The genetics of cognition in schizophrenia

Family and Population Studies

There is evidence from many individual studies, supported by meta-analyses, that unaffected first-degree relatives (FDRs) of those with schizophrenia show generalized impairments of cognitive function but to a lesser degree than those seen in probands (51, 52). However, such findings are not universal. One study of schizophrenia cases and their relatives from relatively highly functioning families found that the siblings of probands did not differ in cognitive performance from a community control sample (53). In addition, a very large study of the Swedish population found no evidence for adolescent cognitive impairment in the siblings of people with schizophrenia (54). These findings raise the possibility that the role of familial risk factors for low IQ in schizophrenia may have been influenced by ascertainment bias in studies of probands and their relatives. Another possible confounder is that families ascertained for having a schizophrenia proband may have higher rates of exposure to environments with an impact on cognitive ability, for example cannabis abuse (55). Finally, there is evidence for assortative mating such that people with schizophrenia who have children on average have partners of lower cognitive ability than controls (56). As a result, assortative mating could contribute to, or even account for, deficits in FDRs rather than these reflecting a substantial overlap between the genetic risk for schizophrenia and low IQ. Thus, while many studies have found evidence for impaired cognition in FDRs, questions remain about how these findings should be interpreted. Genomic studies of parent-proband trios might throw light on this issue.

Population studies have further illuminated the causal relationship between premorbid IQ and genetic risk. In a large study of the Swedish population (57), risk of schizophrenia increased by 3.8% with each one-point decrease in premorbid IQ. The effects were stronger in the lower IQ range, but importantly, the association was monotonic, meaning at no point in the distribution was higher IQ associated with higher risk, although such an effect in people with exceptionally high IQ could not be excluded due to the rarity of those individuals. Overall, these findings, and the magnitude of the effect, were almost identical to those obtained in a meta-analysis of earlier population studies (9). The Swedish study (57) noted that the association between premorbid IQ and risk of schizophrenia was of similar magnitude when onsets within 5 years of testing were excluded, suggesting that it does not reflect potential prodromal effects of declining IQ associated with insidious onset. Again, these findings are in accord with those from earlier population studies which excluded onsets immediately prior to testing (51). Finally, the Swedish study found that that risk of schizophrenia in people with high familial liability to the disorder was substantially modified by premorbid IQ, familial susceptibility having a much stronger impact on risk of illness in those with low IQ.

Since intelligence and schizophrenia are both familial and substantially heritable, the authors of the Swedish study (57) tested the hypothesis that the premorbid IQ-schizophrenia association might be the result of genetic (and family environment) factors that predispose to both traits. They undertook co-relative analyses, comparing the strength of the association between intelligence and schizophrenia within various classes of pairs of relatives whose IQs were different, that is not in the same decile. Siblings share more familial factors (genetic and environmental) than do more distal pairs of relatives, and therefore more of the difference in intelligence between them is likely to be attributable to non-familial factors than is the case for difference in intelligence between more distally related pairs. Accordingly, if the association between low IQ and schizophrenia is substantially due to genes or familial environments that influence both traits, the strength of this association within siblings will be less than in the more distantly related relative pairs. However, the findings were inconsistent with this; association between premorbid IQ and schizophrenia was as strong within siblings as within more distant relatives, and even between people in the general population who are effectively unrelated, suggesting the link between the traits is not the result of heritable genetic or familial environmental risk factors that jointly influence both traits. These findings have been supported by two recent population-based sibling studies (58, 59).

Evidence that non-familial factors may have an important influence on premorbid IQ deficits in schizophrenia also comes from studies showing that risk of schizophrenia may be better indexed by a measure of the extent to which an individual's premorbid cognitive performance is lower than expected based on estimates of familial cognitive aptitude than by their absolute premorbid cognitive performance per se (54, 60, 61). It is conceivable that such deviations from familial expectation might occur in individuals who, by an unlucky roll of the genetic dice, inherit an excess of poorer cognition alleles that have pleiotropic effects on liability to schizophrenia. However, this interpretation was not supported by the finding that, when schizophrenia and control probands were matched for cognitive performance, the siblings of the schizophrenia probands had scholastic aptitudes and IQs that did not differ from population means and which were significantly higher than those of the siblings of control probands (54). Instead, this series of studies from Sweden (54, 57, 60) suggest that an important contribution to risk for schizophrenia comes from neurodevelopmental perturbations that impact cognitive development in people who develop schizophrenia that are not caused by familial factors that typically influence cognition within families. This general conclusion also is indirectly supported by evidence that the heritabilities of a range of cognitive abilities in schizophrenia seem to be lower than in the general population (61).

