A new study last month found prions in the skin of patients with symptomatic sporadic CJD [Orru 2017]. The science here is good: the experiments are thoughtful and well-executed, the right controls are included, and the methods used are standard in the field. I have no quibble with any of the science reported here. I wanted to blog about this study, though, because the paper is framed as raising a possible risk of transmission of prions by surgery (other than brain surgery), and I want to emphasize some of the reasons why the new findings should not be a great cause for alarm.

The authors were careful to couch their conclusions with appropriate caveats. They quantified prions in skin and showed that they were at least 1,000 times lower than in brain. They noted that there is zero evidence for transmission of CJD by casual contact, from this or any other study. And while they cited some of the literature suggesting that CJD might be transmitted by non-brain surgery, they also noted that this literature is largely inconclusive. But some of these wise notes of caution might be lost on a reader who sees only the abstract or, worse, sees only the popular press surrounding this article. And the authors didn’t have room in the article to really go in depth into some of the reasons why the results need to be interpreted with caution.

This blog post will go into some of those reasons. The main two points I want to make concern the sensitivity of prion detection assays, which is so high as to overstate any infection risk in the real world, and the context of this study, in terms of the epidemilogical literature about surgery and CJD risk.

Sensitivity

This paper uses three different technologies to detect misfolded PrP in the skin of CJD patients:

  1. RT-QuIC, an in vitro fibrillization assay based on the conversion of recombinant PrP to amyloid
  2. Western blots for proteinase K-resistant PrP, after four centrifugation steps to enrich for misfolded PrP
  3. Intracerebral inoculation into transgenic mice expressing human PrP

Those of us working in the prion field all realize, though we may not even notice it anymore because it’s just the water we swim in, that these technologies are highly sensitive for detecting small amounts of prion material. It is amazing, actually! I consider it a gift from all the people who’ve worked in this field much longer than I have and have devoted their lives to inventing ways to isolate and detect misfolded PrP. But we need to also remember that the ability to detect prions by one of these means does not necessarily mean that there is a realistic risk of transmission of prions under normal circumstances.

Prions are measured in terms of titer — the number of infectious units in a tissue sample. And they’re measured on a log scale, because the number of infectious units is huge. For instance, the brain of a terminally sick prion-infected hamster has been estimated to contain 109.5, or about 3 billion, infectious units per gram of brain tissue [Prusiner 1982]. The operational definition of those infectious units is that you can take a gram of that brain tissue, grind it up and dilute it out 3 billion-fold, inject it directly into the brain of a healthy hamster, and still have a 50% chance of causing disease.

Whenever you hear about infectious titer measured by bioassay (inoculation into animals), which is the gold standard for measuring prions, that means the number of infectious units measured by intracerebral inoculation — injection directly into the brain. The ability to cause infection by most other routes is orders of magnitude more limited. For instance, one study in hamsters found that 105.5 infectious units, as measured by intracerebral inoculation, only added up to about 1 infectious unit if given orally [Bartz 2003]. Another study in sheep found that about 103 infectious units according to intracerebral inoculation was only about 1 infectious unit by blood transfusion [Andreoletti 2012]. So while the fact that prions are detectable in the skin of CJD patients in a laboratory setting raises a hypothetical risk of prion transmission by surgery, it doesn’t prove that the numbers actually add up to any real risk. The study quantified RT-QuIC seeding activity in skin, and found it lower than in brain, but it didn’t quantify true infectivity in skin, and we also don’t have any data on how many infectious units (defined intracerebrally) it would take to cause disease by a peripheral exposure to a trace amount of material on a cleaned (but not perfectly decontaminated) and re-used surgical instrument.

