That should sound impossible, but it's actually a very mature (and IMHO very underappreciated marvel) technology: pregnancy test kits (and follow-ons for ovulation). That's close to the gold standard in diagnostics: cheap, useful and pretty accurate. Such tests had a major impact on society, empowering women to monitor this important component of their health condition without requiring a visit to a doctor.
A severe challenge for any sort of genomic medicine is to approach this sort of simplicity. The MinION, for example, represents a huge change in the all-in cost and portability of DNA sequencing, but still requires the purification of DNA and multiple laboratory steps as well as a computer and specialized software to analyze the results. A race is on to create simple protocols to answer simple tests. Can nucleic-acid diagnostics be reduced to a straightforward instrument of modest price which generates results quickly, rather than something sent to centralized labs with highly-trained personnel and complex equipment.
Today, a new approach to rapid nucleic acid diagnostics developed at UC Berkeley (and licensed to a company called Nanosens) was published (and had been sent to me in advance under embargo), describing an electronic scheme to detect target molecules via nuclease-inactive Cas9 (aka dCas9). The "CRISPR-Chip" device described in Haijan et al relies on graphene-based transistor elements to sensitively detect native DNA; no amplification steps are used. This device is sensitive to local ionic environment, so a highly charged nucleic acid with its coterie of counterions perturbs that environment sufficiently to generate a signal. A specific chemistry is used to tie dCas9 to the graphene surface.
To assess sensitivity, the group used HEK293 cells with a single-copy blue fluorescent protein transgene (HEK-BFP). Over a range of 300 to 1800 ng of input DNA they show a linear correlation between amount of input and the signal from the device. Signal builds quickly to about half maximum in under 10 minutes (Figure 5D, 5F, 5H). It's too bad they didn't further explore the regime between 0 and 300 to probe the lower limit of detection. Also missing from this plot is the negative control of just HEK293 DNA. No explicit fragmenting of the DNA was performed; just whatever happened during purification and pipetting onto the device.
A set of experiments were used to explore the effects of mixing the transgene-bearing HEK-BFP with wildtype DNA. These experiments are a bit frustrating for what is missing and the odd effects shown, as well as some poor plotting choices (for this paragraph, I'm reviewer #3!). Figures 5 C-H all have the same units on the Y-axis and have a minimum value of 0, but have maxima of 200, 200, 100, 350, 120 and 300 respectively. Several figures show that a signal is still detected when varying amounts of standard HEK293 are spiked in, though the spike-in either causes noise in the assay (Figure 5E & 5G) or greatly affects the signal (5F). Figures 5F and 5H both show real-time signal and that the signal can be significantly attenuated (but not eliminated) by washing. Unfortunately, the static measurements in 5E, 5G and 5I don't have times reported on the figure. But HEK293 DNA alone (Figure 5B) doesn't generate much signal -- but again this is a fixed timepoint rather than a time curve. It's a confusing set of figures with insufficient labeling and controls. One caveat on my complaints is that I didn't notice that I hadn't been sent the supplementary materials until it was much too late -- but conversely anyone with the role of sending out advance copies should know that they are often essential and that the "Supplemental" description is quite unfortunate ("Expanded" would be far better; "Essential but left out due to publishing constraints" perhaps more honest but a bit wordy)
Now, for detecting germline mutations those mixing experiments are perhaps not the most relevant. For this, they designed two guide RNAs to detect two different common deletion alleles causing Duchenne's Muscular Dystrophy (DMD), an X-linked disorder primarily affecting males. Buccal swab samples from three healthy and six affected individuals were used. In these samples, a mostly clear separation of deletion and wild type alleles could be achieved, though variability in the wild type signal did push one unaffected sample close to the detection limit and one diseased individual deletions would have been called for both exons even though one is a false result. As a proof-of-concept study these sort of results suggest further exploration is warranted, but the assay variability suggests that much, much larger panels of samples will be required. My complaints about Y-axes and the failure to report timepoints for non-timeseries plots apply here (Figure 6) as well
So it's an imperfect paper, but does suggest that this method could work. It requires purified DNA, but extracting DNA from buccal samples is well-established. There is some talk in the paper of multiplex probes, though this isn't explained and I suspect they really mean (as shown in the paper) multiple probes on separate chips or perhaps subdividing the chip into regions. As discussed in the paper, this method would be most applicable to detecting large insertions or deletions -- an important class of disease-causing mutations but certainly only a fraction of all mutations.
Going back to my pregnancy test standard, the proposed device wouldn't be as inexpensive and certainly would require power and some sort of box -- but probably a very small one requiring little power. Upstream DNA purification would be required. So this isn't a simple dipstick test, but is the sort of device which could conceivably be inexpensive and compact enough to be placed in large numbers of hospitals and specialized clinics for genetic disease.