We wished to verify whether different TE types might be differentially represented depending on gene function. TE frequency varies with intron length 24 and GC percentage 1. Moreover, in line with previous findings 24, we show that, although differences exist depending on MCS and TE age, conserved sequences have an overall negative effect on TE fixation frequency (Additional data file 1). For each TE type we therefore performed multiple regression analysis on TE number using intronic GC percentage, intron length and conserved sequence length as independent variables. The fitted values were then used to predict the expected TE number per intron (nTEi_exp). For each gene, the TE normalized abundance (Te_na) was calculated as follows:

( ∑ i ∈ g e n e n T E i o b s − ∑ i ∈ g e n e n T E i exp ⁡ ) ( ∑ i ∈ g e n e n T E i o b s − ∑ i ∈ g e n e n T E i exp ⁡ ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8tuQ8FMI8Gi=hEeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@7361@

where nTEi_obs is the observed number of TEs per intron. These calculations were performed for all TE families in both human and mouse.

For each TE family, genes displaying three times more or less TE than expected (TE_na > 0.5 or TE_na < -0.5) were classified as TE-rich or TE-poor, respectively.

We next used GeneMerge 25 to retrieve significant associations; database annotations for the three categories designated by the Gene Ontology (GO) Consortium (molecular function, biological process and cellular component) were employed. Correction for multiple tests was applied to all statistical analyses. For each significant GO term retrieved, genes that are present in the study set and associate (therefore contribute) to the term are designated as 'contributing genes'. We also calculated MCS density and intergenic TE frequency of contributing genes. In particular, for intergenic sequences, TE_na (igTE_na) was calculated as described for introns; for contributing gene sets the fractional igTE_na deviation was then calculated as:

(Mean igTE_na in contributing genes - mean igTE_na in all genes)/|mean igTE_na in all genes|

Similarly, fractional MCS density deviation was calculated for contributing gene sets.

Data concerning significant (Bonferroni-corrected p value < 0.01) GO associations are summarized in Table 1. Three main molecular function categories were found to be associated with genes displaying low TE_na (for more than one TE family). The first one is accounted for by genes involved in nucleic acid binding and transcription; these loci have, on average, high intronic MCS densities and few TEs in their flanking regions. The second functional category is represented by genes coding for cytokines/growth factors/hormones and, more generally, receptor ligands: these genes do not have, as a whole, higher than average intron conservation and, with the exception of LTR-poor genes, tend to have low igTE_na. The last category (not present among Alu-poor genes) is accounted for by structural molecules, mainly represented by ribosomal proteins. These genes have extremely low MCS densities and igTE_na. These same associations were retrieved for mouse genes (supplementary Table 1 in Additional data file 2), although no GO term was significantly associated with L1-depleted mouse genes.