Domain-specific cognitive impairments in humans and flies with reduced CYFIP1 dosage

Abstract

BACKGROUND:

Deletions encompassing a four-gene region on chromosome 15 (BP1-2 at 15q11.2), seen at a population frequency of 1 in 500, are associated with increased risk for schizophrenia, epilepsy, and other common neurodevelopmental disorders. However, little is known in terms of how these common deletions impact cognition.

METHODS:

Here we used a web-based tool to characterize cognitive function in a novel cohort of adult carriers and their non-carrier family members. Results from 31 carrier and 38 non-carrier parents from 40 families were compared to control data from 6,530 individuals who self-registered on the Lumosity platform and opted in to participate in research. We then examined aspects of sensory and cognitive function in flies harboring a mutation in Cyfip, the homolog of one of the genes within the deletion. For work in fly, ten or more groups of 50 individuals per genotype were included.

RESULTS:

Our human studies revealed profound deficits in grammatical reasoning, arithmetic reasoning and working memory in BP1-2 deletion carriers. No such deficits were observed in non-carrier spouses. Our fly studies revealed deficits in associative and non-associative learning despite intact sensory perception.

CONCLUSION:

Our results provide new insights into outcomes associated with BP1-2 deletions and call for a discussion on how to appropriately communicate these findings to “unaffected” carriers. Findings also highlight the utility of an online tool in characterizing cognitive function in a geographically distributed population.

INTRODUCTION

Deletions spanning a four gene region on chromosome 15 (BP1-2 at 15q11.2), present at a population frequency of approximately 1 in 500, are associated with increased risk for multiple neurodevelopmental disorders (NDDs) (1). A large case-control study showed deletion carriers to be at approximately three-fold increased risk for schizophrenia (2, 4, 5). Later work revealed an association between BP1-2 deletions and idiopathic generalized epilepsy, identifying a five-fold increased risk for carriers (6, 7). Separate work has pointed to the deletion as a susceptibility locus for developmental disability (8, 9). Consistent with this result, the rates of speech and motor delays were significantly increased in carrier vs. non-carriers (10). Lastly, case series summarizing presentation in carriers have suggested that the variant may be associated with additional outcomes including dysmorphic features, Autism, Attention Deficit Hyperactivity Disorder (ADHD), and Intellectual Disability (11-16).

Relatively little, however, has been done outside of clinically ascertained populations potentially obscuring the relationship between BP1-2 deletion status and disease risk. Some investigators have looked to outcomes in carrier parents (11-15, 17), and as summarized by Cafferkey and colleagues only 22 of 32 evaluated were reported to be “healthy” (10). Not captured there, however, was phenotype data needed for the interpretation of this observation. Also unaddressed is whether deletions are associated with subclinical features. Results from a separate study identified carriers through population-based genotyping and performed neuropsychological testing determined that deletion carriers struggled more with challenges of daily living than controls (3). Variation in regional grey matter volumes were also observed and separate work has extended this finding to white matter microstructure (3, 18). Stefansson and colleagues also found that deletion carriers reported difficulties with reading and math at higher than expected rates (3). However, results from these reading and math related questionnaires are subjective, and because they are based on historical impressions, they cannot be used in assessing the efficacy of possible interventions. Formal neurocognitive evaluations were also performed but did not identify differences between deletion carriers and non-carriers.

