Pleiotropy, genomic background, and complexity

A recent paper in Nature Reviews: Genetics, "Pleiotropy in complex traits: challenges and strategies" (Solovieff et al.) raises some interesting questions (it's paywalled, so we'll try to summarize succinctly).  Genomewide association studies (GWAS), they say, have identified numerous gene variants that seem to have effects on multiple traits (they call these "cross phenotype" or CP effects).  This seems to be particularly true for autoimmune disorders, neuropsychiatric disorders and cancers.

In some instances this strengthens epidemiological evidence for what seem to be clinically related traits -- schizophrenia and bipolar disorder, for example, have been found to be familial and GWAS results suggest that they share genetic risk factors as well.  A single genetic variant, protein tyrosine phosphatase non-receptor type 22 (PTPN22), has been linked with "rheumatoid arthritis, Crohn's disease, systemic lupus erythematosus and type 1 diabetes", all autoimmune diseases.  The telomerase reverse transcriptase (TERT)–CLPTM1-like (CLPTM1L) locus has been linked with glioma, bladder and lung cancers.

So, it seems that the different traits share a common genetic pathway. 

At first glance, several scenarios fit these observations: distinct effects of the same allele in different cell populations underlying associations with different diseases or disease groups; a single molecular effect having multiple morphological or physiological consequences; or a CP effect tagging two different causal variants within the same gene that result in different functions and affect different phenotypes.

CP effects, the association of a single locus with multiple traits, can be due to other than true pleiotropy, the term referring to the effect of a locus on more than one trait.  Instead they could be reflecting a shared genetic pathway, or they could be spurious, but in any case the paper addresses the question of how to distinguish biological from spurious pleitropy, and describes various statistical approaches for determining CP.

Interesting, but not the most interesting issues, to us.  Humans have only 20-25,000 protein-coding genes and we've defined many thousands of illnesses, and we've got many thousands of normal traits, and there can't be a gene 'for' every disease or trait.  Indeed, it is we the researchers who sometimes create categories of disease that make sense phenotypically but aren't biologically distinct -- while also lumping traits that are biologically distinct.  So, it must be that sometimes pleiotropy is a social construct, so to speak.   

But multigene pathways help to sculpt traits and diseases, and genes that aren't end-product genes (genes that code for mineralization or the lens of the eye, e.g.) have been recruited by multiple pathways for multiple purposes.  So if a gene variant interrupts the production of a protein in one pathway, it's likely to do so in all the pathways it's used in.

Except that that doesn't seem to always be true, which we find more curious than pleiotropy.  Why can a known mutation in, say, fibroblast growth factor receptor 2 (FGFR2) be associated with something as localized as premature closure of cranial sutures (craniosynostosis) but have no apparent effect in other pathways in which it's clearly shown to be expressed?  Half of all the FGFR2's an affected individual's cells make (assuming one copy of the gene is defective, the other normal) will be equally disrupted, after all, wherever they are.  So each cell expressing the gene should be comparably abnormal.  Why do inbred littermates with known  craniosynostosis-associated FGFR2 receptor mutations have different phenotypes, from no effect to several fused sutures?  (Here's a paper describing this.)

There are several issues here.  First, it is assumed but often not true that both copies of a gene that a person has are expressed in the cell that uses the gene.  It is further assumed, but rarely tested and indeed difficult to test, that each copy of the gene is expressed at the same level in the cell.  Second, somatic mutation may mean that the gene is not identical in the cells in an individual and depending on when such a mutation occurred during embryonic development there can be different fractions of descendant, affected cells.

Third, what happens as a consequence of  FGFR2 expression depends on the rest of the animal's genotype -- because the FGFR2 protein is related to signaling (communication) between cells.  Since somatic mutations are occurring all across the genome, and since there are probabilistic aspects of the signal sending and detecting, each animal will be somewhat different -- even with the same genotype.  Another way to see it is the very well known variation between different inbred mouse strains in which the same genetic mutation has been engineered.  This shows clearly that genomic background affects the result.

Indeed, it is of course quite interesting and important to understand if there are any deeper reasons why there are families in which one member has a severe disorder while others with the same genetic variant remain completely unaffected.  Hemochromotosis is an example, periodic paralyses are other examples, but there are many.  For this, the escape-valve term 'penetrance' was invented long ago, to mean a gene that does something every time....except when it doesn't.  It tends to connote weak effects or stochastic effects (since penetrance 'probabilities' are estimated), but it is at least as likely to reflect genomic background variation; in that case it is improper to consider it to be a probability since that suggests a fixed, stochastic effect rather than one highly dependent but perhaps specific, on context.

So the explanation for variable 'penetrance' is presumably genetic background.  That's a very challenging issue, for which we have inadequate means of addressing.  Each of us has his/her own unique background, and it responds differently to external environmental insults, stimuli, and other factors, as well as what's going on inside our cells.  GWAS gives us some evidence, but does not provide a good conceptual way to understand genomic causation.