C O M M E N TA RY
Initiation and propagation of
neurodegeneration
Christian Haass1,2
Although substantial progress has
been made in understanding the
molecular and pathological bases of
neurodegeneration, there have been
few successes in the clinic and a
number of fundamental questions
remain unanswered. Is this skepticism
misplaced, or do the words of Sir Isaac
Newton hold true, that “what we know is
a drop, what we don’t know is an ocean”?
Looking at the amount of literature published
on one topic related to neurodegeneration (for
instance, the generation of amyloid-β (Aβ)
peptide), one might be surprised to read such
a skeptical impression. The identification of
the proteolytic machinery involved in Aβ
generation1 paved the way for amyloid-based
therapeutic strategies. We know a number of
disease-causing genes and risk factors, and
research on neurodegenerative disease has also
provided unexpected discoveries of signaling
and protein degradation pathways linked to
autophagy2, neurogenesis3, myelination4–6 and
regulated intramembrane proteolysis7. Many
important questions, however, remain unan-
swered. Chief among them are the proximal
causes of neurodegeneration and how a more
holistic view of these triggers inform future
therapeutic strategies.
What is the nature of the neurotoxic
amyloid species?
The most relevant question is as follows: what
1DZNE—German Center for Neurodegenerative
Diseases, Munich, Germany. 2Adolf-Butenandt-
Institute, Biochemistry, Ludwig-Maximilians-
University, Munich, Germany.
e-mail: chaass@med.uni-muenchen.de
Published online 21 September 2010;
doi: 10.1038/nm.2223
nature medicine advance online publication 1
is the cause of neurodegeneration? For prion
disorders, it is clear that the scrapie form of the
prion protein (PrPsc) causes neuronal death8,
but in other diseases, such as Alzheimer’s
disease, the toxic protein or species is not as
evident. Moreover, the prion disease field has
an advantage in that its animal models have
a defined and clear endpoint: death. Recent
advances with organotypic brain slices even
permit rapid ex vivo analysis of prion amplifi-
cation and the contribution of other cell types
to amplification, as well as screening of com-
pounds that might protect against neurotoxic-
ity9. Thus, the prion field has a set of tools that
allow the dissection of pathological cascades
downstream of PrPsc and the identification of
targets to stem brain damage.
Researchers investigating Alzheimer’s dis-
ease lack such models and, at best, have models
that can recapitulate mild cognitive impair-
ment (MCI) and the pathology of Alzheimer’s
disease but not true neurodegeneration. As an
Alzheimer’s researcher, I strongly support the
amyloid cascade hypothesis10 (Fig. 1) but feel
we have not obtained definitive proof that Aβ is
the causative agent of neurodegeneration, nor
have we conclusively identified what forms of
Aβ initiate neurodegeneration (Fig. 1). The
pathologic characteristics of Alzheimer’s dis-
ease (namely plaques and tangles) have been
described for over 100 years, and extensive
biochemical and cell biology studies have con-
firmed that both Aβ and tau are neurotoxic.
Yet, plaques are also found in cognitively nor-
mal individuals, and plaque burden does not
correlate with memory decline. A recent study
in a cohort of cognitively intact nuns confirms
these observations and posits that compensa-
tory mechanisms emerge in the face of exten-
sive amyloid burden11, or, more provocatively,
calls into question the amyloid cascade hypoth-
esis. Moreover, successful removal of amyloid
plaques by immunotherapy fails to improve
cognition12. Collectively, do these studies dis-
credit the amyloid cascade hypothesis? Not
necessarily.
Genetic studies have shown that, in auto-
somal dominant familial Alzheimer’s dis-
ease, mutations in the genes that encode the
amyloid-β precursor protein and the two
presenilins (PS1 and PS2) alter Aβ genera-
tion and strongly support the idea that Aβ is
the disease-causing entity1. In mouse mod-
els of Alzheimer’s disease, their limitations
notwithstanding, therapeutic approaches that
target Aβ reduce disease-specific pathology
and ameliorate memory deficits13,14. The
field, however, tends to place undue empha-
sis on oligomeric Aβ as the disease-causing
entity to explain the discrepancies described
above. Do we know the precise biochemical
or structural characteristics and direct func-
tional effects of a disease-initiating oligomer?
Amyloid oligomers vary in size and shape
from small soluble oligomers such as dimers
and trimers, to multimers (such as Aβ*56)
and other assemblies, including Aβ-derived
diffusible ligands, annular complexes and
protofibrils15. Oligomers and fibers are in
equilibrium between aggregation and disag-
gregation, so it is unlikely that one specific
oligomeric species causes neuronal loss. It
seems more likely that many types of soluble
oligomers exert toxicity.
Small oligomers have been purified from
the conditioned medium of cells16 (although
here preferentially one cell line is issued17)
and from human brains18,19. But do soluble
oligomers really exist in vivo, or are they gen-
erated during the isolation protocol, which in
all cases includes procedures during which
isolated molecules are concentrated (a step
that in itself could initiate oligomerization)?
It is hard to characterize the in vivo properties
of oligomers, and new, highly sensitive bio-
physical technologies would be required to
commentary
visualize them directly in vivo. Furthermore,
imagine that a homogenous Aβ oligomer
is isolated that has defined physical and
neurotoxic characteristics and is applied to pri-
mary neurons, leading to neurotoxicity and in
vivo, to a reduction in long-term potentiation
and memory deficits. Do these experiments
provide evidence of a disease-causing entity?
It remains unclear whether these aggregates
are still of the same size and structure in vivo,
or whether they continue to aggregate and
are modified by processes such as oxidation
and cross-linking (Fig. 1). Another important
issue is whether long-term potentiation is the
best readout for these experiments, as it seems
to be extremely sensitive to many kinds of cel-
lular manipulations. Moreover, these oligom-
ers should not only affect synaptic function
but also contribute to the massive neuronal
loss observed in individuals with Alzheimer’s
disease. Although oligomers are apparently
present in mouse models, only limited neu-
ronal loss is seen in these models.
This also raises the question of the specific
downstream effects of Aβ toxicity. Do these
oligomeric Aβ species bind specific receptors?
Although one such receptor could be cellu-
lar PrP (PrPc)20, studies of PrPc-knockout
mice have produced conflicting results21,22.
Alternatively, do the toxic effects of Aβ rely
on nonspecific interactions with other pro-
teins or even lipid membranes (Fig. 1)?
Another point that is not yet fully under-
stood is the reliance of Aβ-dependent toxic-
ity on the presence of tau. Inhibition of tau
expression blocks seizures induced by the
Aβ-mediated overstimulation of excitatory
N-methyl-d-aspartate (NMDA) receptors
and improves survival in a transgenic mouse
model of Alzheimer’s disease23. In that regard,
it is interesting that Aβ toxicity, which is
largely thought to occur on the postsynaptic
membrane, has been shown to be meditated
by the axonal tau protein24. Surprisingly, tau
targets Fyn kinase to postsynaptic densities,
where it phosphorylates and stabilizes the
NMDA receptor subunit 2. Trapping tau in
the soma prevents phosphorylation of NMDA
receptor subunit 2 and blocks excitotoxic-
ity. Of note, the earliest pathological change
observed during the course of Alzheimer’s
disease is the redistribution of tau from the
axon to the somatodendritic compartment25.
In Alzheimer’s disease, enhanced amounts of
Fyn might reach dendritic spines and enhance
Aβ-mediated toxicity. As these events appar-
ently occur before neurofibrillary tangles are
formed, this calls into question the relevance
of these large intracellular inclusions in neu-
ronal cell death (Fig. 1). Recent studies are
consistent with this notion and have provided
2 advance online publication nature medicine
Changes in Aβ metabolism
Increase in total Aβ
Increase in the Aβ42/Aβ40 ratio
Reduced degradation
Oligomerization of Aβ42 and diffuse plaque deposits
What is the nature of the oligomeric species?
Is there a selective Aβ receptor?
Subtle effects of soluble oligomers on synapse function
Is LTP a good readout?
Abnormal distribution of tau and aberrant tau phosphorylation and dephosphorylation
How is Aβ linked to tau?
Oligomerization of tau
What is the nature of the oligomeric species?
Propagation and neuronal dysfunction leading to cell death
What is required for propagation?
Why are certain neurons resistant to cell death?
Which signaling pathways lead to cell death?
Dementia with
plaque and tangle pathology
What is the role of plaques and tangles?
How do environmental factors and aging affect
the development of pathology and dementia?
Figure 1 The amyloid cascade hypothesis. A number of questions need to be addressed to
understand this pathogenic cascade and the link between Aβ and tau pathology. LTP, long-term
potentiation.
evidence that neuronal cell death can occur impetus for additional trials of Aβ-directed
independently of and even before tangle therapeutic agents in people with sporadic
formation26,27. Oligomers of tau might be Alzheimer’s disease.
involved in disease processes, but the physi-
cal nature of such putative oligomers remains A unifying mechanism of disease
to be elucidated (Fig. 1). propagation
An alternative approach to test the amyloid
A preventive trial to prove the amyloid cascade hypothesis is to determine whether
cascade hypothesis disease-causing agents are able to induce
Ultimately, a proof-of-principle clinical spreading or propagation of the disease. What
trial of a therapeutic strategy that lowers Aβ types of Aβ species are sufficient to promote
abundance would provide clear evidence in further aggregation of native proteins and
support of the amyloid cascade hypothesis. initiate a neurodegenerative cascade? Studies
Recent clinical trials of Aβ immunotherapy by Mathias Jucker and his colleagues dem-
failed to show much cognitive improvement onstrated that brain homogenates contain-
despite a reduction in amyloid burden12. The ing Aβ were sufficient to induce Alzheimer’s
clinical symptoms of Alzheimer’s disease are disease–like pathology28. Moreover, patho-
thought to present after a long prodromal logical aggregates of α-synuclein and tau
phase, during which considerable neuronal can also ‘infect’ neighboring cells and induce
loss occurs. Ideally, one would need to treat pathology reminiscent of Parkinson’s dis-
patients with a defined risk (such as carriers of ease and frontotemporal dementia, respec-
presenilin mutations) presymptomatically in tively29–33. Adriano Aguzzi hypothesizes
a small preventive trial. If patients who were that many if not all amyloid proteins such as
treated presymptomatically with an anti-Aβ Aβ, tau, α-synuclein, amyloid A and poly-
vaccine or secretase inhibitor or modulator glutamine proteins are “prionoids” capable
showed improved cognition or a delay in of amplifying themselves via conformational
the onset of disease, this could provide the alterations34. Nevertheless, data from prion