Separate to the identification of deficits resulting from deletions at BP1-2 are the development of appropriate models for interrogation of mechanisms, something that cannot be done through study of deletion carriers. Whereas four genes are present within the human BP1-2 region (TUBGCP5, CYFIP1, NIPA2, and NIPA1), mounting evidence points to the importance of the Cytoplasmic FMR1 Interacting Protein 1 (CYFIP1). Through physical interaction with RAC1, CYFIP1 activates the Wave Regulatory Complex (WRC) and promotes cytoskeletal remodeling, a key aspect of dendritic spine formation (19-21). CYFIP1 also binds to the fragile X mental retardation protein (FMRP), forming a translational inhibitory complex (22-25). Altered CYFIP1 dosage perturbs neuronal morphology (21, 26, 27), synaptic excitation (28, 29), brain patterning and function (30). Importantly, alterations in white matter architecture and functional connectivity, similar to those observed in humans with BP1-2 deletions, were observed in mice and rats heterozygous for a mutation in Cyfip1 (31, 32). Regarding disease, individuals with schizophrenia showed a reduction in CYFIP1 in prefrontal cortex relative to controls (33). Separately, CYFIP1 transcript levels are reportedly upregulated relative to controls in blood from individuals with autism and epilepsy (34, 35). Associations between regulatory variants at the CYFIP1 locus and both schizophrenia and autism have also been observed (36, 37). Lastly, we have shown in human neural progenitor cells that CYFIP1 knockdown resulted in dysregulation of schizophrenia and epilepsy gene networks (38).

Less is known about Tubulin Gamma Complex Associated Protein 5 (TUBGCP5), although its regulation of mitotic spindle formation is done in cooperation with GSK-3β, a protein implicated in schizophrenia (39-44). Separately, a recent case report describing a BP1-2 deletion carrier with a missense variant in TUBGCP5 suggested that this second variant might contribute to the primary microcephaly and mild developmental delay observed (45). NIPA1 and NIPA2 (Non-imprinted in Prader-Willi/Angelman syndrome 1 and 2), encoding proteins involved in magnesium transport, have also each been linked to neurological disease (46). Dominant negative mutations in NIPA1 cause hereditary spastic paraplegia and expanded polyalanine repeats within the protein and are associated with increased risk for amyotrophic lateral sclerosis, a disorder resulting from death of motor in neurons in the brain and spinal cord (47-49). Disruptive mutations in NIPA2 have been identified in childhood absence epilepsy and an association between a possible regulatory variant and schizophrenia reported (36, 50, 51). Towards mechanistic insights we characterized flies heterozygous for a null mutation in the CYFIP1 homolog (Cyfip^85.1/+).

METHODS AND MATERIALS

Human

Subject ascertainment and enrollment

Families in which a child had been identified as a BP1-2 deletion carrier following clinical evaluation were invited to participate in our study. Some families were referred to us from clinicians, whereas others contacted us directly through the “15q11.2 Advocacy, Research, and Support” FaceBook site (https://www.facebook.com/groups/1419874888238714/). Initial screening was based on a review of clinical aCGH results provided to us by subjects or their physicians. Subjects harboring structural variants spanning the adjacent BP2-3 region were excluded, but beyond this, neither secondary variants nor diagnosis were considered in terms of inclusion or exclusion. Our study, consent forms, and procedures for obtaining informed consent were approved by the Einstein Institutional Review Board. All subjects, or their parent in the case of minors, provided written informed consent prior to participation. No genotype or phenotype data generated in the course of our study was made available to participants. Control data corresponds to results from 6,530 anonymized individuals who self-registered on the Lumosity platform and opted in to participate in research. These individuals were selected on the basis of demographic information from individuals within our familial cohort as well as the absence of any self-reported clinical diagnosis. For each subject we enrolled, Lumosity provided us with de-identified results from one hundred or more individuals matched for age, sex, and education level. Like subjects we recruited, these individuals were tested remotely and without supervision.

Genotyping

A saliva or blood sample was obtained from each subject. The primary determinant here was the subject’s geographical proximity to a phlebotomy lab. Regardless, all phenotyping was done online. Saliva samples were collected in Oragene OG-500 kits (DNA Genotek) and blood samples in purple top EDTA-containing tubes (Becton, Dickinson and Company). DNA was extracted on a QuickGene 610 instrument (Autogen) using QuickGene SP kits (Kurabo) for saliva and Puregene Blood kits (Qiagen) for blood. BP1-2 copy number was determined for all subjects using a Taqman assay (Hs01476346_cn, Life Technologies) as described elsewhere (38). DNA samples from Lumosity controls were not available for determination of BP1-2 deletion status.

