referenced the group averaged event related potential at each electrode site to the common averaged reference. Independent Components Analysis (Hyvarinen and Oja, 1997) was used to find the component explaining the majority of the variance within the latency range of interest (225–660 ms). This component explained 21.1% (SNR=5.3) of the variance in the BMI ↓→ group waveform and 21.7% (SNR=9.4) of the variance in the BMI ↑ group. It could be explained [>92% of variance; see (Scherg and Berg, 1991) for methods] by a single, fixed coherent dipole in both groups in the vicinity of the anterior cingulate cortex (Figure 2). For the BMI ↓→ group, the magnitude of the dipole centered at 0.3, 28.5, 81.4 mm was 50.9 μAmm. For the BMI ↑ group, a similar but more anterior dipole centered at −13.0, 53.7, 69.2 mm was only 7.9 μAmm in magnitude.

Figure 2.

Estimated location of the dipole generator of P3 in each group registered to a normal MRI.

DISCUSSION

The goal of the present study was to identify heritable characteristics associated with weight change upon college entry among women. The goal was met. Within a sample of 84 college women, we found that the distal ANKK1 genotype at rs17115439, as well as anterior scalp P3 latencies and amplitudes recorded during a response inhibition task, differentiated those who gained weight from those whose weights declined or remained stable. The variance in weight change collectively explained by ANKK1 genotype and P3 latency was not trivial (R²=0.19). These variables correctly assigned 71.4% of the participants to their groups.

Of course, we recognize that this study is limited. A sample size of 84 participants is less than ideal for a candidate gene association study. A replication with more participants varying in race/ethnicity is needed. Yet, we should not dismiss the present association simply because the sample is small and homogenous. Other studies have associated the ANKK1 gene with other indicators of behavior dyscontrol in large samples of n=1923 participants with/without alcohol dependence (Dick et al., 2007), n=580 participants varying in their frequency of alcohol consumption (Meyers et al., 2012), and n=1615 participants with/without a diagnosis of nicotine dependence (Gelernter et al., 2006). In this context, our discovery of an association between ANKK1 and weight gain is not surprising. Weight gain within these college women could be viewed as a heritable phenotype that is also related to impulsivity, poor cognitive control (i.e., reduced and delayed no-go P3), and the addictions. More information is needed from other studies before we can fully understand the mechanism by which this synonymous ANKK1 SNP may influence other genes and their protein products to alter brain function and impulse control and, in turn, eating habits and weight gain.

Apart from the small size and limited racial/ethnic diversity of the sample is a concern about the exclusive focus on women. We adopted this focus for two reasons. First, because the study was a pilot investigation with attendant limitations in scope and resources, we chose to restrict variability in sex to maximize power. The second reason for excluding men was suggested by the results of previous studies of the association of impulsivity and poor cognitive control with body mass. In those studies (Bauer et al., 2012), impulsivity is correlated with BMI more strongly in women than men.

The present findings are also limited in not indicating the mechanism that connects impulsivity and altered no-go P3 activity to weight gain in these college women. The high heritability and reliability of BIS-11 ratings (Congdon and Canli, 2008) and the no-go P3 (Albrecht et al., 2013; Brunner et al., 2013) argue that these measures are stable traits which are likely to precede and promote weight gain. But, we cannot rule out the opposite possibility in which weight gain is viewed as altering brain white matter (Haltia et al., 2007) and brain function (Bauer et al., 2010a) to effect higher levels of impulsivity and delayed no-go P3 latency.

The associations demonstrated here of more false alarm responses, reduced anterior scalp P3 amplitudes, and delayed anterior scalp P3 latencies with weight gain were hypothesized. Impaired response control and disrupted P3 responses during selective attention and response inhibition tasks are a common finding in the ADHD literature and in other syndromes characterized by poor behavior control and high levels of impulsivity, including childhood overweight/obesity (Bauer et al., 2010b; Tascilar et al., 2011). In a future study, it would be interesting to follow the present sample of college women over time and determine if the P3 and ANKK1 differences associated with weight gain during these early college years predict an overweight/obese body mass at a later time.

The present study made two other contributions to the literature that are worthy of mention. One of these contributions was using the later years of the high school period as the basis for measuring change as opposed to the more common practice wherein the first few months of college are the anchor. The latter choice is convenient. However, it omits the transition to college as well as adaptations which may emerge in preparation for the transition.

The other notable contribution of the present study was its reliance on measured as opposed to self-reported height and weight. Decades of research on obesity/overweight have indicated that participants, including college students (Larsen et al., 2008), will overestimate height and/or underestimate weight. Self-reports are often not valid (Wen and Kowaleski-Jones, 2012).

In sum, the findings of the present study demonstrate the potential of genetic markers and indicators of impulsivity and poor cognitive control for identifying a subgroup of students at-risk for the so-called “freshman 15”. The present findings also suggest a role for these markers and indicators in future studies that examine gene by environment interplay in promoting weight gain during major life transitions (e.g., entering and leaving college, marriage, retirement).

HIGHLIGHTS.

• Correlates of the presence/absence of college weight gain were examined.

• Various indicators of impulsivity and impaired cognitive control were assessed.

• Weight gain was associated with a substance-dependence-associated ANKK1 genotype.

• It was also associated with impulsivity and delays in an event-related EEG response.