Read with caution!

This post was written during early stages of trying to understand a complex scientific problem, and we didn't get everything right. The original author no longer endorses the content of this post. It is being left online for historical reasons, but read at your own risk.

Gene therapy is the use of DNA or RNA to treat a disease.  Gene therapy isn’t just the outlandishly fantastic future (though it may be that too) – it’s here and now, already showing success in clinical trials for some diseases.  This post will aim to introduce the subject, current progress in the field, and the prospects for its implementation in treating prion diseases.

I’ll start by focusing on the use of gene therapy to treat genetic diseases, though it’s not limited to that and may be feasible for non-genetic prion diseases as well.  To understand the use of gene therapy, it’s important to draw a distinction between two types of genetic diseases:

  • loss-of-function: a gene doesn’t do what it should do.
  • gain-of-function: a gene does do something it shouldn’t.

Loss of function diseases are often recessive, because it often takes mutations knocking out both copies of a gene to cause a disease.  If you and your would-be partner-in-parenting get carrier screening done by Counsyl, for example, you’ll be looking for mutations that each of you carries silently in one allele but that your child would have a 25% chance of having in both alleles.  Cystic fibrosis is a common example.  Some loss of function diseases are dominant because the gene is haploinsufficient: one copy isn’t enough.  Others are dominant due to ‘dominant negative inhibition’: one mutant allele can sabotage the other one (by refusing to form dimers, causing it to misfold, etc.)

Gain of function diseases tend to more often be dominant.  Prion diseases are an example.  Gain of function diseases are still quite mysterious: we don’t know exactly what the newly gained toxic function is that makes misfolded prion proteins kill neurons, and that’s also true in Huntington’s and a number of other diseases.

If we could just go in and edit the genome in each cell of your body, we’d be in business to treat both types of diseases.  In practice, no one has yet been able to edit the genomic information already present in an organism (it’s only been done in cell culture, not living creatures).  Rather, all we can do is introduce new genomic information. Therefore these two classes of diseases lend themselves to two very different needs for treatments.  In a loss of function disease, the game is to get a working copy of the gene into the patient.  In a gain of function disease, the game is to suppress the mutant gene.

How to suppress the mutant gene?  The various approaches to this question each hijack a different existing biological mechanism, so I’ll start with an introduction to those mechanisms.

RNA interference (RNAi) is a super general term for when some pieces of RNA prevent other pieces of RNA from being translated to protein.  Confusingly, this term is used both for endogenous biological mechanisms present in all of our cells, and for humans’ attempts to hijack these mechanisms in order to treat diseases.  Also confusingly, it encompasses two distinct, though closely related, biological mechanisms: siRNA and miRNA, covered below.  The most helpful reviews for me in understanding these mechanisms were Carthew & Sontheimer 2009 and Mello & Conte 2004.  The Mello in the latter is Craig Mello, who won the 2006 Nobel Prize in Physiology & Medicine for discovering RNAi in the late 1990s.

siRNA, depending on who you ask, stands for small interfering RNA, short interfering RNA, or silencing RNA.  This pathway was the first form of RNA interference to be discovered and might represent an evolved defense mechanism against viruses.  Cells don’t normally produce double-stranded RNA (dsRNA), so when they do encounter it, it’s likely to belong to a dsRNA virus.  Here’s how cells evolved to deal with such viruses: an enzyme called Dicer recognizes dsRNA and chops it into little bits, 20-25b long (average 21b) but still double-stranded except for a 2-base overhang on the 3′ end.  This little bits are called siRNAs.  You might think that’s enough – hey, we chopped up the virus, let’s go home – but why stop there?  A protein complex called RISC – RNA induced silencing complex – recognizes these little fragments by virtue of the 2-base 3′ overhang, separates the two strands, degrades one (the ‘passenger strand’) and loads the other (the ‘guide strand’) into the protein complex.  Then this ‘activated’ complex bounces around the cell (or perhaps follows some more guided search mechanism, we don’t know yet) until the guide strand finds some complementary RNA to hybridize to, at which point an Argonaute protein cleaves that target RNA, disabling it.  This bit of genius means that when Dicer chops up one virus, it not only disables that virus but also creates guided weapons that can go out and find other copies of the virus and disable them as well.

More recently, people figured out that cells use this same mechanism for regulating their own genes as well, and not just for silencing viruses [reviewed in Golden 2008].  So what I said above was a simplification: actually, cells do produce dsRNA sometimes, either from convergent transcripts (which is when both strands of DNA get transcribed, producing a complementary pair of RNAs) or from repetitive sequences that, once transcribed, fold back onto themselves forming RNA secondary structures that contain long, perfectly base-paired double-stranded stretches.  These dsRNAs then get chopped up by Dicer and go silence specific genes via RISC / Argonaute.

microRNA (miRNA or miR) are another class of short RNAs, average 22b in length, that represent a second endogenous mechanism for gene regulation in cells.  In fact, you have to dig pretty deep into the molecular details to figure out how these are any different from endogenous siRNAs.  Here’s where miRNAs come from: you start with precursors called pri-miRNAs that are transcribed from their own genes, or from intronic RNA that got spliced out of other genes.  These RNAs form stem-loop secondary structures that get pared down by various enzymes until they are just a small hairpin loop, usually with imperfect base-pairing.  Dicer then cleaves the loop, leaving just a short piece of double-stranded RNA, and from there on the process is often the same as for siRNAs: RISC loads the guide strand and, if the guide strand finds an mRNA that is perfectly complementary, Argonaute will cleave it.  Often, though, the guide strand finds an imperfectly complementary sequence and the RISC just sits there, blocking translation but not actually degrading the mRNA.  But even for this to work, the first several bases on the 5′ end of the miRNA still need to be perfectly complementary to the target mRNA – just a one-base mismatch is enough to keep the complex from binding there.