Phenotyping

Parents were asked whether a clinician had ever given them a diagnosis of any of ten NDDs (Table 1). We included subjects reporting Pervasive Developmental Disorder – Not Otherwise Specified within the Autism group. Individuals reporting Dyslexia, Language Delay, or another learning disability were said to have a Learning Disability. Each participant was also provided with a unique login and asked to complete the web-based NeuroCognitive Performance Test (NCPT) (52), at home during a single 20-30 minute session.

Table 1.

Learning Disabilities are present at a higher rate in adult BP1-2 deletion carriers than non-carrier adults from the same homes

Diagnosis	Deletion (n=31)	Non-Carrier (n=38)	P-value a
ADHD b	6 (19.4%)	7 (18.4%)	5.8×10−1
Autism c	1 (3.2%)	0 (0.0%)	4.5×10−1
Bipolar Disorder	2 (6.5%)	0 (0.0%)	2.0×10−1
Depression	6 (19.4%)	7 (18.4%)	5.8×10−1
Intellectual Disability	1 (3.2%)	0 (0.0%)	4.5×10−1
Learning Disabilities d	9 (29.0%)	3 (7.9%)	2.3×10−2
OCD e	2 (6.5%)	1 (2.6%)	4.2×10−1
Schizophrenia	0 (0.0%)	0 (0.0%)	1.0
Seizures or Epilepsy	2 (6.5%)	0 (0.0%)	2.0×10−1
Sensory Integration Disorder	1 (3.2%)	1 (2.6%)	7.0×10−1
Any of above f	14 (45.2%)	16 (42.1%)	5.0×10−1

^aP < 0.05 (uncorrected) are highlighted in bold

^bADHD: Attention Deficit Hyperactivity Disorder

^cAutism: Includes individuals reported to have a Pervasive Developmental Disorder – Not Otherwise Specified diagnosis)

^dIndividuals reporting a diagnosis of Dyslexia, Language Delay, or a non-specified Learning Disability were included here

^eOCD: Obsessive Compulsive Disorder

^fMultiple individuals reported more than one diagnosis, and so percentages do not sum to 100%

Statistical Analyses

A two-tailed Fisher’s exact test was used to evaluate the relationship between deletion status and gender. Frequencies of NDDs were compared between carriers and non-carriers using one-tailed Fisher’s exact tests. For these exploratory analyses, p-values < 0.05 were deemed nominally significant. For NCPT outcomes, raw scores for each subject we enrolled was compared to the mean of ≥100 anonymous controls matched for gender, age, and education.

After excluding values greater than 3 standard deviations from each group mean, one-tailed paired t-tests were used to identify deficits in deletion subjects and non-carrier. Based on self-reported difficulties with reading and mathematics among carriers (3), performance on grammatical reasoning and arithmetic reasoning tasks were evaluated under the a priori hypothesis that deletion carriers would show deficits (Table 1). We used Nyholt’s method to obtain a p-value threshold for these analyses that would ensure appropriate correction for multiple comparisons, but also take into account the correlation between outcomes (Effective test number = 2.73; p_{hypothesis.corrected} < 1.8×10⁻²) (52, 53). This same procedure was used to identify significant effects across the full set of measures available to us (Effective test number = 9.75; p_{all.corrected} < 5.1×10⁻³). Data visualization and statistics were done using R 3.2.0.