Overlap in Common Risk Alleles for Schizophrenia and IQ

There is evidence that some of the common alleles that influence cognitive ability in the general adult population also influence liability to schizophrenia, but the genetic correlation of −0.21 between the two is modest (46). A similar genetic correlation of −0.22 has also been reported between liability to schizophrenia and intelligence estimated in 16 year olds, an age prior to the typical onset of schizophrenia (62). These correlations suggest that, on average, common variant genetic effects associated with lower intelligence in the general population explain around 5% of liability to schizophrenia, an estimate consistent with a longitudinal population-based twin study which estimated this figure at 7% (63). These findings, based on very different study designs, converge on the conclusion that, while there is some shared genetic liability between IQ and risk of schizophrenia, the overlap is small, and in line with the conclusions from the Swedish family studies reviewed above. It should be noted that genetic correlations of this magnitude could potentially be the result of the type of cross-trait assortative mating (64) discussed above (56), as well as other sources of confounding (65), possibilities that warrant further study. Moreover, shared liability does not imply that risk of schizophrenia is mediated by the effects on cognitive ability, it may instead indicate the existence of pleiotropy whereby common alleles independently influence cognitive ability and liability to schizophrenia, perhaps by contributing to the pleiotropic neurodevelopmental perturbations discussed in the previous section.

A study (66) based on modeling the relationships between schizophrenia polygenic liability [indexed by polygenic risk score (PRS)], schizophrenia and cognition (expressed as latent traits) within families found that the best model solutions suggested around a third of genetic risk of schizophrenia could be explained through causal effects on cognition. However, the modeling does not appear to have accounted for the possibility of pleiotropy, and, as the authors noted, the limitations and assumptions mean that their approach cannot prove causality, and longitudinal designs and/or Mendelian randomization approaches are needed. However, to date, the required longitudinal studies have not been conducted and Mendelian randomization methods have not been able to clearly distinguish between causal and pleiotropic hypotheses (67). Moreover, a recent report that almost all variant sites that influence IQ in the general population also influence liability to schizophrenia, but that the specific alleles that increase schizophrenia risk are a mix of lower and higher IQ alleles, suggests that shared liability cannot indicate a simple causal link between lower IQ per se, and schizophrenia (68).

Common Alleles

Separate to the question about the relationship between alleles that influence cognition in the general population and those that influence liability to schizophrenia is what is the relationship of those sets of alleles to cognitive ability in people with the disorder?

There is strong evidence that alleles influencing IQ in the general population also have effects on cognitive ability in individuals with schizophrenia (69–71). In contrast, the evidence for a relationship between cognitive function in schizophrenia and common variant liability for the disorder is inconsistent (69, 70, 72–75). This inconsistency may reflect the relatively modest sample sizes and power of some of the studies as well as differences in duration of illness at the time of cognitive testing and the nature of the cognitive tests employed. Nevertheless, the results suggest that the effects of common schizophrenia risk alleles, as indexed by a polygenic risk score (PRS), on cognition in people with schizophrenia are at best small, and where direct comparisons have been made (69–71), considerably smaller than the effects of common alleles that influence IQ in the general population (as indexed by an IQ PRS). The IQ PRS explains around 9% of variance in IQ (76) in people with schizophrenia. This is similar to estimates in the general population (62), although one study suggested the variance explained in cases is less than in UK Biobank controls, albeit using different cognitive measures in cases and controls (70). These findings, together with those indicating that the genetic correlation between liability to schizophrenia and IQ is modest, and the findings from the large population family studies reviewed above, suggest that variation in premorbid cognitive function in people who later develop schizophrenia is not substantially the result of common genetic risk factors for schizophrenia. There is also clear evidence from genomic studies that common alleles that influence variance in cognitive ability in the general population also do so in people with the disorder. However, in the absence of direct comparisons of cases and controls based on identical cognitive assessments, it is unclear to what extent those alleles can explain the average premorbid cognitive deficit seen in people who develop schizophrenia.

Rare Variants

People with schizophrenia who carry schizophrenia associated CNVs tend to have worse cognitive function than those do not, the average difference between the two groups in measures of cognitive performance being around 0.5–1.0 SD (77). The greatest reductions in cognitive ability are seen in those with CNVs spanning loss-of-function intolerant (LoFI) genes, that is genes in which loss-of-function mutations are highly disadvantageous for reproductive fitness. There is also evidence that in people with schizophrenia, rare coding risk variants (RCVs), particularly protein-truncating variants in LoFI genes, are associated with reduced cognitive function (71, 76), poorer educational attainment (33, 78) and an increased risk of comorbid intellectual disability (37, 78). The effects of rare variants on cognition are largely manifest before illness onset (see below), suggesting that their effects impact neurodevelopmental processes. Further support for neurodevelopmental effects comes from the observation that the effects of RCVs on cognitive function in schizophrenia are on average greater for those within genes that have been implicated in childhood NDDs than for RCVs in other mutation intolerant genes (37, 76, 78). Given evidence reviewed above that non-familial effects are important, it is also notable that the effect sizes of mutations occurring de novo (which by definition are non-familial) are larger than those of transmitted mutations (38, 76, 78).