Humans have thousands of miRNAs in our genome, and they’re numbered, like miR-21, miR-196, etc.  They seem to have really important roles in gene regulation both during embryonic development and adulthood, and might regulate as many as 60% of our genes. You can’t always predict from sequence alone whether a given miR will bind to a given mRNA, because it can only bind if the complementary sequence on the mRNA is exposed in the mRNA’s secondary structure, and our ability to predict mRNA secondary structure computationally is still imperfect.  So people still do big assays in the wet lab to see which miRNAs will bind to a given gene.

Interesting aside: both PrP [Gibbings 2012] and huntingtin [Savas 2008] have been proposed to bind to Argonaute and thus meta-regulate miRNA gene regulation.  It has also been reported that prion disease has a distinctive miRNA signature [Bellingham 2012].

So those two mechanisms – siRNA and miRNA – are how RNAi works.  Many approaches to hijacking this system for gene therapy involve yet another acronym, short hairpin RNA (shRNA).  This involves introducing into the cell a piece of RNA whose largely self-complementary sequence will favor its folding back onto itself, forming a double stranded section with a short loop at the top – a hairpin turn, if you will.  This shRNA looks enough like an siRNA or a miRNA that it gets cleaved by Dicer and loaded into RISC and can then silence mRNAs complementary to its guide strand.

When people do preliminary gene silencing experiments in cell culture, they sometimes just print up a plasmid of DNA coding for the desired shRNA, electroporate it into the cell, and then let the cell transcribe this plasmid DNA into RNA.  In vitro, this actually works.  In vivo, you can’t really electroporate cells, so people encode their desired shRNA gene into the genetic material of a virus and then infect the organism with that virus.  That’s right, in case you didn’t hear the news, viruses, humankind’s nemeses for millennia, have been co-opted for Team Good.

One virus popular for gene therapy applications is adeno-associated virus (AAV).  It’s a single-stranded DNA virus, so your cells won’t chop it up into siRNAs, and it doesn’t cause any diseases that we know of, and most people have little or no immune reaction to it.  Usually, the viral DNA genome just hangs around the cell and gets transcribed and translated just like your own DNA does.  The wild-type virus will also occasionally happen integrate into the human genome at a specific site on chromosome 19, though most gene therapy AAVs have had this capability edited out of them.  AAV is a pretty excellent viral vector for many gene therapy purposes, though it has a couple of drawbacks.  AAV’s genome is only about 4.8kb long, which is too short for some of the genes involved in loss-of-function diseases where the goal of gene therapy is to introduce a working gene rather than just an shRNA.  Another drawback is that because it usually doesn’t integrate into the host genome, it gets broken down eventually, so treatment with AAV will result in expression of the viral genes in the infected cells for something on the order of several weeks – not for a lifetime.  See Daya & Burns 2008 for a review of AAVs as gene therapy vectors.  In spite of its drawbacks, AAV is probably the most widely used virus as an experimental gene therapy vector.

A safety issue involved in any gene therapy involving a virus is the patient’s possible immune response.  While AAVs are not very immunogenic, an immune response is still possible.  By the way, adeno-associated virus is related to, but not the same thing as, adenovirus, which was the viral vector used in a gene therapy trial at UPenn in 2000 that caused a patient’s death due to a massive immune response.  One patient has also died after receiving AAV gene therapy, in 2007, though in that case it is less clear what the cause of death was and whether, or how, AAV might have caused it [Nature, New York Times].

The short lifespan of AAVs in the cell is a potential drawback of AAV as a gene therapy vector, especially in hard-to-reach tissues like the brain.  A way to get more stable long-term expression of a gene is to integrate it into the cell’s genome using a retrovirus.  Retroviruses are made of RNA but use reverse transcriptase to convert their genetic material to DNA and then insert it into the host cell’s genome.  Most retroviruses can only infect cells while the cells are undergoing mitosis, apparently because they can’t get through the nuclear membrane otherwise [Lewis & Emerman 1994].  But a special subset called lentiviruses have figured a way around this problem and are able to infect even post-mitotic cells – including neurons.  Of course, these viruses’ normal mode of action is to insert their genome into yours, trick your cell into printing up a bjillion copies of the virus, and then lyse the cell so that the new viruses may go forth and multiply.  That’s bad.  Therefore most lentiviruses used in gene therapy are modified to have just the genes they need in order to integrate into the host genome but not the genes they need to lyse the cell.  Such modified viruses are called ‘replication-deficient’.