disease research suggests that spreading and
neurotoxic signaling may not necessarily be
linked35. Moreover, do these prionoids act
alone? Probably not, as injection of synthetic
Aβ preparations, including purified oligom-
ers thought to be the neurotoxic entity (see
above), fails to induce spreading, suggesting
additional cofactors may be needed28.
Selective vulnerability and focal
induction of disease propagation
Neuronal degeneration spreads in a stereotyped
fashion, with certain regions of the brain (such
as the cerebellum) being spared until the late
stages of Alzheimer’s disease36. Why are certain
populations of neurons selectively vulnerable
to apoptosis while others remain resistant (Fig.
1)? This selective vulnerability might result in
part from differences in the lipid composition
of neuronal membranes. Pathological Aβ gen-
eration by secretases is strongly affected by the
surrounding lipid composition37. Alternatively,
we might gain some insight from cells that are
seemingly resistant to neurodegeneration, such
as cerebellar neurons in Alzheimer’s disease36.
Systems biology approaches including genom-
ics, proteomics and lipidomics could be used
to identify alterations in proteins or in the lipid
composition of these cells that confer vulner-
ability or resistance to the toxic effects of Aβ.
An integrated picture of disease
pathogenesis
An integrative approach to the analysis of dis-
ease will aid in obtaining a full picture of the
pathology of Alzheimer’s disease, which not
only consists of amyloid plaques and neuro-
fibrillary tangles but also incorporates cere-
bral amyloid angiopathy and inflammatory
responses. Traditionally, neurological diseases
have been separated into mechanistically
distinct families such as neurodegenerative,
inflammatory or vascular conditions. This
classification was predicated on the notion
that clinical phenotypes relate in a categorical
fashion to a discernable disease mechanism.
As a result, research efforts have traditionally
reflected this categorization and tend to focus
on one or another of these mechanisms in iso-
lation, with little or no crosstalk.
Yet, there is a clear interaction between
neurodegeneration, vascular dysfunction and
inflammation. For example, elimination of
microglia severely increases prion titers9, and
microglia are involved in the removal of amy-
loid plaques after vaccination against Aβ38. In
addition, analysis of several neurodegenerative
disorders using systems biology approaches
might reveal that an intricate network of shared
mechanisms initiates the neurodegenerative
cascade. If shared mechanisms of disease can
nature medicine advance online publication 3
be identified, new therapeutic agents can be
targeted to specific nodes within these net-
works. Caution is warranted, as systems biol-
ogy approaches have proven most useful in
analyzing simple model organisms and are only
beginning to be applied to disease-oriented
biomedical research in higher organisms.
Owing to recent advances in genomics
and proteomics, comprehensive quantita-
tive approaches can be applied to smaller and
smaller amounts of material, making an inte-
grative investigation of disease tissue possible.
The emergence of bioimaging and new models
of neurodegenerative disease allow a dynamic
assessment of cellular and molecular interac-
tions from the meso- to the nanoscale. One
caveat is that unbiased research strategies must
always be based on solid, functional readouts,
even on single genes and proteins.
Without that detailed knowledge, holistic
approaches might generate complex interaction
networks that in the end may have little if any
relevance in vivo. The identification of hundreds
of protein-protein interactions will not help us
to understand neurodegeneration unless we
investigate their functional importance at the
cellular level. In other words, we should avoid
producing data instead of knowledge.
Altered proteolysis
One process that might be dysregulated in
protein-folding disorders such as Alzheimer’s,
Parkinson’s and Huntington’s diseases is pro-
tein degradation2. Inhibition of the ubiquitin-
proteasome system is sufficient to induce neu-
ronal death39, and altered proteasome function
has been observed in Alzheimer’s disease40. A
number of proteins that have key roles in neu-
rodegeneration, such as huntingtin, androgen
receptor, ataxin-3, tau and α-synuclein, are
cleared by autophagy2. Autophagic clearance
of small aggregates is affected by aging and
altered in disease41,42, and recent findings
suggest that presenilin mutations affect this
proteolytic pathway43. Are aging-associated
alterations in autophagy an underlying cause
of neurodegeneration?
Alterations in proteolysis would be pre-
dicted to initiate degeneration simultaneously
in many brain regions, but these diseases are
associated with a regional focus in terms of
disease initiation. For example, in individu-
als with frontotemporal lobar degeneration,
autosomal dominant mutations in the pro-
granulin gene44,45 lead to the aggregation and
ubiquitous deposition of the TAR DNA bind-
ing protein-43 in the brain. Imaging studies on
affected individuals, however, reveal that the
disease does not start evenly in all parts of the
brain but begins with asymmetric atrophy46.
This suggests that other factors, in addition to
commentary
the mutant protein, might determine whether
a given anatomical region or cell type is resis-
tant or vulnerable to the toxic effects of the
protein.
Environmental influences—one hit is not
enough
There are many diseases in which the age of
onset of carriers with a specific autosomal
dominant mutation varies substantially, sug-
gesting that there are additional risk factors.
Recent findings support a two-hit hypothesis
for some genetically inherited cases of amyo-
trophic lateral sclerosis47. For example, muta-
tions in fused in sarcoma (FUS) cause this
form of the disease48,49. These mutations lead
to the redistribution of the protein from the
nucleus to the cytosol where, upon addition of
stressors, the protein forms aggregates charac-
teristic of stress granules47. Long-term stress
may precipitate formation of large deposits by
fusion of stress granules and cause disease. As
expression of the mutant protein alone does
not result in aggregate formation, these results
suggest that two events are needed: a gain- or
loss-of-function mutation in a protein, leading
to its altered cellular function or distribution,
and an additional stressor to promote aggre-
gate formation. Environmental factors greatly
affect disease onset in Parkinson’s disease50,
and environmental enrichment reduces the
cognitive deficits observed in a transgenic
mouse model of Alzheimer’s disease51. In
fact, clinicians sometimes suggest physi-
cal and mental exercise to delay the onset of
Alzheimer’s disease. A recent study showed
that Aβ generation is reduced during sleep and
increases during phases of activity52. Are these
two findings counterintuitive? Perhaps not, as
de novo Aβ production (as monitored in the
sleep study) and Aβ clearance (as observed in
the environmental enrichment studies) may be
carefully balanced in vivo, and slight changes
in either pathway could greatly affect disease
pathogenesis.
Translation, that’s what matters
It is clear that although substantial progress
has been made in understanding the cellular
mechanisms of neurodegeneration, problems
arise as soon as these findings are translated
into the clinic. In other words, we are ready
to treat Alzheimer’s disease–like pathology
and probably related memory phenotypes in
animal models, but there are major hurdles to
the treatment of patients. As discussed above,
we need to be able to identify individuals at
risk for neurodegeneration as early as pos-
sible (before profound neuronal cell loss). To
this end, new biomarkers and more sensitive
in vivo imaging techniques that can differ-
commentary
entiate disease-causing oligomeric species
from the large and potentially inert protein
deposits are required.
For those perplexed by the complexity of the
problem at hand, Sir Isaac Newton provides us
with some more guidance: “nature is pleased
with simplicity.” Moving forward, perhaps this
quote can provide a guiding principle for neu-
rodegenerative research.
ACKNOWLEDGMENTS
This article reports on some of the main points
raised at the Herrenhausen Symposium on
Neurodegeneration in Seeon, Germany (May 2010)
during the session on the initiation and propagation
of neurodegeneration. I thank A. Aguzzi,
D. Rubinsztein, S. Sisodia, D. Edbauer and M. Meyer-
Lühmann for reading this manuscript. This work was
supported by the Deutsche Forschungsgemeinschaft.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. Haass, C. EMBO J. 23, 483–488 (2004).
2. Rubinsztein, D.C. Nature 443, 780–786 (2006).
3. Veeraraghavalu, K., Choi, S.H., Zhang, X. & Sisodia,
S.S. J. Neurosci. 30, 6903–6915 (2010).
4. Hu, X. et al. Nat. Neurosci. 9, 1520–1525 (2006).
5. Willem, M. et al. Science 314, 664–666 (2006).
6. Bremer, J. et al. Nat. Neurosci. 13, 310–318
(2010).
4 advance online publication nature medicine
7. Brown, M.S., Ye, J., Rawson, R.B. & Goldstein, J.L.
Cell 100, 391–398 (2000).
8. Aguzzi, A. & Haass, C. Science 302, 814–818
(2003).
9. Falsig, J. et al. Nat. Neurosci. 11, 109–117 (2008).
10. Hardy, J. & Selkoe, D.J. Science 297, 353–356
(2002).
11. Iacono, D. et al. Neurology 73, 665–673 (2009).
12. Holmes, C. et al. Lancet 372, 216–223 (2008).
13. Schenk, D. et al. Nature 400, 173–177 (1999).
14. Schenk, D., Hagen, M. & Seubert, P. Curr. Opin.
Immunol. 16, 599–606 (2004).
15. Haass, C. & Selkoe, D.J. Nat. Rev. Mol. Cell Biol. 8,
101–112 (2007).
16. Walsh, D.M. et al. Nature 416, 535–539 (2002).
17. Podlisny, M.B. et al. Biochemistry 37, 3602–3611
(1998).
18. Lesné, S. et al. Nature 440, 352–357 (2006).
19. Shankar, G.M. et al. Nat. Med. 14, 837–842 (2008).
20. Laurén, J., Gimbel, D.A., Nygaard, H.B., Gilbert, J.W. &
Strittmatter, S.M. Nature 457, 1128–1132 (2009).
21. Calella, A.M. et al. EMBO Mol. Med. 2, 306–314
(2010).
22. Gimbel, D.A. et al. J. Neurosci. 30, 6367–6374
(2010).
23. Roberson, E.D. et al. Science 316, 750–754 (2007).
24. Ittner, L.M. et al. Cell 142, 387–397 (2010).
25. Ballatore, C., Lee, V.M. & Trojanowski, J.Q. Nat. Rev.
Neurosci. 8, 663–672 (2007).
26. de Calignon, A. et al. Nature 464, 1201–1204
(2010).
27. Paquet, D. et al. J. Clin. Invest. 119, 1382–1395
(2009).
28. Meyer-Luehmann, M. et al. Science 313, 1781–1784
(2006).
29. Frost, B., Jacks, R.L. & Diamond, M.I. J. Biol. Chem.
284, 12845–12852 (2009).
30. Clavaguera, F. et al. Nat. Cell Biol. 11, 909–913
(2009).
31. Kordower, J.H., Chu, Y., Hauser, R.A., Freeman, T.B. &
Olanow, C.W. Nat. Med. 14, 504–506 (2008).
32. Li, J.Y. et al. Nat. Med. 14, 501–503 (2008).
33. Aguzzi, A. & Rajendran, L. Neuron 64, 783–790
(2009).
34. Aguzzi, A. Nature 459, 924–925 (2009).
35. Chesebro, B. et al. Science 308, 1435–1439 (2005).
36. Thal, D.R., Capetillo-Zarate, E., Del Tredici, K. & Braak,
H. Sci. SAGE KE 2006, re1 (2006).
37. Fraering, P.C. et al. Biochemistry 43, 9774–9789
(2004).
38. Bard, F. et al. Nat. Med. 6, 916–919 (2000).
39. Winklhofer, K.F., Tatzelt, J. & Haass, C. EMBO J. 27,
336–349 (2008).
40. Keller, J.N., Hanni, K.B. & Markesbery, W.R. J.
Neurochem. 75, 436–439 (2000).
41. Ravikumar, B., Duden, R. & Rubinsztein, D.C. Hum.
Mol. Genet. 11, 1107–1117 (2002).
42. Cataldo, A.M., Hamilton, D.J., Barnett, J.L., Paskevich,
P.A. & Nixon, R.A. J. Neurosci. 16, 186–199 (1996).
43. Lee, J.H. et al. Cell 141, 1146–1158 (2010).
44. Baker, M. et al. Nature 442, 916–919 (2006).
45. Cruts, M. et al. Nature 442, 920–924 (2006).
46. Rohrer, J.D. et al. Neuroimage published online,
doi:10.1016/j.neuroimage.2009.12.088 (4 January
2010).
47. Dormann, D. et al. EMBO J. 29, 2841–2857 (2010).
48. Kwiatkowski, T.J. Jr. et al. Science 323, 1205–1208
(2009).
49. Vance, C. et al. Science 323, 1208–1211 (2009).
50. Di Monte, D.A., Lavasani, M. & Manning-Bog, A.B.
Neurotoxicology 23, 487–502 (2002).
51. Lazarov, O. et al. Cell 120, 701–713 (2005).
52. Kang, J.E. et al. Science 326, 1005–1007 (2009).