Another aspect of sensitivity is that some of these assays are sensitive to types of misfolded PrP that might not actually cause disease. For example, RT-QuIC is sensitive to various types of aggregated PrP that aren’t actually infectious, such as RT-QuIC product itself [Groveman 2017]. PK-resistant PrP can be detected by Western blot in many scenarios where infectivity isn’t present or at least hasn’t been demonstrated — various synthetic amyloids including RT-QuIC product [Wilham 2010], and misfolded conformers that turn up in cellular systems [Ma & Lindquist 1999, Ma & Lindquist 2002]. And Wenquan Zou, the senior author of this study, and Jue Yuan, one of the co-first authors, have previously found that normal control (meaning not prion diseased) brains sometimes contain small amounts of proteinase K-resistant PrP [Yuan 2006], but no one has shown that there is infectivity in those brains, much less that there is sufficient infectivity to pose a biohazard risk.

The paper does, however, show true infectivity by transmission of disease from two skin samples to transgenic mice overexpressing human PrP, and this is the gold standard for assessing the presence of infectivity. Yet first, as stated above, intracerebral inoculation will overstate, often by orders of magnitude, the titer available for infection via peripheral routes. And it’s also worth briefly touching on the details of the two transgenic mouse models they used.

One of the mouse models expresses human PrP with the two glycosylation sites mutated [Haldiman 2013]. We know that altering just one or a few amino acids can have a huge effect on which prion strains a mouse is susceptible to, how fast they get sick, and what the properties of the prions in their brains are [Telling 1995, Giles 2012, Kurt 2014, Kurt 2015, Leske 2017], so the transmission to these mice does not necessarily prove transmissibility to wild-type humans.

The second mouse model used in this study expressed wild-type human PrP. The expression level doesn’t appear to have been characterized in this paper [Orru 2017] nor the original reference [Yuan 2013] but they are likely overexpressers, since the incubation time for human prion brain homogenate was only 157 days [Orru 2017], more rapid than mice expressing normal levels of human PrP [Bishop 2010]. It’s not entirely clear to me whether overexpression increases sensitivity, per se, to low levels of prions in bioassay. Transgenic mice overexpressing PrP are widely used for titer determination in the prion field, but in a quick look at the literature, I didn’t find clear evidence as to whether they give the same answer you’d get if you did the bioassay in wild-type mice. One early seminal study performed endpoint titration in both Tga20 mice and wild-type CD1 mice, and found that the same inoculum (from Tga20 mice at 68 dpi) gave a titer of 7.9 - 8.9 log LD50 per mL of 10% brain homogenate in Tga20 mice (which overexpress PrP), but only 6.1 - 7.7 log LD50 per mL of 10% brain homogenate in CD-1 mice (which express normal levels of PrP) [Fischer 1996, compare Tables II & III]. Whether that difference is significant, I don’t know. Certainly, one can point to examples where inoculation of prions into mice overexpressing human PrP can give rise to transmission not observed at wild-type levels. For instance, chronic wasting disease prions from mule deer appear not to be transmissible to wild-type [Raymond 2007] but were readily transmitted (with 100% attack rate on first passage) to Tga20 mice overexpressing mouse PrP [Sigurdson 2007]. Or there was the report a couple of years ago of transmission of sheep scrapie to mice overexpressing human PrP [Cassard 2014], even though humans have been exposed to scrapie for centuries without a single documented case of transmission, nor any epidemiological evidence of transmission [Brown 1987, van Duijn 1998]. When Olivier Andreoletti presented that work at Prion2014, he said that his findings hadn’t turned him off from eating lamb — rather, his only conclusion was “I would not eat the brain of a sheep that died of scrapie.”

Indeed, Andreoletti’s measured reaction to his own research findings reflects what I think is a healthy attitude I’ve seen among many people involved in prion research and surveillance. Prions pose a potentially serious public health threat, so we have a mandate to study them and understand them and monitor for possible risks. Yet we know that the methods we use to study prions in the laboratory often exaggerate what the risks really are, so findings also need to be taken with a grain of salt.

