Increased Prevalence of Rare Copy Number Variants in Treatment-Resistant Psychosis

Subject Recruitment and Screening

Participants were recruited between April 2015 and August 2019 from five Pennsylvania state hospitals (PASH) and their affiliated long-term structured residence (LTSR) facilities. LTSRs are smaller locked facilities with 24-hour staff coverage and medication management that allow for the “deinstitutionalization” of patients back into the community but are not necessarily an indication of clinical improvement. The study was approved by the Institutional Review Boards at the Drexel University College of Medicine and the University of North Carolina at Chapel Hill, the Chief Medical Officer of the Commonwealth of Pennsylvania, Department of Public Welfare, and individual recruitment sites. Research was carried out in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants (or legal guardian with participant assent).

Subject selection was based on information derived from direct patient interviews and medical records. Inclusion criteria were: 1) ability and willingness to give informed consent (or written consent of a legal guardian with subject assent); 2) age ≥18 years; 3) current chart diagnosis of schizophrenia, schizoaffective disorder, bipolar I disorder with psychotic features, major depressive disorder with psychotic features, or psychotic disorder NOS; 4) continuous psychiatric hospitalization for ≥5 years; and 5) lack of clinical improvement despite ≥3 antipsychotic drug (APD) trials of adequate dose and duration. Exclusion criteria were: 1) a psychotic disorder associated with substance dependence; 2) a medical condition known to cause psychosis; 3) any instance of sustained treatment response. No study participants were prisoners or on forensic units at the time of enrollment. A senior psychiatrist not associated with the study (T.S., see acknowledgments) and one of the senior authors (R.C.J.) reviewed cases to confirm diagnostic inclusion/exclusion criteria (see Supplemental Material for additional details).

Structured interviews were conducted with subjects who were willing and able to participate (see Table 1). The Ammons Quick Test ²⁵ was used to estimate intelligence, and the Positive and Negative Syndrome Scale (PANSS) ²⁶ was administered to document current psychotic symptoms. Demographic and phenotypic data were extracted from medical records and entered into a secure electronic data capture platform (REDCap) ^{27, 28}. Figure 1 provides an overview of subject selection. From 690 initial participants, 509 were included in the final sample.

• Download figure
• Open in new tab

Figure 1: Spectrum of Treatment-Resistance and Study Workflow

Ultra-TRS describes those with TRS who also failed to respond to clozapine treatment; AGRAN, samples taken from a previous study of agranulocytosis in patients with TRS ⁷⁷; PASH, Pennsylvania State Hospital; LTSR, long term structured residence; CLIA, clinical laboratory improvement amendment; SCZ, schizophrenia.

Genomic Analysis

Genomic DNA was extracted from a blood sample and genome-wide SNP genotypes obtained using Illumina Infinium Global Screening Array (v2.0+MD+Psych) per standard protocols (Life & Brain GmbH, Bonn, Germany). CNVs were called from array intensity data using PennCNV ²⁹, QuantiSNP ³⁰, and iPattern ³¹, as packaged by EnsembleCNV ³². CNV calling was performed for chromosomes 1-22 and chromosome X. Individual samples were removed (n=17) due to outliers in quality control (QC) metrics (standard deviations (SD) of log R ratio (LRR), SD of B allele frequency (BAF), wave factor in LRR, BAF drift, and/or the number of CNVs detected). Additional QC steps were then followed to remove low confidence CNV calls spanning <10 probes, confidence score <10 for PennCNV, Log Bayes Factor <10 for QuantiSNP, score <1 or complex CNVs for iPattern, or with >50% reciprocal overlap with large genomic gaps (e.g., centromeres) or regions subject to rearrangement in white blood cells. We annealed adjoining CNVs that appeared to be artificially split by the CNV calling algorithm by recursively joining CNVs if the called region is ≥75% of the entire region to be joined and then removed CNVs with <100kb length. CNVs detected by two of the three calling algorithms were intersected by requiring the same subject, same CNV type, and >50% reciprocal overlap. We excluded CNVs detected by only one algorithm and merged the remaining concordant CNVs using outer probe boundaries. This approach allowed us to improve specificity (detected by ≥2 algorithms) as well as sensitivity (combining all concordant calls among three complementary algorithms). A subset of CNV calls were confirmed using a clinical-grade Agilent comparative genome hybridization array in a CLIA-certified lab (Allele Diagnostics, Spokane, WA).

