ALS and ASOs

ALS and ASOs

Amyotrophic lateral sclerosis, or ALS, feels harder to develop a view on than something like Huntington's, partly because the discussion seems particularly broad when it comes to speculation on causal triggers. In my earlier Huntington's article, the expanded CAG repeat gave a clear shared starting point, even while the steps connecting it to neuronal death remained complicated. With ALS, we can identify a disease-causing mutation in some patients and recognize similar protein pathology across many others, while still having difficulty explaining how each person arrived there. And given the sophistication of modern measurement techniques in many regards, if answers still remain elusive, it makes me wonder if the correct expectations and assumptions have been made in the first place. [1,2]

I think we tend to picture the search for a root cause as a funnel, with one initiating event or a few opening into a widening set of consequences, but perhaps, for certain diseases like ALS and Parkinson's, the funnel runs the other way. A large number of disturbances could converge on a smaller set of cellular failures, which eventually converge again on the loss of motor function that makes the disease recognizable. In that case, the similarity in the symptoms may tell us more about the machinery that failed while leaving all the possible initial triggers as an open set.

A car can be damaged by hail, another car or even a moose that takes offense to its existence, and some of the resulting dents and broken glass may look similar. Because the car's construction and materials constrain how those different events express themselves, I imagine something comparable could happen within a neuron, where many kinds of disturbance encounter the same requirements for energy, transport and maintenance. We might become increasingly good at describing the damage while still finding that its origins differ from one person to the next, based on their unique lives and circumstances. In this case, then, maybe assigning tremendous focus to finding a single root cause or a few broad population-based root causes may pay off less than expected.

And that is not to say that we do not know a lot about ALS. ALS is a disease which progressively damages the motor neurons that control voluntary movement. Upper motor neurons carry signals from the brain toward the brainstem and spinal cord, while lower motor neurons connect those systems to muscles. Their degeneration can produce a combination of weakness and muscle wasting, alongside stiffness and exaggerated reflexes. As the disease spreads, speaking, swallowing and breathing can become difficult. Death from respiratory failure commonly occurs within three to five years of symptom onset, although some patients progress much faster and others survive substantially longer. ^[1]

The first affected muscles also vary, with roughly 70% of patients presenting with weakness in a limb, about 25% with difficulties involving speech or swallowing, and a smaller group with initial trunk or respiratory weakness in the clinical overview referenced here. These symptoms require different kinds of support, even before we consider the molecular differences underneath them. Respiratory assistance, nutritional support and ways to preserve mobility and communication remain central to care alongside attempts to slow the disease itself. ^[1]

ALS also overlaps with frontotemporal dementia, or FTD, which can affect behavior, personality and language. Related molecular disturbances can produce different clinical expressions depending on the brain regions and cell populations involved. Also, because some families include people with ALS, others with FTD, and others with features of both, even the boundary around the disease name can cut across a biological process that extends beyond motor neurons. ^[2]

Approximately 90–95% of ALS cases occur in people with no known family history and are classified as sporadic. Some of these patients may still carry disease-causing changes in their genes such as SOD1 or C9orf72, while others are thought to inherit a combination of genetic changes that seem to collectively increase their risk. There is also increasing exploration around aging and non-genetic cellular dysfunction as causal triggers. [1,2]

And each hypothesis introduces more complexity. For instance, ALS incidence generally peaks in later adulthood, around ages 60–75 in the studies discussed by the PeerView clinical overview. An inherited vulnerability might take decades to become consequential, but changes in cellular maintenance might make an earlier disturbance harder to contain. The existence of younger patients suggests that the time required to reach a similar state can differ substantially, depending on the vulnerability and the processes acting on it. And there are even cases, Stephen Hawking's, for instance, where earlier onset did not preclude a relatively long life. ^[1]

More concretely, over 95% of ALS cases involve abnormal behavior of TDP-43, an RNA-binding protein encoded by the TARDBP gene, though it is worth noting that SOD1-associated and FUS-associated ALS are important exceptions to that pattern. In affected cells with TDP-43 pathology, the protein often becomes depleted from the nucleus and accumulates in the surrounding cytoplasm, sometimes forming protein deposits. ^[2]

Understanding why that movement matters requires some explanation of what RNA is and what TDP-43 does with it. RNA is a chain of nucleotides, usually represented by the letters A, U, G and C, and messenger RNA is one kind of RNA. A cell copies information from DNA into an initial RNA transcript, processes that transcript, and can export the resulting messenger RNA from the nucleus into the cytoplasm. Then ribosomes read its sequence to assemble a protein from amino acids.