Drosophila

Drosophila stocks, rearing conditions, and behavioral analyses

Flies were cultured in vials containing a standard Drosophila medium at 25°0 with 60-80% humidity in a 12h light/dark cycle. Flies heterozygous for a mutation that eliminates two-thirds of the dCyfip-coding region (Cyfip^85.1/+) have been described previously (20). The fly line used as control was wild-type Canton-S w¹¹¹⁸ (iso1CJ). Flies were tested at 3–5 days of age and in all cases animals of each genotype were tested on the same day. All experiments were performed at 25°0 and 75%–80% rela tive humidity with the experimenter blind to genotype. T-tests were performed in GraphPad (Prism) to identify between group differences.

Sensory Perception

Olfactory and shock avoidance assays were performed as described previously to assess sensory function (54). Briefly, ten or more groups of 50 odor naive flies of each genotype were placed in a T-maze. After 90 seconds the number of flies in each arm was counted and an avoidance index calculated. The avoidance index corresponds to the number of flies found in the air-containing compartment minus the number of flies in the odor-containing compartment divided by the total number of flies. Shock avoidance was tested in a similar fashion with 90-volt stimuli 1.25 seconds in duration). In these experiments, air was passed through both arms at constant flow throughout the testing period.

Associative learning

We employed a previously described negative reinforcement paradigm to assay associative learning (54-56). During a 2-minute training phase, ten or more groups of 50-60 flies of each genotype were placed into the single upper arm of a T-maze and exposed to various stimuli. Animals were first exposed to six shocks (90 volts; 1-sec in duration every 5-sec for 30 seconds) and a novel shock-associated odor (CS+). In this same compartment, animals were then exposed to odor free air for 30 seconds, followed by a second unfamiliar odor for 30 seconds (without shock, CS−), and finally odor free air again for a final 30 seconds. Flies were transferred immediately after this training phase to the lower part of the T-maze for testing. A performance index (PI) was calculated as described before (54-56). Briefly, half of the performance index was obtained by calculating the fraction of flies that avoided the shock-associated odor over 90 seconds minus the number of flies avoiding the control odor, divided by the total number of flies. In parallel another population of the same genotype of flies was trained with the CS+ and CS− odors reversed. A final performance index is the average of the two values.

Habituation: electric shock

Habituation, a measure of experience-dependent attenuation of reactivity to a stimulus, was assayed as described previously (57, 58). In a training phase 50-60 flies were placed in the upper arm of a standard T-maze with an electrifiable grid, then exposed to 45-volt shocks over 75-90 seconds (15 shocks of 1.2-sec duration, 4-sec inter-stimulus interval). Air was not drawn through the tube during training to avoid association of the shocks with air. After a 30-sec rest period, animals were transferred over 120-sec to the lower part of the maze and allowed to move freely for 90-sec during which time stimuli were delivered as above to the electrified arm of the maze. At the end of the choice period, the flies in each arm were trapped and counted, and an avoidance index calculated as above.

Habituation: light-off jump reflex

128 male flies per genotype were subjected to 100 light-off stimuli (15-ms each; 1-sec inter-stimulus interval) and a response recorded if a jump occurred during or within 500-msec of the stimulus event. The proportion of responders within each group, or average jump response, was determined for each trial.

RESULTS

Marked cognitive deficits observed in adults harboring BP1-2 deletions

Given high rates of NDDs in clinically ascertained children harboring BP1-2 deletions (10-14), we sought to characterize the frequency of various conditions in carrier and non-carrier parents. Eligibility was based on having a child referred to geneticists and determined to have a BP1-2 deletion. DNA analysis in our laboratory identified 31 carrier and 38 non-carrier parents across 69 individuals from 40 families. Consistent with published work, deletions were largely inherited (10, 16) with only 2 de novo events observed across the 33 families in which inheritance could be resolved. The frequency of deletions was not significantly different between mothers and fathers. As detailed in Table 1, analyses showed that NDDs were reported at similar frequencies in carriers and non-carriers (14/31 vs. 17/38, p=0.58), although learning disabilities were more than three times as common in carriers (9/31 vs. 3/38, p=2.3×10⁻²).