Premorbid Cognitive Impairment and Risk of Schizophrenia

Evidence reviewed above suggests that low premorbid IQ is a risk factor for schizophrenia and that this is not explained by prodromal effects of an insidious onset of the disorder. Moreover, rather than an “U”-shaped relationship between IQ and risk of the disorder, highest risk occurs in people with lowest IQ, lowest risk in those with the highest IQ, and on average, people who subsequently develop schizophrenia have a 0.5 SD deficit in premorbid IQ compared with the general population average.

In principle, the strong evidence that schizophrenia risk alleles can also influence cognitive ability in the general population provides a plausible explanation for low premorbid IQ. Intuitively attractive though this hypothesis may be, it is not supported by the evidence we have discussed. Common variant genetic liability to schizophrenia, which currently explains by far most of the attributable heritability of the disorder (82), does not appear to be associated with premorbid IQ (69, 71), a finding broadly consistent with the small genetic correlations observed between liability to psychosis and premorbid IQ in twins and between schizophrenia and IQ in the general population. Thus, the genomic evidence to date is consistent with that outlined above from genetic epidemiology in suggesting that the premorbid cognitive deficit in schizophrenia is largely explained by non-familial rather than familial factors that jointly influence liability to schizophrenia and cognitive ability. There is strong evidence these non-familial factors include de novo mutations (SNVs and CNVs), but these have been implicated in fewer than 5% of cases (82) and therefore other non-familial risk factors must contribute, including as yet unidentified de novo mutations, such as rare structural variants and non-coding variants, non-familial environmental factors, for example in utero or perinatal birth trauma and infections, and stochastic events (83) that contribute to variation in neurodevelopment.

Some caveats should be noted. First, it is conceivable that studies of the effects of schizophrenia PRS on cognition in cases may underestimate effects due to Berkson's paradox (84), also sometimes known as collider bias (84). Assuming PRS and low IQ to be at least partly independent risk factors, people with exposure to higher PRS will require less exposure to the risk conferred by low IQ in order to manifest the disorder, and vice versa. This can result in a spurious negative correlation between the two risk factors in case only studies, or, where a true population association exists, a reduction in the estimated effect size. Secondly, only a minority of schizophrenia heritability is currently attributable to known types of variant (82) and, in principle, classes of variants responsible for the unexplained heritability could show stronger associations with cognition, although the findings from studies of siblings and other relatives reviewed above (57–59) suggest that this is unlikely to be the case. Thirdly, the findings of the key genetic epidemiological studies require further replication.

Having established that low premorbid IQ is a risk factor for schizophrenia the next question is whether low IQ is per se causal or is instead a risk indicator of a pleiotropic neurodevelopmental abnormality that can (largely) independently manifest as low IQ in childhood and the emergence of schizophrenia in later life. The latter interpretation is supported by the observation that risk of schizophrenia is better indexed by the deviation of cognitive performance from that expected, than by absolute premorbid cognitive performance, which is contrary to the expectation if low IQ per se is directly causal. Molecular genetic studies of common and rare variation are also inconsistent with the idea that low IQ per se is on the causal pathway to schizophrenia. Thus, while all known schizophrenia-associated CNVs, and certain schizophrenia—associated RCVs are associated with low IQ, their impacts on risk in schizophrenia is not contingent on the presence of low IQ (41). The observation that common schizophrenia susceptibility alleles include those associated with higher IQ as well as lower IQ similarly suggests that there is no robust causal link between lower IQ and schizophrenia. There is, however, evidence that low premorbid IQ may sensitize individuals to familial risk factors (57) (Figure 1A) and there is a need to substantiate this finding and, if confirmed, explore possible mechanisms.