If you look carefully at the Wikipedia page on lentivirus you’ll notice that the only lentivirus listed there that is capable of infecting humans is HIV.  As far as I can tell, most or all lentiviral vectors that being tested for human gene therapy today are ultimately modified versions of HIV [see Sakuma 2012, also these docs from OHSU and Kenyon].  The potential risks of this have not gone ignored – one concern is that the virus, even if modified to be replication-deficient, might, once introduced into the patient, somehow acquire replication competence again.  But this seems to be a relatively small concern compared to the immune risks discussed above as well as one other potential safety issue specific to retroviruses: when they integrate into the host genome, they could integrate right into the middle of an existing gene in a way that turns on an oncogene and causes a tumor [see Yi 2011 for a review of retrovirus safety issues].

Indeed, retrovirus-induced cancer has been one of gene therapy’s biggest setbacks in the public eye [see Herzog 2010].  X-SCID is an X-linked immunodeficiency caused by the loss of function of common cytokine receptor gamma-chain (gene: IL2RG), which is required for immune T-cells to function.  Without a working copy of this gene, kids die of infections before their first birthday.  A total of 20 boys were treated with retroviral gene therapy for X-SCID in the early 2000s using murine leukemia virus (MLV; a retrovirus but not a lentivirus): bone marrow was extracted from the kids, transfected with the virus, and then transplanted back in.  5 of the 20 boys got leukemia because the retrovirus (true to its name) had inserted itself in such a location as to activate a nearby oncogene.

That’s a lesson to us about the need to design safer viral vectors for gene therapy. However it should by no means be taken as a lesson to drop gene therapy.  The adverse outcomes in that trial need to be taken in context in light of how fatal X-SCID is and how poor the treatment alternatives were.  The gene therapy approach in this trial was own-tissue transplantation after gene modification – that’s just as invasive as transplantation from someone else’s tissue but virtually guaranteed not to result in transplant rejection or graft-versus-host.  Bone marrow transplants are the prevailing treatment for X-SCID, and when matching donors are available in the same family, this usually works, but when donors are unmatched or unrelated, massive complications arise in transplantation and prognosis is poor.  Most of those 20 boys treated withe gene therapy for X-SCID are still alive today – and that includes most of the ones who got leukemia (it was successfully treated with chemo). They probably would not be alive today without gene therapy.

An entirely different approach to gene therapy that doesn’t make use of RNAi or viral vectors at all is antisense oligonucleotides (ASOs).  These are ~25bp single-stranded DNAs that are designed to be complementary to an RNA you want to knock down.  They can target mRNAs in the cytosol, or they can gain entry to the nucleus, and thus can be used to target pre-splicing, immature RNAs.  ASOs work primarily by recruiting RNAse H, an enzyme in your cells which specializes in breaking down the RNA half of a DNA/RNA duplex, to degrade their unlucky counterpart.  They can also work simply by steric hindrance (fancy speak for being in the way) and thus preventing translation of a target mRNA.  The bad news is that, because cells don’t normally have single-stranded DNA lying around, they’ve evolved mechanisms to clear it.  For that reason, ASOs don’t last long in the cell.   [See Dias & Stein 2002 or Southwell 2012 for a review of ASOs].

A slight variation on that theme is antisense morpholino oligonucleotides (“Morpholinos“), a synthetic version of ASOs.  They are neither DNA nor RNA: they’re strands of morpholino, a similar-but-different nucleic acid-like molecule.  Pro: your cells have never seen it before, so they never evolved a mechanism to break it down.  It sticks around for a long time.  Con: your cells have never seen it before, so it can’t spoof the biological mechanisms of gene regulation – it doesn’t recruit RNAse H and works only by steric hindrance.  Morpholinos were invented by James Summerton and his company Gene Tools holds U.S. Patent 5506337 (issued 1996) on this technology.

And besides morpholinos, there are yet other chemical variations on the ASO theme with different chemical modifications to the DNA backbone to prevent the ASO from being broken down by the cell [reviewed in Southwell 2012].  The (lack of) longevity of ASOs is a particularly large problem because, since they’re made of DNA (or some other chemical) and not RNA, you can’t use a viral vector to express them in the cell over the long term.  A lot of research is currently focused on improving these chemical modifications to make ASOs last as long as possible.

To review what we’ve covered so far, gene therapy treatment of loss-of-function diseases today is likely to involve viral introduction of a working copy of the relevant gene.  Gene therapy treatment of gain-of-function diseases today is likely to involve knocking down the mutant gene either by ASO introduced directly into the body or by viral vectors coding for shRNA to be chopped into siRNAs.

Gene therapy’s biggest success to date has been in the treatment of Leber’s congenital amaurosis (LCA), a genetic disease that causes reduced vision or blindness.  There are actually 11 different forms of LCA, each caused by the loss of function of a different gene.  One form, caused by loss of function of RPE65, has been successfully treated using AAV vectors expressing RPE65 in a couple of different clinical trials [Maguire 2008 at UPenn, Bainbridge 2008 (ft) at UCL].  The RPE65-expressing, replication-deficient AAV is injected under the retina and within a couple of weeks it achieves significant expression of the gene.  A video hosted by NEJM from the Maguire study shows remarkable improvement in a patient’s ability to navigate an obstacle course after the surgery.  Maguire reports that the improvement was still notable 6 months later; the same group has since reported continued success after 1 year and then 1.5 years [Cideciyan 2009, Simonelli 2010].  No adverse reactions were reported in any of the studies.