Context

It’s been known for decades that CJD can be transmitted by certain specific types of medical procedure using materials from an infected donor. These include corneal transplant [Duffy 1974], electrode implantation [Bernoulli 1977], cadaveric human growth hormone (HGH) infusion [Koch 1985], cadaveric dura mater graft [Thadani 1988], and cadaveric human pituitary gondaotropin [Cochius 1990]. Also, only for variant CJD, which is peripherally acquired and has greater distribution of infectivity throughout the body than sporadic CJD, there have been a few cases of blood transfusion transmisison [Hewitt 2006]. In each of these cases, we learned about the potential for transmission because of direct evidence that a specific case of the disease arose from such a medical procedure. For instance, there was a clear link between one patient who underwent electrode implantation and later was diagnosed with CJD, incomplete decontamination of the electrodes, and two patients who subsequently were exposed to those same electrodes and developed CJD [Bernoulli 1977].

Such a direct link has never been established between other types of surgery and CJD. Instead, several studies over the past 30+ years have looked for indirect evidence. In each case, the study design is some variation on the following: find a bunch of CJD patients and a bunch of hopefully well-matched controls, and ask whether the CJD patients have undergone more surgeries on average than the controls have. Under this paradigm, some studies have reported (sometimes with multiple caveats) that there is a correlation [Davanipour 1985, Collins 1999, Ward 2002, Ward 2008, Mahillo-Fernandez 2008, Ruegger 2009, de Pedro-Cuesta 2011, de Pedro-Cuesta 2014] while others have found no correlation [van Duijn 1998, Zerr 2000, Hamaguchi 2009]. But the problem is not just that the literature doesn’t always agree with itself, it’s that even the positive results, while interesting, need to be viewed with some degree of skepticism.

This sort of retrospective, observational study design is probably the best that can be done, but it suffers from several important confounders that mean the results need to be interpreted with caution. And I don’t say any of this as a criticism! You have to remember the historical context here: the UK government denied for years that BSE, the prion disease of cattle, could possibly transmit to humans, but it did, and about 200 people have died as a result. So you could argue that it is reasonable, and some might even argue there exists a duty, to publish the evidence for or against possible sources of acquired prion disease, even if that evidence is imperfect. I don’t blame any of the researchers who published these studies — epidemiological research is hard but important — but I do want to point out a few of the reasons why a correlation between surgery and CJD may not actually mean that CJD is being transmitted by surgery:

  • Surgeries undertaken in response to early CJD symptoms. The early symptoms of CJD can be various [Rabinovici 2006] and often lead to referrals to other types of specialists — eye doctors, psychiatrists, cardiologists, and so on — before the patient finally sees a neurologist [Paterson 2012]. Anecdotally, I have heard stories of patients undergoing some sort of surgery for what was thought to be the problem, before finally getting diagnosed with CJD. Indeed, one study that specifically looked at this issue found that CJD patients were 17 times more likely to undergo minor surgeries in the year leading up to their CJD diagnosis than control individuals were [Cruz 2013]. Note that even for direct brain-to-brain instrument contact, the incubation period for prion disease is at least 18 months [Bernoulli 1977], so the only possible interpretation for the increased rate of peripheral surgeries in the single year prior to diagnosis is that they are a result, not a cause, of the patient’s CJD. The more recent epidemiological studies have caught on to this issue, and have carefully designed their analyses to exclude data from the year immediately preceding CJD diagnosis [Mahillo-Fernandez 2008]. However, several early studies included data right up to the moment of CJD onset or diagnosis [Davanipour 1985, Collins 1999, Ward 2002], and so these surgeries in response to early symptoms could be one factor driving the observed correlation between surgery and CJD.
  • Multiple testing. Many of these studies examined myriad candidate risk factors for prion disease, including a range of potential dietary and occupational hazards [van Duijn 1998, Collins 1999], and didn’t perform any multiple testing correction. For instance, one oft-cited study [Collins 1999] has a P value of 0.007 for surgery in the primary analysis, but that’s out of 14 variables tested, so the Bonferroni-corrected P value is 0.1. Even for studies that considered only surgery as a potential risk factor, multiple testing still looms in the background, because you wonder about publication bias — maybe there are other researchers out there that looked at a dozen other risk factors, got negative results, and didn’t publish them. Finally, some studies have only reported confidence intervals [Mahillo-Fernandez 2008, de Pedro-Cuesta 2014], and not P values, which is reasonable in principle, but it does make it harder to get a sense of how likely or unlikely the results are to have arisen by chance.
  • Recall bias. Many of the studies were based on interviews or questionnaires with spouses or other family members of CJD patients, compared to telephone surveys of randomly selected controls [Davanipour 1985, van Duijn 1998, Collins 1999, Ward 2002, Ward 2008]. One potential issue, though I can’t point to any direct evidence that this is a problem, is that once a person has died of CJD, family members might be more likely to recall other medical issues the person had recently. One study specifically acknowledged this as a potential confounder [Ruegger 2009]. Some more recent studies have gotten around this issue by using national registries or medical records [Mahillo-Fernandez 2008, de Pedro-Cuesta 2011, de Pedro-Cuesta 2014].
  • Imperfectly matched cases and controls. A question with any non-randomized case/control study is whether the cases and controls are directly comparable. These studies all made some effort to match cases and controls on at least basic demographic factors such as sex and age, but it is very difficult to rule out the possibility that there exists some unobserved variable that differs between the two groups [Ward & Knight 2008]. Indeed, one meta-analysis pointed out that the studies that recruited controls from the community tended to find a positive correlation between surgery and CJD, while studies that recruited controls from the hospital tended to find a negative correlation between surgery and CJD [Barash 2008].