Part of RNA processing is splicing, where the cell removes stretches called introns and joins retained stretches called exons. Proteins help determine where those cuts and joins occur, and TDP-43 participates in this regulation by suppressing the inclusion of inappropriate RNA segments that would disrupt the resulting message. Losing nuclear TDP-43 can therefore change the proteins a neuron is able to produce, while abnormal TDP-43 accumulating elsewhere can create additional problems such as potential cell toxicity. The implication here, then, is that any treatment aimed at removing accumulated TDP-43 also has to preserve enough of it in the nucleus to maintain normal RNA processing. [2,3]

Another pathway to consider, perhaps upstream of TDP-43 migration, is RNA conformation. RNA will fold because parts of its nucleotide chain can pair with other parts of the same chain, which can leave a sequence present but less accessible to a protein within that folded structure. Whether TDP-43 can bind a particular sequence therefore depends partly on how the RNA is folded. TDP-43 normally spends most of its time in the nucleus, but some of it moves between the nucleus and cytoplasm, and it is possible that changes in its RNA interactions due to changes in RNA conformation within these regions may affect where it remains. ^[2]

For instance, one possible disturbance is RNA methylation, the addition of a small chemical group that can affect RNA structure and its interactions with proteins. Sami Barmada's work examines a modification called m6A, which was increased at many measured sites in spinal cord tissue from people with ALS. Their proposed connection is that methylation and the proteins recognizing it can expose additional RNA sequences to TDP-43 binding. If that increases binding within the cytoplasm, more TDP-43 could become retained there, leaving less available for its nuclear functions. ^[2]

Barmada's group showed that adding RNA with a sequence that TDP-43 binds readily to could induce its redistribution in cultured neurons. This demonstrates that RNA could then be the factor that changes where TDP-43 accumulates, although whether the cell's own methylation changes produce the same effect remains a hypothesis. Also, once nuclear TDP-43 is depleted, the resulting RNA-processing defects could then compound the original disturbance, creating a feedback loop of problems even if the initial trigger was temporary. ^[2]

Regardless, once TDP-43 accumulates in the cytoplasm, its persistence also seems to depend partly on how quickly the cell can break it down. Through autophagy, a cell encloses material in a membrane-bound structure called an autophagosome, which joins with a lysosome containing enzymes that break the material down. Slower processing could allow cytoplasmic TDP-43 to accumulate, while changes in the protein's physical form could make it increasingly difficult to digest. Barmada's structural observations even raise the possibility that conditions inside these digestive structures themselves may increase the formation of resistant TDP-43 fibrils, although the sequence of events remains open. [2,3]

We therefore have several possible routes to TDP-43 pathology, including altered RNA binding, inadequate clearance and changes in the protein itself. And their relative importance could change throughout the stages of disease progression, if, for example, some initial disruption damages the systems that would ordinarily help the neuron recover. A treatment that improves clearance might then help a neuron overwhelmed by displaced protein, while a neuron continuing to lose nuclear TDP-43 may also need the process causing that loss to be addressed.

It may be that inherited vulnerabilities, lifetime exposures and accumulated damage combine differently in each person, changing how their genes are regulated and expressed, and how their neurons function. Some changes could involve acquired mutations, while others could involve epigenetic regulation that alters gene activity without changing the DNA sequence. These different histories could still produce similar failures because neurons depend on the same basic processes to sustain themselves. So the question may become how a specific individual arrived at that failure, including which changes in gene expression or cellular function continue to sustain their failure and which could then be targeted by treatment. In this case, it would be necessary to begin disease study from a level of personalization, rather than attempt to back out causality from aggregates.

We may see some evidence of this theory in the varied success of antisense oligonucleotides, or ASOs. ASOs offer a way to design treatments around a specific RNA sequence. Molecularly, an ASO is a short, synthetic chain of nucleotides that binds a complementary sequence in a selected RNA, allowing researchers to change how that RNA is processed or how much protein it produces. That makes them promising for neurological diseases in which we can identify a protein that needs to be increased or reduced, as illustrated by nusinersen in spinal muscular atrophy and tofersen in SOD1-associated ALS. [1,4–6]

ASOs can accomplish these changes through different mechanisms, including RNA degradation and altered splicing. A gapmer contains a central DNA-like stretch that pairs with its target RNA and permits an enzyme, RNase H1, to cut the RNA strand. The cell then degrades the transcript, reducing the amount available to make protein. A splice-modifying ASO instead occupies a regulatory sequence without recruiting that cleavage mechanism, changing which RNA segments the cell retains when it assembles the mature message and, in suitable cases, increasing production of a functional protein. [1,4]