View Figure

• Download as Powerpoint
• Download Figure

A further issue that warrants discussion is whether there is a neurodevelopmental subtype of schizophrenia characterized by premorbid cognitive impairment. We have discussed this elsewhere (44) and our view is that the evidence better supports the hypothesis that there is a spectrum of neurodevelopmental impairment in those with schizophrenia, rather than a clear distinction between a form of the disorder with cognitive impairment and one without (15). The monotonic change in risk of schizophrenia across the full IQ range, rather than there being an IQ threshold that is associated with a step change in liability, supports this (9, 57) as does the observation that people with lower cognitive ability in schizophrenia do not substantially differ from those with higher cognitive ability with respect to common alleles that confer risk to schizophrenia generally (70). Finally, in people with CNVs known to affect neurodevelopment, schizophrenia is the result of both the CNV and the common variant liability that is shared with general forms of the disorder (85, 86). While we do not believe that the data support the existence of a distinct neurodevelopmental subtype of schizophrenia, the extent of premorbid cognitive impairment, and more particularly the extent that this deviates from familial expectations, seems to index the degree of underlying neurodevelopmental impairment and may be an important clinical and prognostic marker within the disorder (Figure 1B).

We additionally note that while our focus here is schizophrenia, comparisons of the degree of premorbid deviation from familial cognitive aptitude (60) in different psychiatric disorders support the view that the neurodevelopmental continuum extends across a number of conditions, as previously proposed (15, 87). Thus, the effect size of deviation from familial cognitive aptitude was greatest for ASD, followed by schizophrenia and other non-affective psychoses, and least in bipolar disorder (60) supporting the suggestion (15, 87) that there is a gradient of neurodevelopmental pathology across neurodevelopmental and psychiatric disorders. According to this view schizophrenia occupies an intermediate position between childhood neurodevelopmental conditions and bipolar disorder. This helps explain why cognitive impairment is associated with schizophrenia but to a much lesser extent bipolar disorder despite the high common variant genetic correlation between the two conditions.

It is now important to identify the risk factors for, and the nature of, the neurodevelopmental abnormality underlying premorbid cognitive impairments.

It will also be important to determine how and when neurodevelopmental impairment moderates the impact of familial genetic risk for schizophrenia. One hypothesis with potential implications for interventions is that the effects of divergence from familial cognitive expectations on psychopathology might be mediated by the evocation of disrupted family dynamics and/or impairment of individual's self-esteem (88) (Figure 1).

Genetic Factors Influencing the Degree of Cognitive Impairment in Schizophrenia

As we have seen, recent genomic studies suggest that all classes of allele that influence liability to schizophrenia or to variance in IQ contribute to variation in premorbid cognition and to cross-sectional degree of cognitive impairment in people with established schizophrenia. These types of variant currently explain around 10% of variance in premorbid IQ, of which, as noted above, 90% is explained by IQ PRS and the remainder by rare CNVs and rare damaging coding variants in constrained genes. They explain less variance in cognition, around 6%, in those with established schizophrenia (76), but after allowing for the effects of premorbid cognition, this drops to around 1.6%, equally split between IQ and schizophrenia PRS. Thus, most of the genetic contribution to variation in cognition in schizophrenia is mediated by effects that we have argued above are neurodevelopmental. Also note that, while rare mutations contribute only a small amount to variance in premorbid cognition, they are associated with relatively large impairments of cognitive function (76).

A caveat to our conclusion that most of the genetic influences on cognition in schizophrenia are premorbid and likely neurodevelopmental is that, given our incomplete understanding of genetic architecture, we cannot exclude the existence of as yet unknown types of risk variant that show a different balance of morbid and premorbid effects on cognition. The existence of such variants is plausible given evidence we have discussed that any effects on cognition of common risk alleles for schizophrenia seem to manifest postmorbidly rather than premorbidly (69). Another caveat is there may be specific genetic contributions to cognition in schizophrenia that are independent of those that confer liability to the disorder, or to intelligence in the general population.

Genetic Influences on Postonset Cognitive Decline

As we have seen, uncertainties remain concerning the extent to which there is cognitive decline after onset and if so what proportion of cases are affected and when the decline occurs. There is emerging evidence of decline in cognitive function after schizophrenia onset relative to age matched controls, which appears to be progressive over many years (18) and for substantially increased risks of dementia (22, 23). The presence of late life cognitive decline and increased risk of dementia in people with schizophrenia in the absence of increased genetic liability to dementia suggests that schizophrenia may lead to higher exposure, or greater vulnerability, to the same environmental risk factors that operate in the general population to increase risk of dementia, for example smoking, poor cardiovascular health, and lower cognitive reserve, the latter being a consequence of lower premorbid IQ, social isolation, and low rates of employment. A second explanation is that the increased risk of dementia reflects intrinsic pathogenic mechanisms related to schizophrenia, or environmental exposures, that are relatively specific to those with severe mental illness for example medication effects. This is an area that needs further research that includes measures of genetic and potential environmental risk factors and robust measures of cognition ideally over multiple timepoints (24, 26).