LCA has become a model for gene therapy: people researching almost any other disease aspire to replicate that success.  One reason why this worked for LCA, but has been harder to do in other diseases, is that the eye is ‘immune privileged’ [reviewed in Bennett 2003 (ft)].  One thing you’d ordinarily worry about is that the first AAV injection might have a therapeutic effect but that the patient would subsequently develop antibodies against the virus and so subsequent injections would be ineffective.  None of the LCA studies on humans have done a second round of injections yet (at least nothing to that effect is published), but a study on large animals has suggested that even a second round of injections does not meet with any immune problems [Amado 2010].  So the eye might be a uniquely feasible place to administer therapeutic viruses without an immune reaction.  Immune issues aside, it also helps that the eye is on the outside of the body.  I’m not going to claim that an injection under the retina is non-invasive – actually it sounds terrifying – but it’s a lot easier to do than an injection into the brain.  Accordingly, one of the other potential bright spots for gene therapy is in macular degeneration (see 2011 New York Times Q&A; I couldn’t find any recent reviews in the scholarly literature).

But gene therapy of the brain is under heavy investigation too.  The first clinical trial of viral gene therapy of the brain was for Canavan disease, a recessive disease caused by loss of function mutations in the ASPA gene.  ASPA codes for Aspartoacylase, an enzyme that deacetylates N-acetyl-aspartate (NAA) – patients with ASPA mutations accumulate toxic levels of NAA in their brains.  The clinical trial was announced just over a decade ago [Janson 2002] and just reported long-term followup results last year [Leone 2012].  The AAV itself proved safe, though two patients had serious adverse reactions to the intracranial surgery required to deliver the AAV vector.  At followup, NAA levels were significantly reduced in patients’ brains – a good sign – and brain atrophy was slightly slowed.  In terms of patient clinical symptoms, the results were more mixed: some measures improved modestly and others not at all.  The rate of seizures was reduced, but language abilities did not improve.  Improvements in motor abilities and alertness were modest, though promisingly, the improvements were larger for the younger children in the study, suggesting that the gene therapy can be effective on these measures if administered early enough.

Another ongoing example of viral gene therapy in the brain is the use of AAV to introduce a working copy of TPP1, a gene whose loss-of-function is responsible for late infantile onset Batten disease.  When Genzyme tested this approach on a mouse model [Cabrera-Salazar 2007], the AAV was able to achieve huge levels of TPP1 expression – even far higher than even wild-type mice, let alone knockout mice.  Administration of the AAV to the knockout mice very early on more than doubled their lifespans and dramatically reduced the symptoms of Batten disease, while administration of the AAV later on, once the mice were sick, had only marginal impact.  The use of AAV to express TPP1 and treat Batten disease is now the subject of two clinical trials [1,2; introduced in Crystal 2004] at Weill Medical College (Cornell’s NYC campus).  In these trials they are injecting AAV directly into the ventricles of affected children – you can get an idea of how invasive this procedure is from the description in the second clinical trial:

we propose to perform 2 series of 6 simultaneous administrations of vector for 75 min each. Each subject will receive the assigned dose of AAVrh.10CUhCLN2, divided among 12 locations delivered through 6 burr holes (2 locations at 2 depths through each hole), 3 burr holes per hemisphere.

Results from this study are not in yet.  The plan was to treat several severely affected children first, and then some more moderately affected children later.  A potential confounding factor in the interpretation of these (and other) CNS gene therapy studies is the fact that, because gene therapy is experimental and considered risky, the most severely affected individuals are usually chosen to participate in the studies first.  It is possible that treatments may fail to show much efficacy in the most severely affected patients, even if they would have made a big difference in moderately affected patients – consistent with Cabrera-Salazar 2007‘s finding that AAV gene therapy is more effective when administered before symptoms emerge.

Because this is and fatal familial insomnia is a genetic disease of the brain, I’ve focused here on gene therapy for genetic diseases, especially neurological ones. But it’s worth mentioning that gene therapy is not limited to these parameters.  There are a variety of gene therapy approaches under investigation as cancer treatments and there is even a journal called Cancer Gene Therapy.  And the use of zinc finger nucleases to delete CCR5 from patients’ CD4+ cells, making them resistant to HIV, is also considered a form of gene therapy (read more in the TALENs & ZFNs post).  A review of AAV gene therapy can be had in Mingozzi & High 2011 and a list of current trials is provided in Table 1.

Aside from AAVs, there are also ongoing clinical trials of brain gene therapy using ASO.  One clinical trial using ASOs to knock down mutant SOD1 in familial ALS is infusing ASOs into the cerebrospinal fluid for 12 hours.  Results are not published yet, but a preliminary report presented at a conference said there were no adverse events, but “neurological exams are consistent with ALS disease progression” – i.e. no sign of efficacy so far.

ASOs have shown a lot of promise in mouse models for Huntington’s Disease [Kordasiewicz 2012] and may be headed for a clinical trial in humans soon (see HDBuzz post).  Gene therapy of Huntington’s is especially challenging.  Since HTT knockout mice die before they are born [Duyao 1995], we believe that having at least one working copy of HTT is necessary for our survival.  Therefore gene therapy for HD needs to be allele-specific, knocking down the mutant allele while letting the patient’s normal allele continue to do its job.