Degeneration and repair in central
nervous system disease
Eng H Lo
Divergent disease triggers in
neurodegeneration may induce
convergent endogenous pathways in
neuronal, glial and vascular elements
as the central nervous system (CNS)
attempts to compensate, remodel and
recover. Dissecting these multicellular
mechanisms and the integrative
responses in cerebral blood flow
and metabolism may allow us to
understand the balance between injury
and repair, validate new targets and
define therapeutic time windows for
neurodegeneration.
Remarkable advances have been made in the
last decade in understanding the basic mecha-
nisms of neurodegeneration. Progress has
come on many fronts, including the molecular
biology of cell death, animal models, genetics
and neuroimaging of clinical disease. Many tar-
gets are being tested in a wide range of experi-
mental studies and clinical trials. Along with
the excitement, however, some caution may
be warranted. Advances in our knowledge of
excitotoxicity, oxidative stress, mitochondrial
dysfunction and apoptosis similarly provided
a rich repertoire of targets and drugs for ‘acute
neurodegeneration’ after stroke and traumatic
brain injury1,2. Yet, almost all clinical trials in
stroke and brain trauma have failed thus far.
Broadly speaking, there is still no clinically
validated neuroprotectant. This article reports
on one speculative hypothesis: triggers of
Eng H. Lo is at the Neuroprotection Research
Laboratory, Departments of Neurology and
Radiology, Massachusetts General Hospital and
Harvard Medical School, Boston, Massachusetts,
USA.
e-mail: lo@helix.mgh.harvard.edu
Published online 21 September 2010;
doi: 10.1038/nm.2226
nature medicine advance online publication 1
C O M M E N TA RY
disease in neurodegeneration may also act as
stimuli that induce endogenous compensa-
tion and recovery. Hence, investigations into
the pathophysiology of CNS disease should
take into account not only the primary mecha-
nisms of degeneration but also simultaneous
processes of regeneration as the brain tries to
repair itself.
Divergence in disease, convergence in
repair?
The mammalian nervous system is highly com-
plex. Many things can go wrong in many ways.
Clinically, there are multiple ‘types’ of neuro-
degeneration, comprising major diseases such
as Alzheimer’s disease, Parkinson’s disease,
amyotrophic lateral sclerosis (ALS), multiple
sclerosis, Huntington’s disease and prion dis-
ease. This wide spectrum of neurodegenera-
tive phenotypes reflects key differences in the
proximal triggers and pathologic markers of
disease: plaques and tangles in Alzheimer’s
disease, Lewy bodies in Parkinson’s disease,
motoneuron death in ALS and demyelination
in multiple sclerosis. However, these seemingly
divergent mechanisms also lead to multiple
convergent pathways. Regardless of the initial
triggers, many overlapping downstream and/
or secondary pathways are induced, including
neuroinflammation3. An important challenge
in designing therapies for neurodegeneration is
to not only eliminate the initial disease triggers
but also ameliorate the deleterious inflamma-
tory responses within the CNS once the disease
is under way.
Although the CNS was originally thought
to be an immune-privileged organ, it is now
known that inflammation can be triggered in
injured brain tissue. For example, activation
of microglia and responses in both innate and
adaptive immunity accompany almost all types
of neurodegeneration4,5. From an evolutionary
perspective, inflammation can be interpreted
as part of a highly conserved set of endogenous
responses to injury and disease in any organ6.
The same might be true in the context of neu-
rodegeneration.
In spite of the highly divergent disease trig-
gers in neurodegeneration, is it possible that
these diseases also induce convergent down-
stream mechanisms of compensation, repair
and remodeling (Fig. 1a)? This is obviously a
speculative idea. But some evidence of endog-
enous recovery during neurodegeneration can
indeed be detected in animal models and clini-
cal studies.
It is well known that behavioral and motor
adaptation allows for some individuals with
Parkinson’s disease to compensate during
the initial stages of disease. But beyond com-
pensation, some degree of actual neuronal
remodeling may also take place. Dendritic
sprouting and reafferentation of damaged
areas are known to occur in animal models of
focal brain lesions. The most common animal
models of Parkinson’s disease involve selective
lesioning of dopaminergic neurons within the
substantia nigra. Because the substantia nigra
projects to the striatum, this depletes striatal
dopamine and mimics clinical Parkinson’s dis-
ease. However, it is now recognized that local
sources of dopamine may also exist in the mam-
malian striatum, deriving in part from intrinsic
tyrosine hydroxylase–positive neurons7. After
loss of dopaminergic input from the substan-
tia nigra, local tyrosine hydroxylase–positive
neurons seem to remodel and expand in rodent
and primate models of Parkinson’s disease8,9.
It has been proposed that similar substrates
of neuroplasticity may also exist in individu-
als with Parkinson’s disease10. Whether these
pathways truly restore clinically relevant dop-
amine function remains to be determined11.
Similar substrates of repair and remodel-
ing may also be present in Alzheimer’s dis-
ease. Indeed, one of the early observations in
Alzheimer’s disease was that hippocampal
degeneration is often accompanied by a reor-
commentary
ganization of acetylcholinesterase staining pat-
terns, indicative of cholinergic sprouting from
basal forebrain regions12. More recent stud-
ies now propose that increased expression of
synaptic proteins, such as postsynaptic density
protein-95, may underlie plasticity in prefrontal
and frontal networks as the brain attempts to
cognitively compensate for degenerating neu-
rons13. Functional magnetic resonance imaging
studies suggest that remapping of cortical net-
works may extend to motor systems as well14.
Indeed, the degree of cognitive decline may be
more closely related to an individual’s endog-
enous neuronal plasticity than to the cerebral
abundance of amyloid or tangles per se15. Recent
work is beginning to elucidate the underlying
mechanisms. Perturbations in amyloid pre-
cursor protein processing lead to increased
amyloid-β generation, assembly of toxic oligom-
ers, plaque formation and Alzheimer’s disease
pathology16. However, homeostatic forms and
levels of amyloid-β may also act as functional
regulators of transmitter release and synaptic
function17,18. Ultimately, the balance between
beneficial-adaptive and aberrant-maladaptive
forms of synaptic and neuronal remodeling
may significantly influence how Alzheimer’s
disease pathology disrupts cerebral function as
the disease progresses19.
Endogenous mechanisms of brain repair
and recovery can be classified into several cat-
egories: behavioral compensation, activation
of latent or parallel circuits, neuronal plastic-
ity, remodeling of discrete networks and per-
haps even neurogenesis. Traditionally, it was
thought that no new neurons were generated
after birth in mammalian brains. However,
a wealth of studies have led to a paradigm
shift. It is now generally accepted that pock-
ets of ongoing neurogenesis persist in the
adult brain, including active loci in the sub-
ventricular zone lining the lateral ventricles
and subgranular zone of the dentate gyrus.
Emerging data suggest that neurogenesis is
surprisingly plastic, and increased neurogen-
esis can be observed in models of CNS injury
and disease20. Several suspected molecular
mediators in Alzheimer’s disease, such as
amyloid, presenilin 1, Notch and ErbB4, are
known to participate in the molecular regula-
tion of neurogenesis21. A survey of the animal
model literature reveals that neurogenesis can
sometimes be affected in Alzheimer’s disease,
depending on the stage of disease and the
underlying phenotypes of the mutant mice
involved22. Whether alterations in neurogen-
esis actually occur in human neurodegenera-
tion remains unclear and provides an exciting
frontier for research.
An emerging concept in neuroscience sug-
gests that neuronal responses are intricately
2 advance online publication nature medicine
a Cofactors (genetic, environmental,
physiological, pharmacological)
Endogenous response Prolonged inflammation
Initial disease triggers
to injury and degeneration
Non–cell-autonomous crosstalk
in neuronal, glial
and vascular compartments
b
Cumulative injury
Endogenous compensation
Level of response
and remodeling
Time after disease initiation
Optimal treatment
window
Figure 1 Endogenous responses, cofactors and therapeutic time windows in neurodegeneration. The
wide spectrum of pathology in neurodegeneration reflects a multitude of initial triggers involved in
the induction of disease. But divergent upstream mechanisms also lead to convergent downstream
pathways as the brain responds to disease and injury. Endogenous responses of the CNS may comprise
both deleterious and potentially beneficial mechanisms of inflammation and neurovascular remodeling.
Non–cell-autonomous crosstalk between neuronal, glial and vascular compartments form the basis
for these phenomena, and many cofactors influence the signals and substrates of these endogenous
responses over time. Ultimately, prolonged inflammation and injury lead to a cumulative disease
burden that outstrips any endogenous attempts at compensation or remodeling. Understanding
these transitions may help define therapeutic time windows where candidate treatments for
neurodegeneration should be more effective.
linked to vascular signaling as part of an evolu- inflammatory cascade. To some degree, then,
tionarily conserved phenomenon1,2,23. Hence, angiogenesis in diseased brains may represent
neuroplasticity may occur only within the con- a nonspecific epiphenomenon associated with
text of brain microvessel remodeling24. New neurodegeneration.
cells and networks need new blood supplies.
If endogenous repair and neuronal recovery Multiple mechanisms in multiple cells
take place in neurodegeneration, is there a The neurovascular unit is an increasingly
corresponding angiogenic response as well? accepted conceptual model in neuroscience, in
Although the literature is sparse, there have which cell-to-cell signaling between neuronal,
been a few descriptions of augmented vascular glial and vascular elements underlies both
growth in a wide spectrum of CNS diseases, physiology and pathophysiology in the brain1,2.
including Alzheimer’s disease, Parkinson’s The term ‘neurodegeneration’ implies a purely
disease and multiple sclerosis25–29. Insofar neuronal phenomenon, but this is clearly not
as triggers of disease may conversely act as the case when one examines the signals and
stimuli for repair, it might be interesting to substrates of disease more closely. Disruptions
ask whether mediators of neurodegeneration in cell-cell signaling within the neurovascu-
can promote vascular recovery. For example, lar unit may underlie non–cell-autonomous
amyloid peptides may augment angiogen- mechanisms during the response to injury and
esis by amplifying fibroblast growth factor disease in the CNS32 (Fig. 1a).
signaling30. Furthermore, crosstalk between For example, interactions between neurons
angiogenesis and neurogenesis may occur, and astrocytes are essential for the regulation
as amyloid has been shown to increase rates of glutamate transmission. Without astrocytes,
of neuronal differentiation of bone marrow– neurons become increasingly vulnerable to
derived endothelial precursor cells31. However, excitotoxicity33. Any disruption in astrocytic
it is important to acknowledge that vascular function can markedly promote neurode-
remodeling is a crucial component of any generation. Transgenic Huntington’s disease