To permit standardized and objective quantification of cognitive performance in our global cohort, we had subjects complete the web-based NeuroCognitive Performance Test (NCPT) (52). Composed of classic neuropsychological instruments adapted for computerized administration, the NCPT permits rapid assessment of grammatical reasoning, arithmetic reasoning, psychomotor speed, attention, working memory, non-verbal problem solving, response inhibition, and visual search (Table S1). Good concordance between results from the NCPT and analogous pencil-paper tests is observed (52). Moreover, individuals reporting mild cognitive impairment showed significantly poorer performance than matched controls (52). To surmount issues relating to ascertainment bias, we studied parental carriers recruited on the basis of a child referred for genetic testing. Non-carrier parents were also evaluated to guard against misattributing effects of environment to genotype (59, 60). 22 carriers (7 males and 15 females; Mean age 39.7±9.6 yrs.) and 22 non-carriers (9 males and 13 females; Mean age 42.6±9.0 yrs.) completed the test battery.

Based on work suggesting that carriers reported difficulties with reading and mathematics (3), performance on the Grammatical Reasoning (GR) and the Arithmetic Reasoning (AR) tasks within the NCPT was evaluated first, under the a priori hypothesis that deletion carriers would do worse than population controls (p_{hypothesis.corrected} < 1.8 × 10⁻²). Deficits were observed for both tasks (Table 2 and Figure 1). Deletion carriers took significantly longer than controls to correctly complete trials within the GR task, which has subjects evaluate whether sentences accurately describe the positioning of shapes presented alongside them (Figure 1A, top; p=6.3 × 10⁻³; Cohen’s d=+0.72, 95% CI +0.06 to +1.4). No such effect was observed in non-carriers (Figure 1A, bottom; p=0.66). A near significant decrease in GR score was also observed in carriers (Figure 1B, top; p=2.9 × 10”²; Cohen’s d=−0.55, 95% CI −1.2 to +0.08), but not non-carriers (Figure 1B, bottom; p=0.30). Carriers likewise scored significantly lower than population controls on the AR task in which subjects are presented with an arithmetic problem written out in words and asked to answer using numbers (Figure 1C, top; p=1.3 × 10”²; Cohen’s d=−0.66, 95% CI −1.3 to −0.02). No such effect was observed in non-carriers (Figure 1C, bottom; p=0.17).

Table 2.

Domain specific cognitive impairment in adult BP1-2 deletion carriers but not non-carrier adults from the same homes

	Deletions			Non-Carriers
Test a	Mean Difference (SD)b	Proportion worse than controls c	P-value d	Mean Difference (SD) b	Proportion worse than controls c	P-value d
Grammatical Reasoning (Time per correct trial, msec)	599.23 (1001.79)	16 / 21	6.3 × 10−3	−106.44 (1196.10)	8 / 21	6.6 × 10−1
Grammatical Reasoning (# correct minus # incorrect)	−1.35 (3.15)	15 / 22	2.9 × 10−2	−0.53 (4.67)	12 / 22	3.0 × 10−1
Arithmetic Reasoning (# correct minus # incorrect)	−1.93 (3.76)	15 / 22	1.3 × 10−2	−1.18 (5.77)	14 / 22	1.7 × 10−1
Digit Symbol Coding (# correct minus # incorrect)	−1.86 (5.99)	12 / 22	8.0 × 10−2	−0.32 (8.07)	10 / 22	4.3 × 10−1
Digit Symbol Coding (Time per trial, msec)	80.68 (331.84)	12 / 22	1.3 × 10−1	−62.52 (402.35)	8 / 21	7.6 × 10−1
Divided Visual Attention (Minimum Time, sec)	−40.05 (121.75)	8 / 22	9.3 × 10−1	−34.35 (145.33)	8 / 22	8.6 × 10−1
Reverse Memory Span (# correct)	−0.65 (0.89)	15 / 22	1.2 × 10−3	0.02 (0.76)	10 / 22	5.6 × 10−1
Forward Memory Span (# correct)	−0.05 (1.05)	10 / 22	4.2 × 10−1	−0.46 (0.93)	13 / 21	1.7 × 10−2
Progressive Matrices (# correct)	0.43 (3.44)	8 / 22	7.2 × 10−1	−0.67 (3.36)	10 / 22	1.8 × 10−1
Go/No-Go (Time per trial, msec)	−17.02 (67.28)	8 / 22	8.8 × 10−1	12.39 (55.50)	10 / 22	1.5 × 10−1
Trail Making A e (Time to complete, msec)	−1628.59 (3785.56)	4 / 22	9.7 × 10−1	−3402.18 (4605.57)	5 / 21	1.0
Trail Making B (Time to complete, msec)	−1704.08 (11065.08)	8 / 21	7.6 × 10−1	74.52 (15860.89)	10 / 21	4.9 × 10−1