Kordasiewicz started with the BACHD model of Huntington’s Disease – these mice express a mutant human HTT transgene in addition to their own 2 copies of the mouse Htt gene.  The ASOs, infused continuously into the ventricles for two weeks, were able to reduce mutant human HTT expression to about 38% of its original level, though the protein level lagged and never dropped below about 67% of its original level.  That first experiment used 20nt ASOs that were specific to human HTT mRNA, so the mouse genes weren’t affected.  But they also did another experiment with an ASO that would bind both human and mouse huntingtin mRNA, and knocked them down similarly to 17-31% of original levels.   A single infusion of ASO kept huntingtin mRNA levels low for about three months, and levels didn’t rise back to 100% of normal until four months.

So the ASOs worked – they didn’t knock down mutant huntingtin completely, but they substantially reduced its levels.  Before we move on, let’s return for a second to the method of ASO delivery: continuous infusion into the brain for two weeks.  That sounds like a pretty big deal: yet worse than the one- or two-time infusions of AAVs that were used in the Batten and Canavan clinical trials, and even those had some adverse reactions.  So far this is sounding less-than-awesome.

However, the therapeutic results were really good – good enough to make you think a two-week infusion into the brain might be worth it.  Kordasiewicz was admirably thorough, showing efficacy of ASOs in three HD mouse models – BACHD, YAC128, and R6/2.  The ASO reduced or delayed symptoms when administered presymptomatically, and prolonged the survival of the R6/2 mice.  What’s more incredible, when ASOs were administered to already-symptomatic mice, it didn’t just slow their decline, it actually reversed their symptoms on several measures, bringing them back up to the level of wild-type mice in terms of motor function.  And one thing I find really remarkable is that, in the BACHD mice at least, the phenotypic improvements were sustained even 9 months after ASO administration, long after the ASOs had been degraded in the brain and mutant huntingtin expression had returned to its pre-treatment levels.  It is as though the treatment gave the brain time to clear the damage from mutant huntingtin, resetting the disease clock.

Promisingly, Kordasiewicz also found that ASOs suppressing the wild-type mouse huntingtin in addition to the mutant human huntingtin seemed to be well-tolerated by the mice.  In one experiment, the BACHD mice had their mouse huntingtin mRNA levels reduced to 25% of normal levels for about 4 months, and yet still had phenotypic improvements thanks to the knockdown of the mutant human gene.  This suggests that maybe allele-specific silencing needn’t be perfect in order for gene therapy of HD to be viable.

Finally, the study also demonstrated feasibility of ASO knockdown of huntingtin in rhesus monkeys.  The rhesus were just wild-type, not HD models, and the goal was just to measure the knockdown of huntingtin mRNA to demonstrate feasibility and examine any adverse effects.  Instead of the intraventricular infusion used in the mice, the monkeys received ASOs intrathecally – i.e. at the base of the spine, just under the arachnoid membrane of the spinal cord.  Infusion here is further from the relevant brain tissues than intraventricular infusion but is still in the cerebrospinal fluid and thus bypasses the blood-brain barrier, which ASOs cannot cross.  Kordasiewicz found that the ASOs did indeed reach the brain, though troublingly, the striatum – the region most affected in HD – received relatively little ASO compared to other brain regions, and still expressed 75% of pre-treatment levels of huntingtin.  Still, the demonstration that this less invasive (compared to injections into the brain) method of administration could work at all in non-human primates was an important step in setting the stage for a clinical trial.  And the monkeys also tolerated the knockdown well, suggesting humans too will be likely to tolerate knockdown of their normal huntingtin allele.

Compared to that groundbreaking study in HD, no study has yet brought us quite so close to clinical feasibility of gene therapy for prion diseases.  Nevertheless, a rough picture has begun to emerge of what such therapy might look like.

There are two different potential applications for gene therapy in prion diseases.  In asymptomatic carriers of genetic prion diseases like fatal familial insomnia, gene therapy could be used to knock down, as specifically as possible, the mutant allele.  In patients with full-blown prion diseases – whether genetic, sporadic or variant – gene therapy could be used to knock down both alleles of PrP.  The available evidence suggests that, provided that gene therapy could achieve sufficient knockdown of PrP, these approaches would actually work.

First, there can be no prion disease without PrP expression.  Ever since the first PrP knockout mouse was created twenty years ago, we’ve known that prion infection can’t take hold without PrP expression [Bueler 1993].  Not only do PrP knockout mice not develop prion infections, they actually clear PrPSc injected into the brains within a few days.  Even the heterozygous mice, with only one copy of Prnp, had approximately doubled incubation times upon prion infection, compared to normal mice.  What’s more, PrP knockout mice don’t even show any acute toxicity from the PrPSc injection, even at large doses.  This was shown even more convincingly by Brandner 1996.  Brandner grafted normal, PrP-expressing brain tissue into PrP knockout mice and then injected them with prions.  The grafted tissue generated lots of PrPSc and even secreted it into the rest of the brain, yet no neurodegeneration was seen in the non-PrP-expressing tissues.  This has been taken to mean that it’s not that PrPSc is itself toxic, it’s that the conversion of PrPC to PrPSc is toxic.