model mice that express mutant huntingtin
only in cortical pyramidal neurons do not show
neurodegeneration34, whereas mice that express
huntingtin in astrocytes develop age-dependent
neurological worsening35. In a mouse model
of Niemann-Pick disease, expression of wild-
type Niemann-Pick disease type C1 pro-
tein in astrocytes was sufficient to decrease
the rate of neurodegeneration and increase
life span36. Primary astrocyte and micro-
glial cultures derived from the superoxide
dismutase–mutant mouse model of ALS pro-
duce neurotoxic mediators in conditioned
media that kill wild-type motoneurons37. In
mouse models of Alzheimer’s disease, wide-
spread perturbations of calcium signaling were
detected in astrocyte networks long distances
away from amyloid plaques38. Hence, protect-
ing neurons alone may not be enough, and
strategies to improve glial function may also
be important for CNS disease therapies.
Signaling between endothelial cells and
neurons may have especially crucial roles in
neurodegeneration23,29,39. Endothelial dys-
function and perturbations in microvascular
flow regulation precede the development of
neuronal dysfunction in some mouse models
of Alzheimer’s disease40. Subtle disruptions in
the blood-CNS barrier may be detected before
neuronal death in the superoxide dismutase–
mutant mouse model of ALS41,42. Compromised
blood flow in white-matter regions may precede
lesion development in some individuals with
multiple sclerosis43. Microvessels in the brain
might not be simply an inert system for blood
flow. Instead, the cerebral endothelium may
also serve as an endocrine organ that actively
exchanges signaling mediators and trophic fac-
tors with gray and white matter44–46. Disease
triggers such as amyloid may disrupt these
forms of trophic coupling between the vascular
and neuronal compartments.
If non–cell-autonomous signaling between
neurons, glia and blood vessels contributes to
disease progression, is it possible that com-
pensation and repair also involve coordinated
remodeling in the entire neurovascular unit
(Fig. 1a)? Again, examples can be drawn from
findings in acute brain injury. Angiogenic and
neurogenic plasticity are tightly coregulated
after stroke and brain trauma47. This might
not be surprising, as molecular mechanisms of
neurogenesis and angiogenesis have been evo-
lutionarily conserved so that similar mediators
and pathways are involved in both phenom-
ena48. After experimental stroke in rat mod-
els, newborn neurons are seen to migrate along
‘highways’ that seem to run in close proximity
to remodeling blood vessels49. Promotion of
neural plasticity enhances vascular regrowth;
conversely, angiogenic stimulation enhances
nature medicine advance online publication 3
neural repair50,51. The pathophysiology of
neurodegeneration will surely include analo-
gous interactions between neuronal, glial and
vascular cells.
Taken together, non–cell-autonomous
mechanisms in neurodegeneration may pro-
vide a conceptual framework for asking further
questions. How do these multicell pathways
mediate the convergent mechanisms between
degeneration and regeneration? Can differences
in cell-to-cell signaling help explain the selec-
tive neuronal vulnerability that is the signature
phenotype of so many CNS diseases? What are
the mediators that regulate these processes?
It has been proposed that brain injury and
repair sometimes include biphasic mediators
and mechanisms52. The same molecule or cell
can behave differently at different phases of the
disease. Overactivation of N-methyl-d-aspartic
acid (NMDA) receptors induces excitotoxicity,
but without NMDA signaling, neuronal plas-
ticity cannot take place53. The stress-activated
protein kinase c-Jun N-terminal kinase pro-
motes neuronal apoptosis but is also required
for axonal remodeling54. Activated microglia
are key players in neuroinflammation, but cer-
tain forms of microglia can secrete beneficial
neurotrophic factors55. Aggregated proteins
may represent misfolded pathologic species
or underlie protective endogenous attempts
to sequester toxic molecules56. Understanding
how these biphasic mechanisms are coordi-
nated in multiple cell types will be essential
for dissecting the precarious balance between
neurodegeneration and ongoing compensation
and repair.
Cofactors in risk and recovery
Advances in genetics have revealed a host of
underlying mechanisms that increase risk of
disease. In neurodegeneration, the power of
genetics is clearly indicated in the tremendous
number of sophisticated and targeted transgenic
mouse models that are now available. However,
it might still be useful to ask whether these mice
represent models of disease or, more accurately,
models of mechanisms. These discussions may
not be purely semantic. The initial molecular
triggers of neurodegeneration are cellular, but
pathophysiology is expressed at the organ level,
and clinical disease is usually manifested in
patients with numerous other health problems
requiring concurrent medications.
Would it be useful to consider the devel-
opment of combination approaches whereby
such cofactors are incorporated into the mouse
models (Fig. 1a)? Because aging is a key cofac-
tor in neurodegeneration, future studies should
aim to test leading candidate therapies in aged
transgenic mice. Concomitant vascular dis-
ease may be a crucial cofactor in dementia,
commentary
with overlaps between vascular dementia and
Alzheimer’s disease being especially impor-
tant57. Various Alzheimer’s disease trans-
genic mice could be examined in the context
of hypertension, diabetes and other forms of
vascular dysfunction. Given a body of literature
on the effects of statins on Alzheimer’s disease,
these models could also be used to examine
how cardiovascular medications such as statins
and antihypertensive drugs interact with the
various pathways targeted in neurodegenera-
tion mouse models. Finally, as discussed above,
differences in systemic inflammation may be
crucial for disease pathogenesis. Almost all
mouse models are developed and assessed in
rigorously controlled housing conditions. But
does this also mean that most of our trans-
genic mice have highly ‘clean’ inflammatory
baselines, unlike those of a typical human
with metabolic and vascular diseases? How do
altered inflammatory baselines affect the bal-
ance between degeneration and regeneration?
Beyond promoting the risk of disease, these
cofactors may also contribute to resistance to
disease. Research efforts are largely focused on
identifying factors and mediators that increase
the rate of neurodegeneration. But individu-
als, even with similar loads of disease triggers,
can have widely varying rates of neurological
decline. Is it possible that differences in the
ability to adapt, repair and remodel partly
underlie this phenomenon? And, if so, might it
be productive to screen for genes and cofactors
that promote compensation and recovery and
enhance resistance to neurodegeneration?
Time is brain
The mantra in stroke and trauma clinical tri-
als is “time is brain.” After the initial ischemic,
hemorrhagic or traumatic insult, brain cells
inexorably begin to die. The longer one waits,
the more cells are lost, and the less effective
any putative neuroprotective therapy might
be. Clinical trial design in stroke and trauma
is highly focused on time of treatment, where
only patients admitted within specific times
after onset are enrolled. Is it possible that simi-
lar attention to treatment time windows is also
essential for neurodegeneration (Fig. 1b)?
If one accepts the proposed hypothesis of
ongoing degeneration and regeneration in
CNS disease, then one would predict that ini-
tially, the disease burden would be light and
neuronal dysfunction could be ameliorated
by endogenous mechanisms of compensation,
remodeling and repair. At this stage, patients
might still be asymptomatic or relatively well.
As the disease continued to progress beyond
certain thresholds, the burden of disease would
overcome the brain’s endogenous ability to
cope. During this transition phase, neurode-
commentary
generation would accelerate, and saving the
brain might become extremely difficult, no
matter how potent the therapy may be. Hence,
under some conditions, preventing disease
progression might be more effective than try-
ing to cure the brain after decades of neuronal
dysfunction and death.
An example of this concept of ‘prevention’
versus ‘cure’ can be observed in the mouse
model literature for ALS. Effective neuro­
protection—that is, decreased motor impair-
ment and increased life span—has been reported
for a wide range of targets in superoxide
dismutase–mutant mouse models of ALS58. Yet
a closer look would reveal that the majority of
these studies involved treatment of mice in the
very early stages of disease. In contrast, most
clinical trials in ALS recruit subjects that are
sometimes quite far along in terms of disease
progression and severity. This phenomenon is
by no means unique to ALS. A recent study in
the TG4510 tau-mutant mouse showed that,
during the early stages of disease, many neu-
rons were positive for active caspase but had not
yet undergone apoptosis, suggesting that tau-
bearing neurons are surprisingly long lived59. In
an inducible cell model of tauopathy, turning off
the gene encoding mutant tau allowed neurons
to recover before degenerative cascades were
too far gone60. Similarly, amyloid-β immuno-
therapy may work better at reducing amyloid
load and deposition as a preventive measure
rather than after pathology is established61.
Hence, timing is crucial, and there might be
windows of time in which Alzheimer’s disease
therapies are more effective. Indeed, the major-
ity of experimental treatments for Alzheimer’s
disease have been tested in relatively young
mutant mice, whereas clinical trials may typi-
cally involve a much more heterogeneous pop-
ulation of dementia patients62. However, it is
important to recognize that translating time
windows in mice into relevant times for therapy
in humans may also be difficult.
Ultimately, early treatments for neurodegen-
eration should logically reap greater benefits.
The sooner one tackles the disease, the better.
But how would one find these patients? By
definition, the individuals with earliest-stage
disease may be asymptomatic. Neuroimaging
or tissue biomarkers may be required63. If sur-
rogate markers (imaging of amyloid, tau or
neurotransmitter depletion) can be quantified
and proven to rigorously correlate with clinical
states, then one might envision clinical trials
using biomarkers not only for subject selection
and recruitment but also for the assessment of
secondary outcome measures. No matter how
potent one’s proposed target or drug, biologi-
cal variation will always reveal responders and
nonresponders. Defining the progressively
4 advance online publication nature medicine
worsening balance between degeneration and
regeneration may ultimately allow us to find
patients who still have a chance (Fig. 1b).
Collaborations and consortia
Complex problems require collaborative
solutions. To begin with, the basic molecular
mechanisms of neurodegeneration are highly
multifactorial. If the proposed phenomenon
of simultaneous compensation and repair is
indeed ongoing, this would add yet another
layer of complexity. Traditional approaches
involving a single model, single pathway and
single laboratory may not be enough. To dissect
these underlying mechanisms of degeneration
and regeneration in the entire neurovascular
unit, multidisciplinary networks and consortia
will be useful.
The richness of available models in neuro-
degeneration may be both an advantage and
a disadvantage. The many cellular and animal
models provide diverse and powerful oppor-
tunities for generating and testing hypotheses
at the mechanistic level. However, in terms of
serving as preclinical platforms for drug testing,
they can also pose a challenge. Which models
should one use for screening, development and
optimization? Model systems span a wide phy-
logenetic range, comprising individual cells,
Caenorhabditis elegans, Drosophila melano-
gaster, zebrafish, mice and perhaps even non-
human primates. The workhorse in the field
remains the transgenic mouse, but the major-
ity of these highly targeted mice may represent
models of mechanisms rather than standalone
models of human disease. Without collabora-
tions, the increasing heterogeneity of models
and techniques may prove intractable.
A recent meta-analysis of the Alzheimer’s
disease mouse model literature clearly showed
that many of the models being tested involved
presymptomatic mice62. Perhaps most impor-
tant, many studies were underpowered and
lacked proper randomization, blinding and
statistical controls. It is notable that similar
conclusions were reached in a meta-analysis of
animal stroke models64. An emerging consen-
sus in the stroke field suggests that consortia
are needed to unravel these difficult differences
between models and laboratories65. Would col-
laborative consortia also be productive in neu-
rodegeneration? Early success in the Alzheimer
Disease Neuroimaging Initiative suggests that
broadly collaborative networks can indeed be
built, where data are shared and findings can be
rapidly disseminated and leveraged.
Questions and opportunities
Many questions remain, but these translational
challenges also present us with new opportuni-
ties. Collaborative approaches should be useful
for conducting rigorously powered, statistically
controlled screening of candidate mechanisms,
targets and drugs. Findings could be replicated
in multiple models across multiple laborato-
ries. The role of cofactors (age, gender, envi-
ronment, baseline inflammation, metabolic
and cardiovascular disease, and concurrent
medications) could be more easily assessed.
Different laboratories could also contribute
expertise in dissecting non–cell-autonomous
crosstalk between various neuronal, glial and
vascular cell types in the mammalian CNS.
Molecular mechanisms of selective neuronal
death may not make sense unless interpreted in
the context of integrated responses in cerebral
blood flow and metabolism.
In spite of divergent disease triggers, it
is likely that convergent downstream path-
ways of secondary response and remodeling
will be activated during neurodegeneration.
Dissecting the interplay between injury and
repair may allow us to find therapies that
maximize neuroprotection and neurorepair.
Ultimately, understanding these transitions
and mechanisms may help us find ways to
identify patients with compensatory reserves
who still have salvageable brains.
ACKNOWLEDGMENTS
The author apologizes to colleagues whose important
work could not be directly cited. Because of space
limitations, mostly review articles were used as
starting points for discussion. Many thanks to
B. Bacskai, D. Selkoe, B. Hyman, M. Schwarzschild
and M.M. Ning for helpful discussions; and F. Beal,
D. Cleveland, E. Mandelkow, W. Robberecht and
all participants in the Herrenhausen Symposium
on Neurodegeneration for a wonderful educational
experience.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. Lo, E.H., Dalkara, T. & Moskowitz, M.A. Nat. Rev.
Neurosci. 4, 399–415 (2003).
2. Moskowitz, M.A., Lo, E.H. & Iadecola, C. Neuron 67,
181–198 (2010).
3. Glass, C.K., Saijo, K., Winner, B., Marchetto, M.C. &
Gage, F.H. Cell 140, 918–934 (2010).
4. Block, M.L. & Hong, J.S. Prog. Neurobiol. 76, 77–98
(2005).
5. Lucin, K.M. & Wyss-Coray, T. Neuron 64, 110–122
(2009).
6. Medzhitov, R. Nature 454, 428–435 (2008).
7. Betarbet, R. et al. J. Neurosci. 17, 6761–6768
(1997).
8. Blanchard, V. et al. J. Neurochem. 64, 1669–1679
(1995).
9. Tandé, D. et al. Brain 129, 1194–1200 (2006).
10. Porritt, M.J. et al. Lancet 356, 44–45 (2000).
11. Huot, P., Levesque, M. & Parent, A. Brain 130, 222–
232 (2007).
12. Hyman, B.T., Kromer, L.J. & Van Hoesen, G.W. Ann.
Neurol. 21, 259–267 (1987).
13. Leuba, G. et al. Neurobiol. Dis. 30, 408–419 (2008).
14. Agosta, F. et al. Hum. Brain Mapp. 31, 515–525
(2010).
15. Iacono, D. et al. Neurology 73, 665–673 (2009).
16. Selkoe, D.J. Behav. Brain Res. 192, 106–113
(2008).
17. Abramov, E. et al. Nat. Neurosci. 12, 1567–1576
(2009).