^aBased on previously published work (3), performance on the grammatical reasoning and arithmetic reasoning tests was evaluated under the a priori hypothesis that deletion carriers would do worse than population controls. Results for these tests are highlighted in grey.

^bPerformance for each subject was compared to a population of controls matched for age, education, and gender and differences calculated. For score based measures, negative values correspond to relatively worse performance. For time based measures, positive values correspond to relatively poorer performance. SD corresponds to Standard Deviations.

^cOutliers (greater than 3 SD from the mean) were removed prior to analyses, resulting in differences from endpoint to endpoint in the number of subjects considered.

^dFor a priori tests (grey), p-values surviving correction for multiple comparisons are bolded (p_threshold < 1.8×10⁻²). For the remaining endpoints, p-values surviving correction for the full set of measures evaluated are bolded and italicized (p_threshold < 5.1×10⁻³). In each case, one tailed tests were carried out to identify deficits present in either group.

^eSurprisingly, non-carrier subjects performed significantly better than matched population controls on the Trail Making A test (p=2.9 × 10⁻³). Performance on this task was also nominally better in deletion carriers (p=5.7 ×10⁻²).

Figure 1. BP1-2 deletion specific impairments in grammatical reasoning, arithmetic reasoning, and working memory.

Task performance for individuals harboring deletions (red circles) and non-carrier adult family members (green triangles) was compared to matched population controls and difference scores were obtained (Y axis). (A) Based on previously published work (3), performance on the grammatical reasoning and arithmetic reasoning tests was evaluated under the a priori hypothesis that deletion carriers would do worse than population controls (p_{hypothesis.correted} < 1.8 × 10⁻²). Relative to population controls, deletion carriers took significantly longer to answer individual questions within a grammatical reasoning task (p=6.3 × 10⁻³; Cohen’s d=+0.72, 95% CI +0.06 to +1.4). No such effect was observed in non-carriers (p=0.66). (B) A near significant decrease in grammatical reasoning score (number correct minus number incorrect) was observed in deletion carriers (p=2.9 × 10⁻²; Cohen’s d=−0.55, 95% CI −1.2 to +0.08), but not non-carriers (p=0.30) (C) Deletion carriers scored significantly lower than population controls on an arithmetic reasoning tasks (p=1.3 × 10⁻²; Cohen’s d=−0.66, 95% CI −1.3 to −0.02). No such effect was observed in non-carriers (p=0.17). (D)Performance on additional tasks was also examined, but a more stringent p-value threshold employed to correct for all 12 endpoints evaluated (p_{all.corrected} < 5.1 × 10⁻³). Deletion carriers scored significantly worse than population controls on a reverse memory span task (p=1.2 × 10⁻³; Cohen’s d=−1.0, 95% CI −1.7 to −0.36). No such effect was observed in non-carriers (p=0.18). (E)Neither deletion subjects nor non-carriers showed deficits on the less taxing forward memory span task (p=0.42 and 0.02, respectively). (F) Performance IQ as assessed by the Raven’s Progressive Matrices task was similar to controls in both deletion carriers (p=0.72) and non-carriers (p=0.18).