Second, even if we can’t achieve complete knockout of PrP, even knockdown may be good enough.  This was demonstrated in an awesome series of experiments by Tremblay 1998.  Tremblay created mice conditionally express PrP so that the gene could be effectively “turned off” after scrapie infection had already started.  I find conditional gene expression to be one of humankind’s most awe-inspiring scientific achievements, so I’m going to digress for a moment.

We humans have been using doxycycline and other tetracycline antibiotics for about 50 years now to treat all sorts of things like chlamydia, acne – we even feed it to our livestock.  Tetracyclines work as antibiotics by being a wrench in the gears of bacterial ribosomes and thus blocking protein translation (see Wikipedia; and FYI we mammals have different enough ribosomes that tetracyclines don’t bother us.)  Bacteria are no slouches at evolving, and they quickly evolved tetracycline resistance genes.  Specifically, they evolved what we call the Tn10 tetracycline resistance operon.  One DNA-binding protein called tetR (R for repressor) is always expressed and it normally binds to the promoter of the operon, turning off the expression of the genes inside the operon.  When tetracyclines are introduced into the cell, they bind to tetR and induce conformational change so that it no longer binds to DNA, and thus the resistance genes in the operon begin to be transcribed.  Those resistance genes might include transporters that sit in the membrane and pump tetracyclines out of the cell, or other proteins that protect the ribosomes themselves.  Several groups tried to figure out how to harness the Tn10 operon to create conditionally expressed genes in mammalian cells.  Gossen & Bujard 1992 (ft) found by far the best solution: they created a fusion protein with the tetracycline-binding and DNA-binding parts of tetR and the transcription-activating part of VP16, a viral protein that herpes simplex virus uses to turn on expression of its own genes.  This fusion protein binds tetracycline just like tetR and also binds the Tn10 operon promoters (called tetO, O for operator) like tetR but, thanks to that piece of VP16, it activates rather than repressing gene transcription.  This new fusion protein, called tTA (tetracycline transactivator) turns on genes under the tetO promoter, but lets them shut off  again when it’s bound by tetracyclines.

This system is now called a Tet-off switch and has been used to study dozens of genes and diseases.  The fact that humans commandeered genes from antibiotic-resistant bacteria and herpes virus in order to make this possible fills me with hope and admiration for the power of science.

Back to the subject of prion gene therapy.  Tremblay 1998 put mouse Prnp under the control of a Tet-off switch, raised mice expressing PrP, then knocked down PrP expression by administering doxycycline, and infected the mice with scrapie.  The small residual amount of PrP expression was enough to allow the mice to continually produce low levels of PrPSc, but they didn’t show any behavioral or histopathological signs of prion disease.  So apparently a low rate of PrPC-to-PrPSc conversion not too intolerably toxic, and at this low level animals can clear PrPSc about as fast as they create it, since accumulation was not observed.  Tremblay’s mice normally expressed PrP at about twice the level of wild-type mice, but expressed only ~15% the PrP of wild-type mice when treated with doxycycline, suggesting that in humans, perhaps an 85% knockdown of PrP would be enough to halt the symptoms of prion disease.

A slight confounder here is that doxycycline itself has also been shown to have some therapeutic value against prion disease.  But effect sizes in those studies were small, at least for established prion infections in the brain, so we can probably assume that most or all of the effect seen in Tremblay’s study was due to turning off PrP expression.

Tremblay only looked at turning off PrP expression before prion infection, leaving open the question of whether this would work once prion infection had begun.  Mallucci 2003 finally answered this question using a different conditional gene expression system based on Cre recombinase.  Cre recombinase is a cool technology that we stole from bacteriophages.  It clips out a piece of DNA ‘floxed’ (bookended) by two particular binding sites.  Mallucci engineered a mouse with ‘floxed’ Prnp as well as Cre under the promoter of NFH, a gene not expressed until mouse adolescence.  Therefore the mice expressed prion protein for the first 10-12 weeks of life and then turned on their Cre recombinase genes which knocked out prion protein.  Mallucci infected the mice with prions early in life, giving them plenty of time to develop multiple signs of prion disease - PrPSc accumulation, astrocytosis, spongiform degeneration – before Cre was activated, turning off PrP expression.  Not only did these mice survive, they actually appeared to reverse the spongiform degeneration that had started before PrP expression was turned off, suggesting that at least some degree of recovery from the damage of prion disease is possible.  Mice (and humans) don’t grow too many new neurons once they’re adults – neurogenesis is limited to a few brain regions – so it would be optimistic to think that patients could fully recover from really advanced prion diseases when many neurons are already lost.  But Mallucci’s results at least show that halting prion infection is possible if PrP is knocked down enough, and that some degree of recovery from at least the early stages of symptomatic disease may be possible.

In both Tremblay’s and Mallucci’s studies, PrP was kept turned off for as long as the mice lived.  I don’t know of any studies where PrP expression has been turned back on again once the mice had recovered.  Because PrP knockdown was not absolute in either study, and some PrPSc was produced, it seems likely that re-expressing PrP would have started the disease course again.  In humans with full-blown prion diseases, odds are that PrP would have to be knocked down for good in order to reverse the disease.