18. Pearson, H.A. & Peers, C. J. Physiol. (Lond.) 575, 5–10
(2006).
19. Palop, J.J. et al. Neuron 55, 697–711 (2007).
20. Koch, P., Kokaia, Z., Lindvall, O. & Brustle, O. Lancet
Neurol. 8, 819–829 (2009).
21. Lazarov, O. & Marr, R.A. Exp. Neurol. 223, 267–281
(2010).
22. Marlatt, M.W. & Lucassen, P.J. Curr. Alzheimer Res. 7,
113–125 (2010).
23. Zacchigna, S., Lambrechts, D. & Carmeliet, P. Nat. Rev.
Neurosci. 9, 169–181 (2008).
24. Madri, J.A. J. Physiol. Pharmacol. 60 Suppl 4, 95–104
(2009).
25. Barcia, C., Emborg, M.E., Hirsch, E.C. & Herrero, M.T.
Front. Biosci. 9, 277–282 (2004).
26. Desai, B.S., Schneider, J.A., Li, J.L., Carvey, P.M. &
Hendey, B. J. Neural Transm. 116, 587–597 (2009).
27. Holley, J.E., Newcombe, J., Whatmore, J.L. & Gutowski,
N.J. Neurosci. Lett. 470, 65–70 (2010).
28. Vagnucci, A.H. Jr. & Li, W.W. Lancet 361, 605–608
(2003).
29. Zlokovic, B.V. Trends Neurosci. 28, 202–208 (2005).
30. Cantara, S. et al. FASEB J. 18, 1943–1945 (2004).
31. Jin, H.K., Bae, J.S., Furuya, S. & Carter, J.E. Cell Prolif.
42, 571–586 (2009).
32. Ilieva, H., Polymenidou, M. & Cleveland, D.W. J. Cell
Biol. 187, 761–772 (2009).
33. Rosenberg, P.A. & Aizenman, E. Neurosci. Lett. 103,
nature medicine advance online publication 5
162–168 (1989).
34. Gu, X. et al. Neuron 46, 433–444 (2005).
35. Bradford, J. et al. J. Biol. Chem. 285, 10653–10661
(2010).
36. Zhang, M. et al. J. Neurosci. Res. 86, 2848–2856
(2008).
37. Nagai, M. et al. Nat. Neurosci. 10, 615–622 (2007).
38. Kuchibhotla, K.V., Lattarulo, C.R., Hyman, B.T. &
Bacskai, B.J. Science 323, 1211–1215 (2009).
39. Iadecola, C. Nat. Rev. Neurosci. 5, 347–360 (2004).
40. Iadecola, C. Cell. Mol. Neurobiol. 23, 681–689
(2003).
41. Garbuzova-Davis, S. et al. PLoS One 2, e1205
(2007).
42. Zhong, Z. et al. Nat. Neurosci. 11, 420–422 (2008).
43. Varga, A.W. et al. J. Neurol. Sci. 282, 28–33 (2009).
44. Arai, K. & Lo, E.H. J. Neurosci. 29, 4351–4355
(2009).
45. Dugas, J.C. et al. J. Neurosci. 28, 8294–8305
(2008).
46. Guo, S. et al. Proc. Natl. Acad. Sci. USA 105, 7582–
7587 (2008).
47. Arai, K., Jin, G., Navaratna, D. & Lo, E.H. FEBS J. 276,
4644–4652 (2009).
48. Carmeliet, P. & Tessier-Lavigne, M. Nature 436, 193–
200 (2005).
49. Thored, P. et al. Stroke 38, 3032–3039 (2007).
50. Ohab, J.J., Fleming, S., Blesch, A. & Carmichael, S.T.
commentary
J. Neurosci. 26, 13007–13016 (2006).
51. Taguchi, A. et al. J. Clin. Invest. 114, 330–338
(2004).
52. Lo, E.H. Nat. Med. 14, 497–500 (2008).
53. Ikonomidou, C. & Turski, L. Lancet Neurol. 1, 383–386
(2002).
54. Waetzig, V., Zhao, Y. & Herdegen, T. Prog. Neurobiol.
80, 84–97 (2006).
55. Perry, V.H., Nicoll, J.A. & Holmes, C. Nat. Rev. Neurol.
6, 193–201 (2010).
56. Williams, A.J. & Paulson, H.L. Trends Neurosci. 31,
521–528 (2008).
57. Fotuhi, M., Hachinski, V. & Whitehouse, P.J. Nat. Rev.
Neurol. 5, 649–658 (2009).
58. Benatar, M. Neurobiol. Dis. 26, 1–13 (2007).
59. de Calignon, A. et al. Nature 464, 1201–1204
(2010).
60. Wang, Y., Kruger, U., Mandelkow, E. & Mandelkow, E.M.
Neurodegener. Dis. 7, 103–107 (2010).
61. Lemere, C.A. & Masliah, E. Nat. Rev. Neurol. 6, 108–
119 (2010).
62. Zahs, K.R. & Ashe, K.H. Trends Neurosci. 33, 381–399
(2010).
63. Hampel, H. et al. Nat. Rev. Drug Discov. 9, 560–574
(2010).
64. O’Collins, V.E. et al. Ann. Neurol. 59, 467–477
(2006).
65. Fisher, M. et al. Stroke 40, 2244–2250 (2009).

The benefits and limitations of animal
models for translational research in
neurodegenerative diseases
Mathias Jucker
Age-related neurodegenerative
diseases are largely limited to humans
and rarely occur spontaneously in
animals. Genetically engineered mouse
models recapitulate aspects of the
corresponding human diseases and
are instrumental in studying disease
mechanisms and testing therapeutic
strategies. If considered within the
range of their validity, mouse models
have been predictive of clinical
outcome. Translational failure is less the
result of the incomplete nature of the
models than of inadequate preclinical
studies and misinterpretation of the
models. This commentary summarizes
current models and highlights key
questions we should be asking about
animal models, as well as questions that
cannot be answered with the current
models.
Natural animal models
Age-related neurodegenerative diseases are
largely human-specific diseases. Although
aged nonhuman primates and some other
higher-order animal species show aspects
similar to those of human brain aging, these
animals generally do not readily develop the
full neuropathological or clinical phenotypes
seen in humans1,2 (with a few possible excep-
Mathias Jucker is at the Department of Cellular
Neurology, Hertie Institute for Clinical Brain
Research, University of Tübingen, and the DZNE-
German Center for Neurodegenerative Diseases,
Tübingen, Germany.
e-mail: mathias.jucker@uni-tuebingen.de
Published online 21 September 2010;
doi: 10.1038/nm.2224
nature medicine advance online publication 1
C O M M E N TA RY
tions, such as mutant SOD-linked canine
degenerative myelopathy, resembling amyo-
trophic lateral sclerosis (ALS))3.
Aged mammals, similar to aged, cognitively
normal humans, can develop the neuropatho-
logical lesions that define human Alzheimer’s
disease. For example, cerebral β-amyloidosis
spontaneously occurs in aged nonhuman
primates, bears and dogs4–7. Neurofibrillary
tangles have been described in aged nonhu-
man primates, bears and sheep4,8–10. No such
spontaneously occurring lesions have been
reported in aged laboratory rodents or lower-
order species.
Paradoxically, the paucity of age-related
neurodegenerative diseases in nonhuman
species could yield some mechanistic insights
into the etiology of the human diseases. At
least for nonhuman primates, the amino acid
sequences of the disease-defining pathogenic
proteins are very similar to, or predict an even
more pathogenic protein than, the human
sequences11–14. Differences in chaperones,
proteasomal and autophagosomal clearance
mechanisms, lifespan or comorbid conditions
are considered causes for the apparent resis-
tance of nonhumans to age-related neurode-
generation.
Familial cases are similar to idiopathic
cases
The majority of human neurodegenera-
tive diseases are sporadic and of unknown
etiology. However, aside from age of onset,
idiopathic Alzheimer’s disease, Parkinson’s
disease, frontotemporal lobar degeneration
(FTLD) and ALS are clinically and neuro-
pathologically similar to their most common
familial forms15–18. Phenotypic similarities
between genetic and idiopathic variants, in
combination with a paucity of natural ani-
mal models, have driven the development of
genetically modified animal models based on
familial disease mutations.
Gentically engineered animal models
Drosophila melanogaster, Caenorhabditis elegans
and zebrafish have been useful in dissecting
basic disease mechanisms and in screening com-
pounds that target such basic mechanisms19,20.
However, the much simpler nervous systems of
nonmammalian species and the species’ phylo-
genetic distance from humans limit their use as
translational models of human-specific neuro-
degenerative diseases.
The most popular animal models for study-
ing human age-related neurodegenerative
diseases are models involving genetically engi-
neered mice17,19,21–27. For an updated list of
current models, see http://www.alzforum.org/
res/com/tra/default.asp. Some general findings
relevant to translational research with current
mouse models are highlighted below.
Alzheimer’s disease and cerebral
β-amyloidosis. The discovery that mutations in
the genes encoding amyloid-β precursor protein
(APP) and presenilins 1 or 2 (PSEN1 or PSEN2)
cause autosomal dominant Alzheimer’s disease
has bolstered the amyloid-β cascade hypothesis
of Alzheimer’s disease28. Today, there are at least
25 known mutations or genomic duplications
of APP and over 170 mutations in PSEN1 or
PSEN229. Most transgenic mouse models
overexpress mutant human APP, or PSEN1 or
PSEN2, or a combination of these. The overly
hopeful expectation that the introduction
or overexpression of wild-type or mutated
human proteins would be sufficient to induce
a human-like form of dementia in mice has not
been realized.
What these transgenic mice develop remark-
ably well, however, is cerebral β-amyloidosis,
commentary

1999–2000
a b c d e 2004–2005
2008–2009

Number of APP-transgenic models

80

Blinded plaque load estimation

120 100 7
70 100

per study (% publications)

Number of mice per group
Gender (% publications)
100 80 60 6
(% publications)
Randomization

(% publications)
80
50 5
80 60

(median)
40 4 60
60
40 30 3
40 40
20 2
20 20
20 10 1
0 0 0 0 0
Yes NS Male Female Both NS Yes NS n=1 n≥2

Figure 1 Studies from PubMed were retrieved using the keywords ‘mouse’, ‘amyloid’ and ‘Alzheimer’s disease’ for the years 1996–2009. Judged from the
abstracts, studies were selected that showed modulation of Aβ levels, Aβ deposition or both either through a genetic cross or through pharmacological or
nutritional interventions. Only studies in high-impact journals were selected (defined as having an impact factor >7.48), and these studies were fully and
carefully read. Shown are the data for the years 1999–2000 (n = 13), 2004–2005 (n = 43) and 2008–2009 (n = 59). (a) Less than 20% of the studies
indicate that mice have been randomly assigned to groups (NS, not specified). (b) The gender of the mice in most studies is not indicated, and mice of both
sexes were used in less than 10% of the studies. (c) The investigator seemed to be blinded toward treatment allocations and evaluation outcome in only one
third of the studies. (d) The number of mice per group varied from experiment to experiment within a study, and there were often only four to six mice per
group. (e) Two or more mouse models were examined in only 20% of the studies.