Exome sequencing data were available for 478 subjects and used as a second approach to CNV analysis, and as the primary source for CNV calling in cases whose samples were excluded from SNP-based CNV analysis. Whole exome sequencing was performed using Agilent SureSelectXT Clinical Research Exome capture and was sequenced using paired end 150 bp reads on Illumina NovaSeq (GeneByGene, Houston, Texas, USA). Sequences were aligned to hg19, variants called using the GATK pipeline ³³ and annotated using Variant Effect Predictor ³⁴. CNVs were called using XHMM ³⁵ and CN-Learn ³⁶. For the training input to CN-Learn, previously CLIA-confirmed CNVs from the current study (n=21) along with other high confidence CNVs called from the Global Screening Array (n=10) were used (recommended minimum number of training CNVs for CN-Learn is 31). See Supplemental Material for diagram of CNV calling workflow.

Called CNVs were cross-referenced with a curated list of 74 CNVs associated with neurodevelopmental disorders (NDD CNVs) based on large case-control studies, including 12 associated with schizophrenia (SCZ CNVs; see Supplemental Table S2 for complete list with genomic coordinates) ^{15, 37, 38}. Large (>1 Mb) CNVs that did not overlap with known NDD CNV loci are also considered to have potential clinical value and are reported here ³⁹. Three cases with CNVs >1 Mb in size were considered incidental findings and not included in this prevalence analysis as they are unlikely to be relevant to psychiatric disease. These included a mosaic copy loss of the Y chromosome, a 17p12 deletion associated with Hereditary Neuropathy with Liability to Pressure Palsies, and a 16p11.2-p11.1 duplication in a region non-sensitive to copy number dosage changes (hg19 chr16:34,197,143-35,257,261; https://dosage.clinicalgenome.org/).

Statistical Analysis

To test whether SCZ CNVs, in aggregate, increase risk for TRS, we compared the prevalence of SCZ CNVs in this sample to previously published data from a large schizophrenia sample (21,094 subjects not selected for treatment resistance) ¹⁵ using a chi-squared test of independence. These same data were used to test for associations with individual SCZ CNVs using 2-sided Fisher exact tests, with p-values adjusted to correct for multiple testing using Benjamini–Hochberg false discovery rate (BHFDR; adjusted p-values <0.05 were considered statistically significant). Only variants observed at least once in this TRS sample, and for which data were available in Marshall et al. ¹⁵ (CNV probe-level results), were included in loci-based analysis. Additionally, we checked for demographic differences between CNV carriers and non-carriers across all variables listed in Table 1; only Positive PANSS scores were significantly different between groups t(361) = 1.9958; uncorrected p = 0.0467), with CNV carriers (M = 20.97; SD = 4.824967) having higher average scores than non-carriers (M = 18.91; SD = 5.72927). All statistical analyses were performed using R Statistical Software (v3.6.3) ⁴⁰ and the epitools R package (v0.5-10.1) ⁴¹.

Known Risk CNVs for Neurodevelopmental/Psychotic Phenotypes

Of 24 NDD CNVs identified, 21 affected one of the 12 SCZ CNV loci (see Table 2). The prevalence of SCZ CNVs in this TRS sample (4.1%) represents a statistically significant increase compared to a large previously published schizophrenia cohort (2.2%) ¹⁵, (χ2 (1, N = 21,603) = 7.77, p = 0.005322; OR = 1.93; 95% CI, 1.24-3.02). Importantly, the combined prevalence of these 12 SCZ CNVs has been replicated with relatively consistent findings across multiple schizophrenia samples (∼2.1 - 2.3%) ^{21, 24, 37}. This is indicative of an association between this group of CNVs and increased risk for treatment resistance in psychosis.

To identify associations of individual SCZ CNVs with TRS, we independently tested the prevalence of each SCZ CNV in our sample against that observed in Marshall et al. ¹⁵. Duplications at 15q11.2-q13.1 and 16p11.2 were significantly associated with TRS after correction for multiple testing. We found a greater than 11-fold increase in the prevalence of 15q11.2-q13.1 duplications (0.78%; corrected p = 0.00624; OR = 11.1; 95% CI, 3.68-33.66) and an almost four-fold increase in the prevalence of 16p11.2 duplications (1.18%; corrected p = 0.0199; OR = 3.98; 95% CI, 1.72-9.24). When these two CNVs were removed from the aggregate analysis, there was no longer a statistically significant increase in the combined prevalence of SCZ CNVs (χ2 (1, N= 21,515) = 0.2195, p = 0.6394), suggesting that these CNVs drive the collective increase in prevalence of SCZ CNVs in this sample.