This raises the question of what would happen to people without PrP.  All of us are walking around with some of our genes inactivated due to rare loss-of-function mutations [MacArthur 2012] but so far we’ve never found anyone with both copies of PRNP inactivated, so we don’t really know what the human phenotype of PrP knockout (or extreme knockdown) would be.  We have to guess based on what we see in knockout animals we’ve created.  Bueler’s mouse showed us that PrP knockout is viable – in fact, at first the knockout mice appeared no different from wild-type mice at all, though studies since then have revealed a variety of PrP knockout phenotypes [reviewed in Steele 2007].  PrP knockout mice have disrupted sleep, and are more vulnerable to strokes and seizures.  None of those are happy things, but are clearly better than dying of prion infection.  For a while after Bueler’s knockout mouse, some scientists speculated that PrP knockouts were only viable because, during embryonic development, they were able to somehow adapt to not having PrP, whereas turning off PrP during adulthood would be fatal.  Tremblay and Mallucci’s studies have pretty much put an end to that line of thinking.

On the other hand, clearly, if you’re a genetic prion disease carrier and you know it, it would be preferable to turn off your mutant allele and keep expressing your healthy allele if possible.  There is no guarantee that this would prevent prion disease, but the fact that prion infectivity is so tightly associated with PrP expression levels (knockout heterozygotes have long incubation times [Bueler 1993] and mice overexpressing PrP have increased vulnerability to prion infection [Fischer 1996]) certainly suggests that a lower dose of mutant PrP would reduce risk and delay onset.

Knocking down mutant PRNP while keeping expression of the healthy allele is what’s called allele-specific gene silencing.  That’s a challenge since most disease-causing PRNP mutations are single SNPs.  The tools we have at hand for gene therapy – basically, shRNA or siRNA introduced via viral vectors, or ASOs – all involve antisense hybridization to the PRNP mRNA, and antisense strands of nucleic acids have a tendency to still bind even if one or two bases don’t match.  So you could design an ASO or an siRNA that matches your disease allele exactly, and it would indeed knock that allele down, but it would also knock down your healthy allele by a considerable amount as well.

But allele-specific gene silencing, even with just one SNP to discriminate between alleles, may not be impossible.  It has been reported that siRNAs require absolute matching in bases ~2 through ~7 in order to do their job, but that they can tolerate a great deal of mismatch elsewhere in the sequence [Birmingham 2006].  The upside is that by designing an siRNA with the disease-causing SNP somewhere in bases 2-7, you may be able to prevent knockdown of the healthy allele.  The downside of Birmingham’s finding is that much of siRNA specificity is determined by just those six or seven bases, which means they’re not as specific as we’d like, and your anti-PRNP siRNAs may end up knocking down other genes too, with nasty side effects.  One possible solution to that problem is “siRNA redundancy” – instead of delivering a large dose of one siRNA, to design multiple siRNAs against PRNP, and deliver each one in smaller doses, so that off-target effects are reduced but the collective effect on PRNP is still large.  This and other approaches are reviewed in Jackson & Linsley 2010.

So far I have made the case that sufficient knockdown of PrP, if it could be achieved with gene therapy, would be able to prevent or reverse prion disease.  But this post has also given a sense of some of the potential challenges of gene therapy, such as potential immune responses, reaching the relevant brain tissues without severe surgical complications, obtaining a sufficiently large knockdown of the targeted gene.  In short, the principle is sound, but the implementation is nontrivial.  In terms of demonstrating actual realistic feasibility, prion diseases aren’t as far along as Huntington’s Disease looks to be, but several experiments have already helped to pave the way.

White & Mallucci 2009 review the progress to date.  Not long after the RNAi pathway was discovered, one basic experiment simply showed that RNAi against PrP was capable of reducing PrP expression: Tilly 2003 expressed mouse and sheep PrP in rabbit kidney cells, transfected them with RNAi plasmids and was able to reduce PrP expression to just 4-19% of original levels when a perfectly matched RNAi sequence was used; when a sequence with four mismatches was used, PrP expression was still slightly reduced, to 85% of original levels.  OK, so RNAi against PrP is feasible and potentially quite effective.  Golding 2006 went a step further, virally introducing anti-PrP shRNA into goat and cow cells, knocking PrP down to ~10% of original levels.  Golding then incorporated the cells into blastocysts, suggesting the feasibility of creating transgenic livestock resistant to prions (though this could also be done via knockout).  Daude 2003 (ft) also showed that, as you’d expect, knocking down PrP also reduces PrP-res formation: RNAi against mouse PrP in ScN2a (scrapie-infected mouse neuroblastoma) cells resulted in less PrP-res production.

So far, those were all cell culture experiments.  The first study of anti-PrP RNAi in live animals was Pfeifer 2006.  Pfeifer created lentiviruses expressing anti-PrP shRNAs and raised chimeric transgenic mice expressing these shRNAs in some – but not all – cells of their bodies.  Promisingly, the few highly chimeric mice expressing the viral shRNA in at least ~2/3 of their cells lived ~50% longer than controls, even though the less chimeric mice didn’t live longer.  That’s promising because it says that if you can achieve lentiviral shRNA expression, you can increase survival.  But that’s a big ‘if’, and Pfeifer’s data suggested that if the virus only got into a few cells, that wouldn’t help too much.  Pfeifer had also injected lentivirus into adult mouse brains and Figure 1 shows lentiviral integration based on a GFP reporter gene.  Yet Pfeifer chose to base the main survival experiment of the paper on the chimeric mice rather than on this lentiviral injection model, even though the latter is clearly a far more accurate model of what would happen in real human use of gene therapy.  Why that decision?  Probably because, when injected, the virus only got into a very small fraction of cells in the brain, and Pfeifer suspected (or tried, and learned, but didn’t write in the paper) that this would not significantly prolong survival.