one of the hallmark pathological features of
Alzheimer’s disease. Currently, a variety of
mouse models generate β-amyloid plaques,
β-amyloid angiopathy or a mixture of both21.
The amyloid lesions are associated with dystro-
phic neurites, loss of dendritic spines, synaptic
degeneration and robust neuroinflammation.
Amyloid in the vessel wall leads to a loss of
smooth muscle cells and can lead to vascular
rupture and microbleeds. All of these patholo-
gies closely resemble the amyloid-associated
lesions in the Alzheimer’s disease brain21,30.
Despite these similarities, the commonly used
positron-emission tomogrophy (PET) imaging
ligand Pittsburgh Compound B (PIB) reveals
much stronger binding to the amyloid lesions
in Alzheimer’s disease brains than in mouse
brains31, suggesting structural and/or biochem-
ical differences between mouse and human
lesions32,33. PIB also binds poorly to the amy-
loid in the nonhuman primate brain and does
not make nonhuman primates a better model
for this aspect of the disease34. Interestingly,
structural and biochemical β-amyloid variants
(including poor PIB binding) have also recently
been identified in the human brain35–37.
Unlike in Alzheimer’s disease38,39, neuronal
loss in APP-transgenic mice is modest and is
confined mainly to the hippocampus21,40,41.
Also, unlike in Alzheimer’s disease42, many
APP-transgenic mouse lines are behaviorally
impaired even before they deposit Aβ in the
brain43,44. The findings that the infusion of
soluble oligomeric forms of Aβ impairs cog-
nition45,46 and that Aβ-specific antibodies
and γ-secretase inhibitors rapidly attenuate
the memory impairment of APP-transgenic
mice47,48 support the argument that that soluble
Aβ species, independent of Aβ deposition, are
sufficient to impair cognition in mice. Until the
role of soluble Aβ species in human cognitive
decline is clarified, the relevance of behavioral
impairments in APP-transgenic mouse models
to the cognitive impairment in Alzheimer’s dis- likely that cortical synucleinopathy drives the
ease remains uncertain. DLB as a complication of Parkinson’s disease
Parkinson’s disease and synucleinopathies. or as a distinct disease itself50. Thus, reports of
α-synuclein lesions are hallmarks of Parkinson’s behavioral abnormalities in α-synuclein trans-
disease, dementia with Lewy bodies (DLB) genic mice have stimulated interest in using
and multiple system atrophy. The discovery of this model to analyze nonmotor sequelae of
point mutations and genomic multiplications α-synucleinopathy23.
in the gene encoding α-synuclein in familial The issue of distinguishing between changes
Parkinson’s disease has sparked the genera- that cause disease and those that are comorbid is
tion of various α-synuclein–transgenic mice24. also complicated by the recognition of familial
Common haplotypes in the α-synuclein gene forms of Parkinson’s disease linked to mutations
have now been associated with an increased in LRRK2, PARKIN, PINK1 and PARK7 (also
risk of developing Parkinson’s disease49, known as DJ-1)18. Degeneration of nigrostri-
strengthening the concept of a common patho- atal dopaminergic neurons is common to all of
genesis of both familial and sporadic forms of these familial forms of Parkinson’s disease, but
the disease. the underlying mechanisms may be distinctly
The overexpression of human wild-type and different, as Lewy bodies are not found in all
mutant α-synuclein in transgenic mice leads to of these familial forms of Parkinson’s disease24.
synuclein lesions that resemble the Lewy bodies Mouse models exploiting these more recent
and Lewy neurites found in Parkinson’s disease mutations have been generated, and although
and DLB and the glial inclusions found in multi- these mice models have provided mechanistic
ple system atrophy22,24. α-synuclein–transgenic insights, they do not appear to show dopamin-
mice show progressive age-dependent neuropa- ergic cell loss in the substantia nigra.
thology and cognitive and locomotor dysfunc- FTLD-tau and tauopathies. The aberrant
tions, but the relationship of these symptoms polymerization of the microtubule-associated
to the α-synuclein lesions is not clear24. Even protein tau (MAPT) is a defining lesion in
in mouse models with α-synuclein expression Alzheimer’s disease, progressive supranuclear
confined to the substantia nigra, dopaminergic palsy, corticobasal degeneration and about
neuron loss in the nigrostriatal system resem- 40% of FTLD (now called FTLD-tau), includ-
bling that in Parkinson’s disease is not appar- ing Pick’s disease and frontotemporal dementia
ent22,50. with parkinsonism linked to chromosome 17
After an initial wave of creation of (FTDP-17). Tau in the human brain is expressed
α-synuclein–transgenic mice, progress in in three-repeat and four-repeat splice variants
transgenic modeling of Parkinson’s disease in a cell type–specific manner51. In contrast,
has decelerated because of uncertainty about mice express only the four-repeat tau isoforms.
what is causing the disease and what is simply Tau aggregation is dependent on the isoforms
related to the symptoms of Parkinson’s disease. involved, and the cell- and region-specific com-
The view that Parkinson’s disease is an exclu- position of tau filaments is distinct for the vari-
sively motor disease affecting the dopaminergic ous tauopathies51.
system has recently been challenged50. Whereas The discovery of MAPT mutations in FTDP-
dopaminergic neuron loss in the substantia 1752,53 initiated the generation of various trans-
nigra is the probable cause of the early-onset genic mice overexpressing mutated human tau
motor problems in Parkinson’s disease, it is isoforms. These mice develop neurofibrillary

