Let’s call this ncDNA for non-coding DNA. I don’t like the name “Junk DNA” because it implies “not worth sequencing.” Also, there is a semantic problem with “junk.” It doesn’t properly address the complexity of genomes and the evolutionary processes going on. Your analogy of genomes to gut bacteria is apt. Genomes are ecologies. An old ecology that could go back to a time before DNA. Variety is a key part. Even related species with the same genome size can have huge differences in the characteristics of their respective ncDNA.  Parts of this ecology are host/associate processes that run the range from predator to parasite to a type of symbiosis. In response the genome has built defenses and through exaptation, has recruited various elements, especially new regulatory elements and sometimes new genes[3].

So I ask myself a question: Why is it that bacterial genomes don’t have ncDNA, especially:

introns
telemeres
centromeres

The null hypothesis would be they got there by accident, contingency.

Ten minutes on Google is a revelation. There are no structures on eukaryote genomes that doesn’t already exist in some form on bacterial genomes. Multiple chromosomes, linear chromosomes, group I and group II introns, retrotrasposons, precursors to centromeres[4]. The tell is that there are three ways bacteria solve the halting problem for linear chromosomes while eukaryote chromosomes only use one, telomere repeats[2]. So my bet would be contingency.

ncDNA as structure:
Seems like there is correlation between genome size and cell size. Also, that newer species tend to have smaller genomes (like the onions). I know that attempts to correlate different phenotypes and genome size in plants have not had very good results. Also, genome folding compartmentalizes gene rich and gene poor regions[1]. Seems the direct relationship is to physiology but I’m a little unclear as to what that means. I took the c-values of some of the various animal groups and did a frequency distribution. Large genomes are rare and small genomes common, the frequency of large genomes fall into a very long tail. Insects especially. If you look at the frequency of plant host families to Lepidoptera species you get a long-tailed graph. Specialists are common and generalists are rare. Frequency of gene introgression vs distance from cline center across a hybrid zone or seed dispersal vs distance, all long-tailed. This means that there is a barrier to larger genomes but the barrier is permeable. This would suggest that large genomes are being recruited for some function but it could be very different function depending on the group.

Comprehensive Mapping of Long-Range Interactions Reveals Folding Principles of the Human Genome, Erez Lieberman-Aiden, Nynke L. van Berkum, Louise Williams, Maxim Imakaev, Tobias Ragoczy, Agnes Telling, Ido Amit, Bryan R. Lajoie, Peter J. Sabo, Michael O. Dorschner, Richard Sandstrom, Bradley Bernstein, M. A. Bender, Mark Groudine, Andreas Gnirke, John Stamatoyannopoulos, Leonid A. Mirny, Eric S. Lander, Job Dekker, Science Oct 2009.
Complete nucleotide sequence of the chlorarachniophyte nucleomorph: Nature’s smallest nucleus, Paul R. Gilson, Vanessa Su, Claudio H. Slamovits, Michael E. Reith, Patrick J. Keeling, Geoffrey I. McFadden, PNAS June 20, 2006.
DNA Transposons and the Evolution of Eukaryotic Genomes, Cédric Feschotte and Ellen J. Pritham, Annu Rev Genet. 2007.
Distribution of Centromere-Like parS Sites in Bacteria: Insights from Comparative Genomics, Jonathan Livny, Yoshiharu Yamaichi, and Matthew K. Waldor, J Bacteriol. 2007 December

