Red, theory; black, fact.
No junk DNA?
The junk-DNA concept is quite dead, killed by the finding that the noncoding sections (sections that do not specify functional proteins) have base-pair sequences that are highly conserved in evolution and are therefore doing something useful.
What does long non-coding RNA do?
This DNA is useful because the RNA transcripts made from it are useful, serving as controllers of the transcription process itself and thus, indirectly, of protein expression. (Changes in protein expression may be considered the immediate precursor of a cell's response to its environment, analogous to muscle contractions in an intact human.) Small noncoding RNAs seem to be repressors of transcription and long noncoding RNAs (lncRNA) may either repress or promote. (lncRNA molecules also control chromatin remodeling, but this may be an aspect of stem-cell differentiation during development.) Despite the accumulation of much biochemical information, summaries of what lncRNA seems to do for the cell have been vague, unfocussed, and unsatisfactory (to me).
Control of gene expression: background
The classical scheme of protein expression, due to Jacob and Monod, was discovered in bacteria, in which a signal molecule from the environment (lactose in the original discovery) acts by binding to a protein to change its conformation (folding pattern). The changed protein loses the ability to bind to DNA upstream from the sequence that specifies the lactase enzyme, where it normally acts to block transcription. The changed protein then desorbs from DNA, which triggers transcription of lactase messenger RNA, which is then translated into lactase enzyme, which confers on the bacterium the ability to digest lactose. Thus, the bacterium adapts to the availability of this food source.
Since I have a neuroscience degree, I naturally wonder if all this can be modelled in neurobiological terms. Clearly, it's a reflex, comparable to the spinal reflexes in vertebrates. An elementary sensorium goes in, and an elementary response comes out. But vertebrates also have something higher than spinal reflexes: operations by the brain. (Don't worry, I am not about to go off the deep end on you.)
My neuron-inspired theory of long non-coding RNA
I propose a coordinating role for the noncoding RNAs: rather than relying on a bunch of independently acting reflexes, eukaryotic cells can sense many promoter signals at once, as a gestalt, and respond with the expression of many proteins at once, as another gestalt. You do not need an entire brain to model this process, just one neuron. The synaptic inputs to the dendrites of the neuron can model the multiple promoter activations, and the eventual output of a nerve impulse (action potential) can represent the signal to co-express a certain set of proteins, which is hard-wired to that metaphorical neuron by axon collaterals. In real neurons, action potentials are generated by a positive feedback between membrane depolarization and activation of the voltage-gated sodium channel, which causes further depolarization, so around we go. This potential positive feedback can be translated into terms of molecular biology by supposing a cyclic, autocatalytic pattern of lncRNA transcription, in which each lncRNA transcript in the cycle activates the enhancer (which is like a promoter) of the DNA of the next lncRNA in the cycle. The neuron model suggests that the entire cycle has a low level of baseline activity (is "constitutively active" to some extent) but the inhibitory effect of the small noncoding RNAs (analogous to what is called the rheobase current in neurons) suppresses explosive activation. However, when substantially all the promoters in the cycle are activated simultaneously, such explosive transcription does occur. The messenger RNA of the proteins to be co-expressed as the coordinated response is generated as a co-product of lncRNA hyper-transcription, and the various DNA coding regions involved do not have to be on the same chromosome.