A final summary point is that all of these studies are seeking to correlate surgery and CJD, and correlation does not imply causation. So even if all of the above issues are solved, a positive result in one of these studies still doesn’t mean that CJD is being transmitted by surgery. Indeed, at least one of the more recent studies has done a really careful job of excluding data from the year before onset, using national registry data to avoid recall bias, and making every effort to match cases and controls [Mahillo-Fernandez 2008]. But there are still ways that some hidden confounder, rather than genuine surgical transmission of disease, could drive an association.

As one example, consider the U.S., where there exists an immense disparity in health care access. Having been through it firsthand, I can tell you that getting a loved one diagnosed with prion disease is a labor of love. While they are slipping into a deep dementia before your eyes, you are taking days and weeks off of work, taking them from hospital admission to hospital admission, running test after test, and getting sent to ever higher and more geographically distant referral centers for a consultation by a more specialized neurologist. I wager that while CJD strikes without regard to your health insurance status, people with better health insurance to help pay for all these visits are more likely to be correctly diagnosed. And I wager that those same people are also more likely to have undergone more surgeries throughout their life, rather than, say, just making do with that bad hip. For this reason, if you did a case/control study in the U.S., I strongly suspect that you’d find that CJD patients have on average undergone more surgeries in their lives, even if those surgeries are in no way the cause of their CJD. I don’t know whether this exact issue could be a major problem in the studies cited here, since most of the studies were done in countries with nationalized health care and presumably less disparity in health care access. But this example at least illustrates the point that there are ways a correlation between A and B can arise even when A does not cause B.

Conclusion

I’ve argued here that the new skin study [Orru 2017], while scientifically sound, does not prove that there is sufficient infectivity in skin to pose a realistic risk for transmission of CJD through peripheral (non-brain) surgeries, and that the epidemiological studies arguing for CJD transmission via surgery suffer from several confounders and do not yet provide unmabiguous evidence of risk. Therefore, the new study is best viewed as raising the possibility of a hypothetical risk, rather than as providing an explanation for a demonstrated risk.

That doesn’t mean the new findings aren’t valuable. The possibility of risk raised here, while still hypothetical as far as we know, might still help to inform conversations about reasonable ways to reduce risk in hospital settings. And the ability to detect prions in skin might prove useful in diagnostics. Byron Caughey has pointed out that postmortem skin biopsy might be considered more acceptable to families who refuse autopsy, and so could provide at least some confirmation of CJD diagnosis. This could be important in China, for example, where autopsy is done in <1% of prion disease cases [Gao 2011]. Further studies will be required to confirm the sensitivity and specificity of skin-based diagnosis, and to determine whether there are clinical settings where skin biopsy might substitute for CSF, brain biopsy, or other tissue sampling procedures.