Performance on additional tasks was also examined (Table 2), but a more stringent p-value threshold employed to correct for multiple comparisons (p_{all.corrected} < 5.1 × 10⁻³). These analyses determined that deletion carriers scored significantly worse than population controls on a Reverse Memory Span (RMS) task that has subjects click on rectangles in the opposite order to which they changed color (Figure 1D, top; p=1.2 × 10”³; Cohen’s d=−1.0, 95% CI −1.7 to −0.36). No such working memory effect was observed in non-carriers (Figure 1D, bottom; p=0.18). Neither deletion carriers nor non-carriers showed deficits on a cognitively less taxing Forward Memory Span (FMS) task (Figure 1E; p=0.42 and 0.02, respectively). Performance IQ as assessed by the Raven’s Progressive Matrices task was similar to matched controls in carriers and non-carriers (Figure 1F; p=0.72 and p=0.18, respectively). Similarly, no deficits were observed in either group for additional endpoints examined (Table 2 and Figure S1), although non-carriers were faster than matched population controls on the Trail Making A test used to assess visual search ability (p_two.tailed=2.9 × 10⁻³; Cohen’s d=−1.0, 95% CI −1.7 to −0.36). For this test, subjects were tasked with tracing a path between numbered circles as rapidly as possible. Performance on this task was nominally better in deletion carriers relative to controls, although this difference was not significant (p_two.tailed=5.7 ×10⁻²; Cohen’s d=−0.50, 95% CI −0.13 to +1.1).

Cyfip haploinsufficiency in flies revealed associative learning and habituation deficits

Given published work linking one of four genes within the BP1-2 deletion to synaptic plasticity and LTP (20, 21, 23, 24, 29), we reasoned that the cognitive deficits we observed in carriers might be attributable to haploinsufficiency of CYFIP1. To test this hypothesis, we characterized associative and non-associative learning paradigms in Cyfip^85.1/+ flies, heterozygous for a null mutation in the Drosophila homolog of the human CYFIP1 gene. No between group differences were evident for avoidance of aversive odors (octanol and benzaldehyde) or an electric shock (Figure 2A, p>0.05), demonstrating intact sensory capabilities in Cyfip^85.1/+ flies. Behavior in an associative learning paradigm, however, identified a deficit in Cyfip^85.1/+ animals. More specifically, avoidance of a shock-paired odor was significantly reduced in Cyfip^85.1/+ flies relative to controls (Figure 2B, p<1.0 × 10⁻³). Furthermore, while control flies showed significantly reduced habituation to electric shock, a form of non-associative learning (Figure 2C, p<1.0 × 10⁻³), mutant flies were resistant to habituation. Behaviour in the light-off jump reflex assay was similar between the two genotypes (Figure 2D, p>0.05), however, suggesting that at least some forms of short-term habituation remained intact.

Figure 2: Associative learning and habituation deficits in flies with reduced Cyfip dosage.

(A) Control and Cyfip^85.1/+ flies avoided an environment containing octanol or benzaldehyde. Both genotypes likewise avoided entering an environment in which they received an electric shock, indicating intact sensory perception. No significant between group differences were observed (p>0.05; ≥11 groups of 50 flies per genotype). (B) Cyfip^85.1/+ mutant flies showed significantly impaired associative learning in a negative reinforcement paradigm in comparison to control animals (p<1.0 × 10⁻³; 17 groups of 50 flies per genotype). (C) Cyfip^85.1/+ flies showed diminished habituation to shock, a deficit in non-associative learning, when compared to control animals (p<1.0 × 10⁻³; 13 groups of 50 flies per genotype). (D) No difference between control and Cyfip^85.1/+ flies was observed in a second test of habituation, the light-off jump assay (p>0.05).