2 advance online publication nature medicine


tangles (NFTs) that resemble those seen in
FTLD-tau and Alzheimer’s disease. The initial
models showed high tau expression in brain-
stem and motor neurons and developed severe
motor phenotypes with premature morbid-
ity. More recent models develop NFTs in the
hippocampus and neocortex and model more
closely the lesions in FTLD-tau and Alzheimer’s
disease19, with the caveat that the NFTs in
Alzheimer’s disease and most cases of FTLD-
tau consist of wild-type tau. NFTs consisting of
wild-type tau were successfully generated in at
least two studies, but these models are compli-
cated by a complex genetic background54 and
the fact that NFTs appear only after 20 months
of age55. Transgenic mice expressing mutated
tau show cognitive impairments and neuronal
death, although a discrepancy between tau-
bearing neurons and cell death suggests that
soluble tau or the expression of the tau trans-
gene, rather than the presence of NFTs, is the
neurotoxic factor25,26,56,57.
FTLD-ALS disease spectrum and TDP-43
proteinopathies. TDP-43–positive cytoplasmic
inclusions are hallmarks of most sporadic ALS
and the most common subtypes of FTLD (now
called FTLD-TDP) and underscore the link
between the two diseases, which is the occur-
rence of motor neuron disease in many indi-
viduals with FTLD and cognitive impairment
in many individuals with ALS27. Mutations in
the TDP-43 gene (TARDBP) cosegregate with
both ALS and FTLD-TDP53. The majority of
cases of familial FTLD-TDP, however, are linked
to mutations in the progranulin gene (GRN),
underlining the heterogeneous pathomecha-
nism of the disease27. Infrequent forms of
familial ALS are caused by mutations in the
superoxide dismutase-1 (SOD1) gene and, more
rarely, the fused in sarcoma (FUS) gene58.
The first mouse models in the new and rap-
idly developing field of TDP-43 proteinopathies
have shown that overexpression of both wild-
type and mutant TDP-43 can lead to TDP-43
inclusions as well as cortical and motor neuron
loss and therefore resemble some features of
the FTLD-TDP and ALS disease spectrum59,60.
Similar to synuclein- and tau-transgenic mouse
models, the neurotoxic entity and the causal
link between pathogenic protein and neuron
death have not been established. GRN muta-
tions are thought to be loss-of-function muta-
tions, but the Grn-null mice analyzed thus far
have no clear FTLD-TDP phenotype61,62.
SOD1 mutations in ALS were first discovered
almost two decades ago63, and myriad transgenic
mutant SOD1-overexpressing mice have since
been generated17. These various mouse lines
show progressive hindlimb tremor and weak-
ness, locomotor deficits and paralysis followed
by premature death. Neuropathologically, these
nature medicine advance online publication 3
Box 1 Predictive validity of mouse models for clinical trials
Internal validity
Randomization
Blinded evaluation
Gender and genetic background
Sample size and statistics
Health status of mice
mice show SOD1-positive inclusions, motor
neuron loss and gliosis. Neurodegeneration in
the models is mediated through a convergence
of several gain-of-toxic-function mechanisms
acting in neurons and also through a non–cell-
autonomous dysfunction of glial cells64. Thus,
the mutant SOD1 models replicate remarkably
well the key clinical and neuropathological fea-
tures of this familial form of ALS. However, the
absence of TDP-43 inclusions in this familial
variant may indicate a pathomechanism dif-
ferent from idiopathic ALS, so it is possible
that treatments that work in these models may
not be applicable to the sporadic form of the
disease, which represents the vast majority of
affected individuals.
The advantages of incomplete mouse
models
Today’s genetically engineered mouse mod-
els recapitulate primarily the disease-defining
lesions and associated pathologies. Neuron loss
is apparent in some of these models, but the
actual neurotoxic entity is often uncertain. With
the exception of SOD1-mutant ALS mice, ani-
mal models only partially recapitulate the com-
plex clinical features of the human diseases.
Large-scale efforts are ongoing to function-
ally annotate every mouse gene in the context
of the whole organism and of various environ-
ments65, and entire pathways and organs in
the mouse have already been humanized66,67.
Thus, the next generation of mouse models
is expected to recapitulate human neurode-
generative diseases more closely. However, the
extent to which complex, human-specific neu-
rodegenerative diseases can be phenotypically
modeled in rodents remains an open question.
As it is important to distinguish disease causes
from effects, incomplete models of diseases may
actually be advantageous models of specific dis-
ease processes.
For example, although the spatial and tempo-
ral expression of mutated genes in the human
brain can be modeled remarkably well using
genomic-based transgenic mouse models68,
mouse models using cell- and region-specific
promoters have shown that lesions and entire
disease processes are non–cell autonomous.
For instance, the cell type–specific expression
of mutant SOD1 has demonstrated that SOD1-
commentary
External validity
Models predict what they model
Use of several mouse models
Outcome measure of clinical relevance
Hypothesis-driven design and statistics
Proof-of-principle versus preclinical trial
induced toxicity in glial cells acts on motor
neurons and accelerates disease progression64.
Therefore, mutant SOD1 in neurons deter-
mines disease onset and the early progression
of disease, but the later course of the disease
is governed instead by neighboring glial cells.
Non–cell-autonomous toxicity may also occur
in synucleinopathies and tauopathies69–72,
and this concept has garnered more support
by the recent findings of exogenous induc-
tion and transcellular spread of proteopathic
lesions73–75.
Mouse models bearing multiple transgenes
and showing multiple lesions can mimic the
neuropathology of the human diseases more
closely than do models bearing single trans-
genes (for an example, see ref. 76). However, the
interpretation of such models is complicated by
uncertainty about the interrelationship of the
various transgenes and lesions. The observation
that APP-transgenic mice do not develop NFTs
and thus fail to fully model Alzheimer’s disease
pathology might be seen as a disappointment.
The mechanistic interplay between Aβ and tau
lesions can, however, be studied exemplarily by
cross-breeding APP-transgenic mice with tau-
transgenic mice and comparing the resulting
offspring with the parental single-transgenic
mouse lines77–80. The enhancement of NFT
formation in the presence of Aβ aggregates has
become a milestone observation, albeit one that
is mechanistically still unclear. Unraveling these
mechanisms is likely to provide prime targets
for disease-modifying drugs.
Incomplete mouse models in translational
research
Successful translation. An example of a rapid
and fairly predictive translation of preclinical
findings into clinical studies comes from APP-
transgenic mice. Only four years after the gen-
eration of the first successful APP-transgenic
mouse model81, it was reported that both
active and passive immunization against Aβ
reduces cerebral amyloidosis in APP-transgenic
mice82,83. This finding has been replicated by
many laboratories using different mouse mod-
els and immunization protocols84.
Translation of this milestone laboratory
finding to the clinic was remarkably rapid.
In 2003, a case report provided evidence that,
commentary
consistent with previous mouse studies, active
immunization of patients with moderate
Alzheimer’s disease reduces cerebral β-amyloid
levels, presumably through a microglia-medi-
ated mechanism85. Unfortunately, the clini-
cal trials were halted due to the unexpected
occurrence of acute meningoencephalitis in
a subset of patients86. Encephalitis was not
predicted by the mouse models, and only later
was this side effect reported to occur in mice
under rare circumstances87,88. However, the
mouse models predicted that immunization
clears primarily parenchymal amyloid rather
than vascular amyloid and, in fact, that it may
increase cerebral amyloid angiopathy-related
micro­bleeds89,90. There is now convincing
evidence that immunization reduces cerebral
amyloid load in humans with, at least tempo-
rarily, an increase in cerebral amyloid angiop-
athy and microbleeds91,92. Moreover, detailed
postmortem analyses of immunized human
cases have found not only a reduction of amy-
loid lesions but also of the amyloid-associated
neuropathology, again showing remarkably
similar results to those predicted from mouse
studies91,93,94. Recently, vasogenic edema was
reported as an adverse side effect of passive
immunization that was evident by magnetic
resonance imaging in humans95. The mouse
counterpart of this phenomenon is currently
not clear, as imaging in immunized mice has
rarely been done, and the neuropathological
features of vasogenic edema in immunized
humans await further analysis.
APP-transgenic mouse models of cerebral
β-amyloidosis have predicted the outcome of
Alzheimer’s disease immunization trials rea-
sonably well. Whether APP-transgenic mice
are useful for modeling the cognitive decline in
Alzheimer’s disease is less clear. If it turns out
that the current vaccination approach benefits
cognition in humans, the predictive value of
the mouse models for this critical endpoint
will also be affirmed (as suggested by D. Schenk
(Elan Pharmaceutical) at the Herrenhausen
Symposium on Neurodegeneration).
Failed translation? Despite the success in
predicting important outcomes of Aβ immu-
nization in Alzheimer’s disease clinical trials, it
has been argued that mouse models of human
neurodegenerative diseases have largely failed
to predict the efficacy of clinical trials17,96.
However, this apparent inconsistency appears to
represent less a failure of the mouse models per
se than a consequence of inadequate preclinical
studies and misinterpretation of the models.
The translational failure of apparently prom-
ising animal studies has at least partly been
attributed to the inadequate internal and exter-
nal validity of preclinical studies97. Indeed, sys-
tematic analysis of the internal validity of studies
4 advance online publication nature medicine
with APP-transgenic mice (Box 1 and Fig. 1a)
revealed that only a minority of the studies state
that the mice were randomly assigned to groups
before treatment. In most studies, the gender
of the mice was not indicated, even though Aβ
deposition in mouse models is often gender
dependent (Box 1 and Fig. 1b)98,99. In many
instances, it is unclear whether the investigators
were blinded toward treatment allocation and
evaluation of the outcome (Box 1 and Fig. 1c).
The group sizes were often as low as four to six
mice per group (Fig. 1d), and this number var-
ied from one experiment to the next within a
paper, raising the concern that mice were added
to a group until the desired effect was achieved
without proper statistical sample size calcula-
tions. Rarely were two or more mouse models
examined in any given study (Fig. 1e), and the
health status of the mice was rarely indicated
(Box 1).
The odds of ‘positive’ results are higher in
studies with methodological flaws97, implying
that many animal studies may not withstand
rigorous replication. Indeed, apart from immu-
nization studies of APP-transgenic mice, there
are more opposing than confirmatory reports
found in the literature, although a publication
bias toward new results cannot be excluded96,97.
A similar assessment has been made of recent
analyses of preclinical studies using SOD1
mutant mouse models of ALS100,101.
The external validity of a model refers to what
a model is good for97. However, the limitations
of models are rarely acknowledged, and the
predictive validity of these models for humans
is often overstated or misinterpreted, raising
false or premature hopes for clinical efficacy.
APP-transgenic mice are models of cerebral
β-amyloidosis and not of Alzheimer’s disease.
Furthermore, a treatment that is effective in
female mice may not necessarily predict efficacy
in male subjects. Similarly, a treatment assessed
in one mouse model may not be replicable in
another model, and the outcome measure of a
treatment in a mouse model may be of no use
for clinical translation (Box 1).
Therapeutic intervention studies in SOD1-
transgenic mouse models have shown that in
almost all instances of an apparent benefit in
extending survival, treatment was initiated
early and delayed the onset, but not the rate,
of disease progression100; nevertheless, most
clinical studies have been initiated well after
disease onset. Although some researchers have
used failure of the human trials to discount the
utility of the transgenic mice, others (such as
D. Cleveland (University of California–San
Diego) at the Herrenhausen Symposium on
Neurodegeneration) argue the mouse has been
a near perfect predictor of success at trial—
treatments that fail to slow disease progres-
sion in mice, fail to slow disease progression
in humans. Similarly, in light of the limited
efficacy of intervention trials in humans with
ongoing Alzheimer’s disease, the success of
prevention experiments in mice argues per-
suasively for preventive immunotherapy trials
in humans96.
Need for large preclinical mouse studies
A promising avenue to improve the predic-
tive value of mouse models for clinical studies
would be to adapt aspects of the design, quality
control and transparency of clinical studies in
humans.
For example, the internal validity of preclini-
cal mouse studies could be improved by estab-
lishing a checklist for quality control issues
(Box 1)97,101. In addition, a minimal standard-
ization in methodology and quality control of
preclinical mouse trials for later comparisons,
interpretation and potential meta-analysis tech-
niques would be desirable, including the report
of negative results.
The external validity of mouse studies could
benefit from including more than one mouse
model, testing at multiple sites and using ade-
quately powered designs to confirm treatment
effects and to refute statistical flukes. Recently,
for example, the extension of the lifespan of mice
by rapamycin was replicated simultaneously
at three test sites on two genetically different
mouse lines of different gender and according
to predetermined standardization102.
Most importantly, the outcome measures of
preclinical studies should focus on parameters
and biomarkers with proven relevance in clini-
cal settings or at least those which can be tested
in the clinic. For example, β-amyloid imaging
and cerebrospinal fluid (CSF) measurement of
Aβ and tau have become valuable biomarkers
for the diagnosis and progression of Alzheimer’s
disease103,104. However, few mouse studies have
analyzed Aβ and tau in CSF, despite the pros-
pect that such studies may help to clarify the
mechanisms governing changes in protein lev-
els in human CSF. Rather, a primary endpoint
of most mouse studies is postmortem neuro-
pathology, which is arguably not very useful for
translational purposes.
How could such large and admittedly expen-
sive preclinical studies be done? One option is
to establish specialized testing facilities similar
to the various mouse phenotype clinics that
already exist105. Such centers could accommo-
date large colonies of aged mice and use differ-
ent mouse models for validation. The inclusion
of more aged mice in general is worth consid-
ering, as aging is the strongest risk factor for
neurodegenerative diseases, and neurodegen-
erative disease phenotypes can differ in young
and aged individuals50.