Additional CNV Findings

As shown in Table 3, 11 participants (2.2% of the total sample) carried large CNVs (>1 Mb) in regions not previously associated with neurodevelopmental/psychiatric phenotypes in case-control studies (one case carried two large CNVs). For most of these loci, case reports exist which describe neurodevelopmental phenotypes in individuals with overlapping CNVs, including one CNV which was previously reported in patients with psychotic symptoms (see Supplemental Table S3). Many of these loci involve genes highly expressed in the brain or have been implicated in central nervous system or neurodevelopmental disease (Table 3).

Finally, 12 participants (2.4%) carried one or more VUS CNVs which do not fit the above categories but nevertheless may influence phenotypic expression in carriers (see Table 4). Seven carried duplications in regions where deletions are known to be associated with neurodevelopmental phenotypes, including 2q37, 15q11.2, 16p12.2, and distal 22q11.2. Six carried CNVs which overlapped with known NDD risk CNVs by less than 50%, including deletions of 1q21.1 and 22q11.2, as well as deletions and duplications of 15q13.3. These regions have known susceptibility to structural rearrangements that vary in size and breakpoint location due to the presence of several genomic low copy repeats, with some data supporting the phenotypic effects of smaller or atypical CNVs ^42-45. One case was a carrier of both 15q13.3 and 2q37 duplications.

Generalized vs Specific Risk

From the perspective of generalized risk, any effect that disease-associated CNVs exert on TRS risk may be mediated first through early neurocognitive impact. There is compelling evidence for neurodevelopmental factors being implicated in TRS. Treatment resistance has been genetically correlated with traits related to intelligence and cognition in terms of having overlap in polygenic risk alleles ¹². Studies have also found lower premorbid IQ, younger age at onset, and poorer premorbid social adjustment are associated with TRS ^{10, 11, 48}. Similar clinical features are also predictive of which individuals with schizophrenia carry clinically relevant CNVs. Specifically, a high prevalence of disease-associated CNVs are found in schizophrenia patients with comorbid intellectual disability (24%; IQ < 70), and/or childhood onset psychosis (11.9%; onset < 13 years of age) ^{11, 49, 50}.

A report from Kushima et al. ²⁴ provides more direct evidence that neurodevelopmental burden in CNV carriers may be related to APD response. Among patients with schizophrenia carrying disease-associated CNVs, treatment resistance was seen more than twice as often in patients with co-occurring congenital and/or developmental phenotypes (48% had TRS) versus those without developmental phenotypes (21.4% had TRS). Importantly, no CNVs have been identified which increase the risk for psychosis alone. All SCZ CNVs also increase risk for various other neurodevelopmental disorders (e.g., developmental delay, intellectual disability, autism spectrum disorder), and often appear at higher rates in neurodevelopmental cohorts than in schizophrenia cohorts. In combination with the findings from our TRS sample, these data suggest that SCZ CNVs may offer unique opportunities to study processes and shared mechanisms between neurodevelopmental burden, psychosis, and APD response.

Conversely, there is an important aspect of specificity in the contribution of CNVs to neuropsychiatric outcomes, especially regarding psychotic phenotypes. Of the most highly established NDD CNVs, only a small subset confer risk for schizophrenia ^{15, 37}. It is likely that the distinct effects that CNVs have on neurobiology underly such genotype-specific outcomes ^{51, 52}. From the 74 NDD CNVs that we screened for (Table S2), 88% (21/24) of the NDD CNVs identified in our TRS sample affected one of the 12 SCZ CNV loci (Table 2). Furthermore, the overall increase in SCZ CNVs observed in this sample is driven by duplications at only two loci; 16p11.2 and 15q11.2-q13.1. Also of note, CNVs occurring within the larger 16p13.11-p11.2 and 15q11.2-q13.3 genomic regions collectively affect 5.1% of our sample and account for 26 of the 47 CNV carriers in our sample overall (n = 10 and n = 16 with 16p13.11-p11.2 and 15q11.2-q13.3 CNVs, respectively). If replicated, these two CNV loci may represent important sites to focus exploration into TRS biomechanisms.

15q11.2-q13.1 and 16p11.2 Duplications

Duplications at 16p11.2 and 15q11.2-q13.1 are both associated with variable early clinical outcomes such as global developmental delays, neurological disturbances, and autism spectrum disorder among others) which have been extensively studied ^53-59. Beside a few isolated reports, however, the longitudinal effects of these two variants have remained largely unexplored in clinically affected adults, particularly regarding APD response. Kushima et al. ²⁴ identified 16p11.2 duplications in three adult patients with schizophrenia, of which two were reported as treatment-resistant with prominent positive symptoms, while the third (age 18) had no history of antipsychotic treatment. Another 16p11.2 duplication carrier (age 18) reported by Filges et al. ⁶⁰, presented with psychiatric symptoms including anxiety, mood lability, and paranoid ideation which were minimally responsive to treatment with quetiapine and sertraline. Our findings add six additional 16p11.2 duplication cases with TRS to the literature.