DISCUSSION

BP1-2 deletions are associated with increased risk for neurodevelopmental disorders and present in 1 in 500 individuals. We show here that carriers, recruited for this study because of genotype as opposed to clinical diagnosis, show marked deficits in grammatical and mathematical reasoning. These results are consistent with an earlier questionnaire-based study that determined that BP1-2 deletion carriers reported significantly greater difficulties than non-carriers in reading (ARHQ) and math (AMHQ) (3). Whereas Stefansson and colleagues inferred competency from self-report, we assessed subject performance directly. This is an important distinction in that subjectivity is reduced and inherent biases diminished. For example, ARHQ scores are highly sensitive to subject age (61). It’s unclear whether we would have identified deficits in our much smaller cohort had we relied on self-report. It’s also still unclear what specific aspects of cognition are compromised in deletion carriers. Stefansson and colleagues found no effect of genotype on verbal IQ as measured by the Wechsler Abbreviated Scale of Intelligence or verbal fluency as determined by the Controlled Oral Word Association Test. Similarly, we saw no effect on the Digit Symbol Coding task despite a strong correlation between outcomes here and those for the Arithmetic Reasoning task (Pearsons’ r=0.48). Underscoring the complexity here, this correlation is greater than that between performance on the grammatical and arithmetic reasoning tasks in which deficits were identified (Pearsons’ r=0.4). Additional clarity may emerge from brain imaging studies, in that rare and common variants at the BP1-2 locus have been associated with variation in grey matter volume and white matter integrity (3, 18, 62).

We also report for the first-time impairments in working memory which could underlie the observed deficits in reading and math-related performance. No memory-related findings were observed by Stefansson and his colleagues. In their study, memory was assessed using the Logical Memory subtest from the Wechsler Memory Scale III, which has subjects retell stories told to them either immediately or after a delay. Working memory was also evaluated by them by having subjects perform the Spatial Working Memory subtest within the Cambridge Neuropsychological Test Automated Battery (CANTAB); subjects are presented with colored squares and asked to identify hidden tokens through a process of elimination. The effects we observed are large, with Cohen’s d values as high as 1.0, corresponding to a 76% chance that a randomly selected BP1-2 deletion will perform worse than a non-carrier (63). As such, results argue against the deletion being either benign or of unknown significance and suggest that findings be shared with families. This said, because outcomes for individuals cannot be predicted with good precision, accurate communication of results will be challenging.

Results from fly, demonstrating deficits in associative learning and habituation in Cyfip^85.1/+ animals, show that in this species, reduced dosage of this single gene is sufficient to cause the cognitive deficits we observed in human deletion carriers. Consistent with this idea is that individuals with Fragile X Syndrome, a disorder arising from dysregulation of the CYFIP1 interacting protein FMRP, do especially poorly on tasks involving arithmetic processing and memory span (64-66). Also consistent is that regulatory variants in CYFIP1 linked to FOXP2 are associated with variation in surface area of the language-related supramarginal gyrus (62). In this context, additional study of Cyfip^85.1/+ animals may identify new strategies for addressing the cognitive impairments observed in disorders like schizophrenia for which individuals with BP1-2 deletions are at increased risk. A limitation here, however, is that it’s not yet clear the degree to which Cyfip associated cognitive deficits in fly will inform our understanding of the impact of BP1-2 deletions in humans. As reviewed immediately above, this issue is particularly relevant for BP1-2 deletion associated deficits given distinct findings from subtly different neuropsychological tests.

Finally, results from this study demonstrate the utility of a web-based approach in the characterization of a geographically distributed population. Given ever falling costs in generating and analyzing complex genetic data, acquiring phenotype data at scale represents the primary bottleneck in uncovering new genotype-phenotype relationships. The application of high throughput remote phenotyping represents a potential solution here and may make possible studies that could not have been undertaken otherwise.