1. Species exposed to intense UV (e.g., on land or in freshwater in the Arctic, pelagic plankton, etc.) do not appear to have large genomes. On the other hand, some very large genomes are found in deep-sea invertebrates.
I don’t think that either no-function or protective hypotheses need to explain very large genomes if it is accepted that they are there for no particular “reason” but that their existence may reduce the required fidelity of repair mechanisms (in the case of protective). What is the life cycle of plankton – a few days? Short enough perhaps to not be compromised by UV?
2. Among vertebrates, species with high metabolic rates, and presumably more free oxygen radicals, have smaller genomes than species with lower metabolic rates.
Even the smaller genomes are considerably non-coding (hummingbiord?) and would be sufficient to provide protection to allow survival long enough for reproduction in these short life-span animals. I’m trying not to dismiss out of hand…is there a proposed reason for smaller genomes linked to higher metabolic rates that would make the “cost” of genome size reduction worth it?
3. There can be substantial differences in genome size among similar organisms, for example as in onion and its relatives or among salamanders.
Again – I don’t think that either no-function or protective needs to explain vastly differing genome sizes. Neither propose that genome size evolved to fit specific needs, rather, the protective hypothesis would propose that the repair mechanisms evolved to the necessary fidelity required to cope with whatever circumstances existed. Both no-function and protective accept that the non-coding DNA happened by chance and was allowed to build up because there is not strong reason to get rid of it, since it is not such an energy consuming resource to replicate it etc (except maybe for high metabolic rates where some needs to be discarded?)
4. Despite claims to the contrary based on small and questionable analyses, there are no clear relationships between genome size and lifespan.
Ditto – except that obviously, for a protective role, there would need to be sufficient non-coding DNA to support a life-span long enough to reproduce. This does not predict though that genome size is tightly linked to lifespan – for protection, a genome with a majority of non-coding DNA would be necessary but not sufficient for an adequate life-span.
5. Transposable elements, which are the primary contributor to genome size, can cause a range of mutations through insertion into genes or by causing large deletions by illegitimate recombination, the latter of which is especially likely with the long terminal repeat (LTR) elements that are common in plants.
Mismatch detection and repair systems are very well developed and highly (not 100%) efficient – the greatest cause of DNA damage, by orders of magnitude, is endogenous, so problems related to transposable elements, copying long repeats, etc would presumably be coped with. Also not all mutation is bad even at the gene level, we are full of SNPs, insertions, deletions, CNVs, etc. Most damage is handled, if not the next protection is apoptosis if insertions corrupt crucial genes. Sometimes insertion into some crucial genes though may not be detected as aberrant and could lead to loss of cell growth control, or reduce damage repair fidelity, or anything that may lead eventually to transformed cells. But this could also be an argument for non-coding DNA being useful for protection – the presence of which makes it much less likely that insertions will happen in genes.
6. DNA content obviously can be amplified in somatic cells by endoreduplication, but this tends to be in cells involved in ion exchange, protein production, etc., and not ones exposed most to mutagens (such as the skin exposed to UV).
Not sure what the answer to this is, or if it needs one, given the “passive” role proposed for non-coding genome size regarding protection. For UV other protection mechanisms evolved (e.g. melanin)

David: One way of testing the relationship between genome size and mutation rate (without worrying about selection) would be to look at polyploid plants, by generating tetraploids. This would double the size of the genome, and allow for comparision of the mutation rate in parents and F1

I’d be surprised if that had never been attempted, but I’ll let you search the botanical literature!

I am interested in retrotransposition, and exaptation of retrotransposed elements. In most genomes the distribution is non-random, and indeed different elements may display different patterns of distibution.
Whilst not proving or disproving the hypothesis either way, anecdotally I have come across examples of repeats within repeats within repeats (like a russian doll), suggesting that the initial repeat was a target for a subsequent integration and so on.

One way of testing the relationship between genome size and mutation rate (without worrying about selection) would be to look at polyploid plants, by generating tetraploids. This would double the size of the genome, and allow for comparision of the mutation rate in parents and F1 

Keith Grimaldi: The null hypothesis is the simplest – i.e. no real function, but the protective hypothesis (NOT MINE, just one i think could be possible) is the next simplest that needs to be explained away.

I’m fine with continuing to consider this as a viable hypothesis — my goal here is not to refute it but to move it forward and to point out what it will need in order to be accepted.  However, I have an issue with your logic here.  The simplest explanation is non-function.  The second-simplest is bodyguard function.  The latter is simple and therefore needs to be refuted with evidence.  But the first, which is even simpler, apparently does not.  Evidence for the bodyguard hypothesis would be evidence against the non-functional hypothesis, so you’re stuck with needing positive evidence for your idea before we worry about refuting it.