As an aside, I find that increasing production of a protein that a patient lacks is more convincing as a therapeutic target than lowering its normal and harmful forms together. For instance, nusinersen (Spinraza) alters splicing to increase functional SMN protein in spinal muscular atrophy from abnormally low levels, whereas the FDA-approved tofersen (Qalsody) lowers both mutant and normal SOD1 in ALS. [4–6] As argued in A Consensus on Precision Therapy, precision should probably be a reference to how much normal function a treatment co-targets. Harm caused by suppressing that function would be on-target toxicity, even if the drug binds exactly where intended. Partial SOD1 reduction may leave enough for healthy function, but presymptomatic treatment, as will be investigated in the ongoing ATLAS study, will make normal SOD1 suppression much more of a consideration. ^[8]

Beyond that tradeoff in even an approved drug, the unsuccessful C9orf72 and ataxin-2 ALS ASO programs raise, in my mind, the questions I have about how patients are grouped for treatment. The C9orf72 repeat expansion is considered an established genetic cause of ALS and FTD, while ataxin-2 was implicated through genetic associations and experiments linking its reduction to less TDP-43 toxicity. [9–11] Yet BIIB078, targeting C9orf72 sense RNA, showed no clinical benefit, and BIIB105 lowered ataxin-2 without improving clinical outcomes or reducing neurofilament. Both were discontinued. [10,12] Their failures leave open the possibility that population-level findings concealed differences in how much individual patients depended on those targets for continued disease progression. Other genetic differences, changes in gene expression and accumulated cellular dysfunction could have mattered more to their treatment response than the shared feature used to select the target.

In other words, returning to the inverted funnel, a shared genetic error could still be identifying a relatively wide part of the process. Two people carrying the same major variant may differ in the other processes sustaining their disease, in how far those processes have progressed, and in how much function remains recoverable. Lowering one product could then have different consequences despite the shared mutation. So a genetic finding that seems conclusive in explaining commonalities in why people develop ALS may still tell us too little about which patients will respond to a drug aimed at that gene.

And this is why I suspect the search may need to become more individualized. Sequencing more patients might identify narrower genetic subsets, especially when combined with measurements of the cellular defects those variants actually produce. Perhaps ALS includes many more groups requiring different treatments than we currently recognize, and we will need to distinguish them by the particular genetic and cellular defects a drug can address. Even patients with the same disease-causing mutation might need different treatments if other parts of their disease have become more important to its progression, dependent upon environmental and inherited factors unique to each individual.

The SOD1 experience offers some encouragement for that approach, and the n-Lorem Foundation takes personalization further by developing experimental ASOs for exceptionally rare genetic conditions. Its patients have defined mutations that may occur in only 1–30 known people worldwide, with treatment selected according to what the mutation does and whether an ASO can plausibly alter it. The foundation has reported early clinical benefit in several rare conditions, although these observations come from individual treatment programs with limited follow-up. Perhaps some treatments that fail across a broad population would become useful if we could identify the smaller group whose disease depends on the process being targeted. The argument, in other words, is that the n-Lorem approach might be the correct one for an entire disease grouping. [7,13,14]

Greater specificity could also mean focusing more on shared downstream failures among patients with different genetic backgrounds. If several routes lead to a persistent RNA-processing defect or an inability to clear a harmful protein form, treatment might become possible at that narrower stage of the funnel. In this case, though, genetics may not be a target but a starting point. Patients would share a treatment because they have the same ongoing cellular problem, regardless of whether different mutations or life histories brought them there. In this sense, the argument here is that overweighting genetic association data as the source of causal truth may actually be a source of false specificity.

The economic difficulty is that this could produce many small treatment populations where the conventional development model expects a few large ones. Each ASO still needs manufacturing, safety work and clinical follow-up, even when the eligible population is tiny. n-Lorem's nonprofit model, supplying experimental medicines free for life, offers one way to organize that work, though it is unlikely to become mainstream in current capital funding structures, even if it became revenue-seeking. But it would be a societal failure if patients in these small groups were left without a treatment even when a plausible approach exists, simply due to ingrained abstractions of net present value calculations. [13,14]

The pursuit of scale becomes a problem when it begins deciding which biological differences we are willing to accommodate, especially if honest reflection starts to show that perhaps certain diseases which express similarly have inherently unique root causes. A person whose disease requires a more specific treatment has the same interest in recovering function as someone whose condition happens to fit a large commercial market. If the biology repeatedly leads us toward small populations, then making their treatment feasible becomes part of the problem that drug development needs to solve.

For ALS, this would put more emphasis on identifying what is still sustaining the disease in an individual patient, even if reconstructing its entire origin remains beyond reach or is simply due to random occurrences over the course of an individual's life. ASOs provide a way to act on some of those findings, particularly where we can restore a deficient function or selectively reduce a harmful product. But their failures also show the limitations of causality hypotheses based on statistical genetics, even with strong ensuing experimental findings.