To our knowledge, no reports exist which document APD treatment response in adult 15q11.2-q13.1 duplications carriers, although one study ⁶¹ indicated a relative inefficacy of benzodiazepines in the treatment of seizures in 15q11.2-q13.1 duplication carriers, which was hypothesized to be related to disruption in GABAergic signaling (several GABAβ3 receptor genes are located in the 15q11.2-13.1 region). Likewise, mouse models of the 16p11.2 duplication highlight behaviorally-relevant changes in to GABAergic interneuron and cortical pyramidal neuron activity leading to excitatory-inhibitory imbalances ^62-64. Disruption in GABAergic inhibitory signaling has previously been implicated in psychosis and APD response, and thus may represent one possible convergent mechanism of these CNV sites for TRS ^{65, 66}. Importantly, both regions contain several dozen genes, and therefore mechanisms of action related to APD response may be complex and multigenic. Yet, our findings suggest that 16p11.2 and 15q11.2-q13.1 duplications may provide opportunities to model treatment resistance within experimental systems, with the aim of uncovering bio-mechanisms or potential drug targets.

Clinical Implications

The potential role of specific CNVs in treatment-resistance is also relevant to discussions surrounding the use of genetic testing in psychiatry. Genome wide screening technologies are routinely used to identify genomic etiologies in pediatric medical/neurodevelopmental settings and the benefits of testing in these populations are well established ^{67, 68}. However, it remains an open question as to whether testing is similarly justified for other psychiatric conditions based on considerations of potential harms and benefits ⁶⁹. As we have previously argued ^{70, 71}, there is a shortage of the type of detailed clinical data needed to investigate the impact that large-effect CNVs have on health comorbidities and treatment outcomes in adult psychiatric patients. This limited knowledgebase has made it impossible to assess areas in which specific genetic etiologies may provide guidance to patients’ treatment management and understanding of individual health risks ⁷².

The 22q11.2 deletion syndrome (22q11DS) stands out among the group of SCZ CNVs, in that thorough investigations have been undertaken to characterize the associated phenotype in pediatric and adult carriers alike. As a result, clinical care guidelines have been established ⁷³, and a diagnosis of 22q11DS even has implications for APD prescribing practices ⁷⁴. These resources will become increasingly important in adult psychiatric practice as children diagnosed with 22q11DS (and other CNV syndromes) enter adulthood and seek out psychiatric care. We did not identify a significant increase of 22q11DS in our TRS cohort, which is consistent with the prevailing idea that psychosis associated with 22q11DS responds to APD treatment at rates similar to idiopathic schizophrenia ^{75, 76}. Whether other SCZ CNVs (such as duplications of 15q11.2-q13.1 and 16p11.2) influence treatment-response, and through which biological pathways, should be the focus of future research efforts.

Strengths and Limitations

There remains pervasive heterogeneity in the use of varying definitions of TRS in research. Since large, well-defined TRS cohorts are difficult to acquire, many studies base subject selection on surrogate measures of TRS, such as filled clozapine prescriptions or the use of polypharmacy, without any direct patient contact to confirm the TRS diagnoses or rule out instances of pseudo-TRS (i.e., medication noncompliance, substance abuse). Importantly, each subject in our TRS cohort resided in the protected environment of the PASH system for at least five years with continuous clinical care, medication management, adequate sustenance, and minimal access to substances of abuse.

Our approach to CNV detection (combining research-grade microarrays, exome sequencing, and clinical CLIA lab confirmation) also allowed us to maximize sensitivity while minimizing false positives. This combined approach may have captured CNVs that microarray alone would have missed, which must be considered when making comparisons of our findings to studies that used differing methods. For example, of the six 16p11.2 duplications identified in our TRS sample, only two were called from SNP-based microarray data, whereas all six were called from analysis of exome sequencing data and confirmed by a CLIA-certified lab. Larger studies will be necessary to confirm and further investigate the association of specific CNVs with TRS; however, this will likely necessitate significant collaborative efforts. Furthermore, there is a need for a greater depth of phenotypic data regarding adult psychiatric patients with disease-associated CNVs.