99 independent genetic loci influencing IQ/brain health/structure (n=280360)

'Highly expressed' here simply means that the variants in question tend to be located in/affecting genes that are more active in CNS stuff, it doesn't refer to the variance explained. You could identify a gene which is active only in cells in a particular organ yet variants inside that gene might affect little in the way of differences from person to person. This is useful for giving mechanistic hints as to how the variants are having any causal effect; it's also a useful sanity check, as if the IQ GWASes were turning up SNP hits mostly in non-nervous-system expressed genes, it would be highly suspicious.

That said, one of the more counterintuitive aspects of GWAS is that it is possible to identify many variants which affect intelligence while explaining trivial amounts of variance - this is the central limit theorem in action on the thousands of variants, because on net in normal people most variants just cancel out and the genetic contribution to intelligence is relatively few variants (most people differ by only 10 or 15 IQ points, so even if most of individual differences are due to different sets of variants, and we know from GWASes that single variants typically explain something like 0.10 points...), you have to identify with great precision the effect of a great many variants before you start being able to predict anything in normal people (This is weird and counterintuitive and I don't yet have any good metaphors or examples which can make it immediately understandable.)

Finally, on the gripping hand, 4% is weirdly small because the same people with a smaller sample (albeit boosted using MTAG/genetic correlations) were able to produce a PGS explaining almost double the variance, 7%, in "A combined analysis of genetically correlated traits identifies 107 loci associated with intelligence" http://www.biorxiv.org/content/early/2017/07/07/160291.1 , Hill et al 2017, and just yesterday "Large-Scale [n=107k] cognitive GWAS Meta-analysis Reveals Tissue-Specific Neural Expression and Potential Nootropic Drug Targets", Lam et al 2017 http://www.biorxiv.org/content/early/2017/08/16/176842 turned in the same 4% PGS despite different analysis & sample size, so I am suspicious that the validation sample being used is misleading in some way such as using lower-quality IQ measurements. (One of the problems in using independent validation samples rather than something like cross-validation.)