If I was reading that in 2006, I’d be fairly worried that gene therapy for prion diseases, though theoretically feasible, might be practically difficult just because it is hard to get viruses into enough parts of the brain.  But not so fast: White 2008 did a successful experiment based on viral injection into adult mouse brains.  I like this experiment because the therapy wasn’t administered until prion infection was pretty advanced – in this way it was an accurate model of the stage of intervention that is available for sporadic CJD, which comprises the majority of prion disease cases.  At 56 days post-infection (dpi), White injected anti-PrP shRNA lentivirus into both hippocampi of the mice.  Just 1 week later, the control mice started to show symptoms of prion infection (decreased burrowing), while the lentivirally treated mice didn’t. And the treated mice lived ~20% longer (avg. 105 dpi vs. 88 or 85 dpi for two control groups).  This is impressive because the lentiviruses were only injected into the hippocampus – most of the brain was totally untreated.  The key is that the prion strain that White used (RML prions) preferentially affects the hippocampus.  White speculates that the hippocampus might therefore represent a key early area from which prion infection spreads to the rest of the brain.

The same authors’ later review [White & Mallucci 2009] argues, therefore, for the importance of both space and time in gene therapy for prion diseases.  If we can target a key brain region (different for each prion strain – for instance the thalamus in FFI) during a window of intervention in the early symptomatic phase, then we may be able to slow the disease down even if we can’t reach every neuron.

If true, that’s a promising idea, because the difficulty of delivering viruses to the brain seems to be perhaps the most challenging aspect of gene therapy.  It’s quite clear at this point that (1) RNAi can reduce PrP expression by a lot – down to 10 or 20% of original levels, and that (2) reducing PrP expression delays or even reverses disease.  There is no question that the remaining challenges lie in safe and effective delivery.   ‘Safe’ here means avoiding immune response, minimizing the risk of cancer from transgenes turning on oncogenes, minimizing off-target knockdown effects against other crucial genes, and minimizing infections and other surgical complications from the need to deliver the viruses (or ASOs) directly into the brain or at least the cerebrospinal fluid.  And ‘effective’ means reaching enough neurons and having the viral genes (or ASOs) stick around long enough.

With that in mind, here’s why I’m an optimist about gene therapy: there is almost nothing disease-specific about these problems of safe and effective delivery.  The biggest remaining hurdles to gene therapy today are universal enough that progress in one disease means progress for all.

Contrast this with small molecule drug discovery.  Here in the U.S., the FDA approves on the order of 20 novel molecules as drugs each year [FDA 2010], and most of those drugs will have spent 15 years in development prior to approval.  Meanwhile, NIH’s Office of Rare Diseases Research database lists 7,000 rare diseases.  Does this mean it will take 350 years to develop approved treatments for all of these rare diseases? No.  It means we need a paradigm shift.  That paradigm shift might ultimately involve something new that we can’t begin to imagine yet, but gene therapy looks like the best candidate so far.  If we can solve a few fundamental problems of delivery, we’ll have unlocked the treatment for virtually every genetic disease – and a pile of non-genetic diseases too.  This should be among our top priorities as a species.

My optimism is focused on the slightly longer run, because both the science and the regulatory environment still have a ways to advance.  Today, U.S. and E.U. regulations require that every gene therapy product receive separate regulatory approval, making them no different from drugs in terms of regulatory burden [see C. Bernard's commentary in Nature, 2012].  Gene therapy is a much more modular approach than drug development – the same viral vector or ASO delivery approach can be adapted to target a different gene simply by changing the DNA sequence, and even within treatments for one disease, sequences will sometimes need to be customized for patients’ individual polymorphisms.  Right now, every distinct sequence to be used in gene therapy would require separate FDA approval, a daunting prospect.  This seems to negate much of the benefit of the modular nature of gene therapy.  Yet there is validity to it as well: changing just one base in an siRNA or ASO could alter the off-target effects in a way that could, rarely, make the difference between minimal side effects and lethal side effects.

I do think that regulatory treatment of gene therapy will need to change eventually, and although it would be nice if it did so now so as to encourage development of gene therapy, it is also probably a fact that the science of gene therapy needs to advance more in order to show that it deserves better regulatory treatment.  With the possible exception of the AAV treatment for Leber’s congenital amaurosis, no gene therapy treatments have really shown enough evidence of efficacy and safety yet to be considered more than an ‘experimental’ treatment.  Certainly not in the central nervous system.

But that’s where we need to go.  It’s too early to place any bets as to whether gene therapy will succeed, but it certainly strikes me that, with the potential cure for every genetic disease at stake, we as a species can surely be clever enough to solve the problems of safe and effective delivery of gene therapy vectors.