Alternatively, individual laboratories could
seek partner laboratories and initiate inter-
ventions at multiple sites with different mouse
lines and according to predefined criteria. Such
multicenter approaches increase quality con-
trol and have the advantage of including mice
housed in different environments and possibly
mice of different health status106. In this way,
both the internal and external validity of the
studies would be enhanced. However, although
the workload would be divided among labora-
tories, it may still be financially prohibitive to
conduct such large studies on aged animals in
an academic university environment.
Small, innovative treatment studies by indi-
vidual laboratories should by no means be
replaced by large, centralized or multicenter
preclinical mouse studies. However, the limita-
tions of such smaller studies should be clearly
acknowledged by, for example, differentiating
between proof-of-concept studies and preclini-
cal testing studies101. Clear editorial policies
established by journals on these issues, and their
enforcement by critical reviewers, will improve
the translational value of animal studies for
neurodegenerative diseases.
ACKNOWLEDGMENTS
I would like to thank L. Walker and M. Staufenbiel for
various discussions and help with this manuscript.
The systematic review of the literature by R. Radde is
greatly acknowledged. I also thank M. Neumann,
V. Lee, J. McLaurin, D. Schenk, D. Thal, J. Götz,
D. DiMonte, M. Goedert, T. Gasser and P. Kahle for
comments on various parts of this commentary and
the attendees of the Herrenhausen Symposium on
Neurodegeneration for the inspiring discussions from
which this commentary has emerged. The work was
supported by the German National Genome Network
(NGFNPlus).
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. Walker, L.C. & Cork, L.C. in Alzheimer Disease (eds.
R.D. Terry et al.) 233–243 (Lippincott Williams and
Wilkins, Philadelphia, Pennsylvania, USA, 1999).
2. Gerlach, M. & Riederer, P. J. Neural Transm. 103,
987–1041 (1996).
3. Awano, T. et al. Proc. Natl. Acad. Sci. USA 106,
2794–2799 (2009).
4. Cork, L.C. et al. J. Neuropathol. Exp. Neurol. 47,
629–641 (1988).
5. Price, D.L. et al. Brain Pathol. 1, 287–296 (1991).
6. Cummings, B.J., Su, J.H., Cotman, C.W., White,
R. & Russell, M. J. Neurobiol. Aging 14, 547–560
(1993).
7. Walker, L.C. Brain Res. Brain Res. Rev. 25, 70–84
(1997).
8. Nelson, P.T., Greenberg, S.G. & Saper, C.B. Neurosci.
Lett. 170, 187–190 (1994).
9. Schultz, C., Hubbard, G.B., Tredici, K.D., Braak, E. &
Braak, H. Adv. Exp. Med. Biol. 487, 59–69 (2001).
10. Rosen, R.F. et al. J. Comp. Neurol. 509, 259–270
(2008).
11. Holzer, M., Craxton, M., Jakes, R., Arendt, T. &
Goedert, M. Gene 341, 313–322 (2004).
12. Podlisny, M.B., Tolan, D.R. & Selkoe, D.J. Am. J.
Pathol. 138, 1423–1435 (1991).
nature medicine advance online publication 5
13. Fukuhara, R., Tezuka, T. & Kageyama, T. Gene 296,
99–109 (2002).
14. Hamilton, B.A. Genomics 83, 739–742 (2004).
15. Ryan, N.S. & Rossor, M.N. Biomark. Med. 4, 99–112
(2010).
16. Mackenzie, I.R.A. Lancet Neurol. (in the press).
17. Turner, B.J. & Talbot, K. Prog. Neurobiol. 85, 94–134
(2008).
18. Gasser, T. Expert Rev. Mol. Med. 11, e22 (2009).
19. Götz, J. & Ittner, L.M. Nat. Rev. Neurosci. 9, 532–544
(2008).
20. Teschendorf, D. & Link, C.D. Mol. Neurodegener. 4,
38 (2009).
21. Duyckaerts, C., Potier, M.C. & Delatour, B. Acta
Neuropathol. 115, 5–38 (2008).
22. Kahle, P.J. Acta Neuropathol. 115, 87–95 (2008).
23. Chesselet, M.F. Exp. Neurol. 209, 22–27 (2008).
24. Dawson, T.M., Ko, H.S. & Dawson, V.L. Neuron 66,
646–661 (2010).
25. Zilka, N., Korenova, M. & Novak, M. Acta Neuropathol.
118, 71–86 (2009).
26. Denk, F. & Wade-Martins, R. Neurobiol. Aging 30,
1–13 (2009).
27. Chen-Plotkin, A.S., Lee, V.M. & Trojanowski, J.Q. Nat.
Rev. Neurol. 6, 211–220 (2010).
28. Hardy, J. & Selkoe, D.J. Science 297, 353–356
(2002).
29. Bertram, L. & Tanzi, R.E. Nat. Rev. Neurosci. 9, 768–
778 (2008)
30. Herzig, M.C., Van Nostrand, W.E. & Jucker, M. Brain
Pathol. 16, 40–54 (2006).
31. Klunk, W.E. et al. J. Neurosci. 25, 10598–10606
(2005).
32. Kuo, Y.M. et al. J. Biol. Chem. 276, 12991–12998
(2001).
33. Maeda, J. et al. J. Neurosci. 27, 10957–10968
(2007).
34. Rosen, R.F., Walker, L.C. & Levine, H. 3rd. Neurobiol.
Aging published online, doi:10.1016/j.neurobio
laging.2009.02.011 (27 March 2009).
35. Piccini, A. et al. J. Biol. Chem. 280, 34186–34192
(2005).
36. Levine, H. III & Walker, L.C. Neurobiol. Aging 31,
542–548 (2010).
37. Rosen, R.F. et al. Acta Neuropathol. 119, 221–233
(2010).
38. Gómez-Isla, T. et al. J. Neurosci. 16, 4491–4500
(1996).
39. West, M.J., Coleman, P.D., Flood, D.G. & Troncoso,
J.C. Lancet 344, 769–772 (1994).
40. Calhoun, M.E. et al. Nature 395, 755–756 (1998).
41. Rupp, N.J., Wegenast-Braun, B.M., Radde, R.,
Calhoun, M.E. & Jucker, M. Neurobiol. Aging (in the
press).
42. Morris, J.C. et al. Arch. Neurol. 66, 1469–1475
(2009).
43. Ashe, K.H. Learn. Mem. 8, 301–308 (2001).
44. Chen, G. et al. Nature 408, 975–979 (2000).
45. Lesné, S. et al. Nature 440, 352–357 (2006).
46. Shankar, G.M. et al. Nat. Med. 14, 837–842
(2008).
47. Dodart, J.C. et al. Nat. Neurosci. 5, 452–457
(2002).
48. Comery, T.A. et al. J. Neurosci. 25, 8898–8902
(2005).
49. Simón-Sánchez, J. et al. Nat. Genet. 41, 1308–1312
(2009).
50. Obeso, J.A. et al. Nat. Med. 16, 653–661 (2010).
51. Goedert, M. & Spillantini, M.G. Science 314, 777–781
(2006).
52. Heutink, P. Hum. Mol. Genet. 9, 979–986 (2000).
53. Sleegers, K., Cruts, M. & Van Broeckhoven, C. Annu.
Rev. Neurosci. 33, 71–88 (2010).
54. Andorfer, C. et al. J. Neurochem. 86, 582–590
(2003).
55. Ishihara, T. et al. Am. J. Pathol. 158, 555–562
(2001).
56. Santacruz, K. et al. Science 309, 476–481 (2005).
57. de Calignon, A. et al. Nature 464, 1201–1204
(2010).
58. Bento-Abreu, A., Van Damme, P., Van Den Bosch, L.
commentary
& Robberecht, W. Eur. J. Neurosci. 31, 2247–2265
(2010).
59. Wils, H. et al. Proc. Natl. Acad. Sci. USA 107, 3858–
3863 (2010).
60. Wegorzewska, I., Bell, S., Cairns, N.J., Miller, T.M. &
Baloh, R.H. Proc. Natl. Acad. Sci. USA 106, 18809–
18814 (2009).
61. Ahmed, Z. et al. Am. J. Pathol. 177, 311–324
(2010).
62. Yin, F. et al. FASEB J. published online, doi:10.1096/
fj.10-161471 (28 July 2010).
63. Rosen, D.R. et al. Nature 362, 59–62 (1993).
64. Ilieva, H., Polymenidou, M. & Cleveland, D.W. J. Cell
Biol. 187, 761–772 (2009).
65. Abbott, A. Nature 465, 410 (2010).
66. Beckers, J., Wurst, W. & de Angelis, M.H. Nat. Rev.
Genet. 10, 371–380 (2009).
67. Traggiai, E. et al. Science 304, 104–107 (2004).
68. Hock, B.J. Jr. & Lamb, B.T. Trends Genet. 17, S7–S12
(2001).
69. Yazawa, I. et al. Neuron 45, 847–859 (2005).
70. Shults, C.W. et al. J. Neurosci. 25, 10689–10699
(2005).
71. Forman, M.S. et al. J. Neurosci. 25, 3539–3550
(2005).
72. Higuchi, M. et al. J. Neurosci. 25, 9434–9443
(2005).
73. Meyer-Luehmann, M. et al. Science 313, 1781–1784
(2006).
74. Clavaguera, F. et al. Nat. Cell Biol. 11, 909–913
(2009).
75. Aguzzi, A. & Rajendran, L. Neuron 64, 783–790
(2009).
76. Oddo, S. et al. Neuron 39, 409–421 (2003).
77. Lewis, J. et al. Science 293, 1487–1491 (2001).
78. Bolmont, T. et al. Am. J. Pathol. 171, 2012–2020
(2007).
79. Terwel, D. et al. Am. J. Pathol. 172, 786–798
(2008).
80. Coomaraswamy, J. et al. Proc. Natl. Acad. Sci. USA
107, 7969–7974 (2010).
81. Games, D. et al. Nature 373, 523–527 (1995).
82. Schenk, D. et al. Nature 400, 173–177 (1999).
83. Bard, F. et al. Nat. Med. 6, 916–919 (2000).
84. Brody, D.L. & Holtzman, D.M. Annu. Rev. Neurosci.
31, 175–193 (2008).
85. Nicoll, J.A. et al. Nat. Med. 9, 448–452 (2003).
86. Orgogozo, J.M. et al. Neurology 61, 46–54 (2003).
87. Furlan, R. et al. Brain 126, 285–291 (2003).
88. Lee, E.B., Leng, L.Z., Lee, V.M. & Trojanowski, J.Q.
FEBS Lett. 579, 2564–2568 (2005).
89. Pfeifer, M. et al. Science 298, 1379 (2002).
90. Racke, M.M. et al. J. Neurosci. 25, 629–636
(2005).
91. Boche, D. et al. Brain 131, 3299–3310 (2008).
92. Rinne, J.O. et al. Lancet Neurol. 9, 363–372
(2010).
93. Holmes, C. et al. Lancet 372, 216–223 (2008).
94. Serrano-Pozo, A. et al. Brain 133, 1312–1327
(2010).
95. Salloway, S. et al. Neurology 73, 2061–2070
(2009).
96. Zahs, K.R. & Ashe, K.H. Trends Neurosci. 33, 381–
389 (2010).
97. van der Worp, H.B. et al. PLoS Med. 7, e1000245
(2010).
98. Callahan, M.J. et al. Am. J. Pathol. 158, 1173–1177
(2001).
99. Wang, J., Tanila, H., Puolivali, J., Kadish, I. & van
Groen, T. Neurobiol. Dis. 14, 318–327 (2003).
100. Benatar, M. Neurobiol. Dis. 26, 1–13 (2007).
101. Ludolph, A.C. et al. Amyotroph. Lateral Scler. 11,
38–45 (2010).
102. Harrison, D.E. et al. Nature 460, 392–395 (2009).
103. Blennow, K. Nat. Med. published online, doi:10.1038/
nm.2221 (21 September 2010).
104. Perrin, R.J., Fagan, A.M. & Holtzman, D.M. Nature
461, 916–922 (2009).
105. Fuchs, H. et al. Curr. Pharm. Biotechnol. 10, 236–243
(2009).
106. Martin, B., Ji, S., Maudsley, S. & Mattson, M.P. Proc.
Natl. Acad. Sci. USA 107, 6127–6133 (2010).

The future of genetic research on
neurodegeneration
Christine Van Broeckhoven
Why, with all the progress in the field
of neurodegeneration, do we still lack
disease-modifying drugs that tackle the
primary defect of severe cell loss? How
much progress has been made toward
this goal? Have we spent our time and
resources wisely? And, most important,
is there room for improvement? This
commentary highlights several problems
faced by researchers in studying the
genetic etiology of neurodegenerative
diseases and seeks to provide direction
in overcoming some of these obstacles.
Over the past two decades the field of molecu-
lar genetics has contributed substantially to
our understanding of neurodegeneration.
The genetic basis of neurodegenerative dis-
eases such as Huntington’s disease, Alzheimer’s
disease, Parkinson’s disease, frontotemporal
lobar degeneration (FTLD) and amyotrophic
lateral sclerosis (ALS) is no longer questioned.
Genetic analysis of multiple generations of
affected families paved the way in identifying
new pathways and pathogenic cascades. For
instance, in Alzheimer’s disease, a consider-
able proportion of our current understanding
of the pathogenesis originates from genetic
breakthroughs in the early 1990s in a handful
of families with a rare early-onset dominant
form of the disease1–8.
Christine Van Broeckhoven is at the
Neurodegenerative Brain Diseases Group,
Department of Molecular Genetics, VIB, Antwerpen,
Belgium, and the Laboratory of Neurogenetics,
Institute Born-Bunge, University of Antwerp,
Antwerpen, Belgium.
e-mail: christine.vanbroeckhoven@molgen.vib-ua.be
Published online 21 September 2010;
doi: 10.1038/nm.2225
nature medicine advance online publication 1
C O M M E N TA RY
Yet, at least in the case of Alzheimer’s disease,
these early days of so-called ‘low-hanging fruit’
were followed by nearly 15 years without real
breakthroughs in genetics. Linkage analyses in
individuals with familial late-onset Alzheimer’s
disease revealed chromosomal regions (on chro-
mosomes 9, 10 and 12) that probably harbor a
gene associated with Alzheimer’s disease9, but,
ten years later, the genes explaining the observed
linkage signals have yet to be found. A flurry
of candidate gene–based association studies
were published, but for every positive report,
negative replication studies followed. Even the
initial wave of high-density chromosome- and
genome-wide association studies (GWASs) on
Alzheimer’s disease failed to deliver substan-
tial insight into the disease. Slight differences
in design notwithstanding, the only convinc-
ing association to be detected in these studies
was with apolipoprotein E (APOE) variant ε4,
the single widely accepted genetic risk factor
for Alzheimer’s disease. Although an optimist
could interpret this as a proof of concept, it is
distressing that all these efforts and resources
only produced confirmatory evidence for a risk
factor that was identified in 1993 (ref. 10).
New approaches for genetic analyses
Many obstacles have contributed to this slow
progress. A central concern is the inherent
difference between identifying a pathogenic
mutation that segregates with disease in a
monogenic family and finding a genetic risk
factor that explains no more than 1% of the
heritability of a trait in a population11. An
intuitive approach (albeit difficult to imple-
ment) in identifying risk factors with small
effect sizes is to increase the size of the study
population. In Alzheimer’s disease, thousands
of samples from different populations and
study designs have been pulled together, which
led to the identification of at least one (clus-
terin (CLU)), but probably three (complement
receptor 1 (CR1) and phosphatidylinositol-
binding clathrin assembly protein (PICALM))
previously undescribed risk genes12,13. Another
approach is to homogenize the study popula-
tion on the basis of a shared characteristic. This
was done in a GWAS on FTLD characterized
by pathology involving the TAR DNA-binding
protein TDP43, and led to the identification
of what might be the first risk gene for FTLD,
transmembrane protein-106B (TMEM106B)14.
To obtain a sufficient number of subjects for
statistical analyses, several concessions were
made in this study14. Clustering subjects by
pathology meant that the study population
was more variable than ideal in terms of clini-
cal presentation, population background and
inclusion of carriers of highly penetrant muta-
tions in genes giving rise to TDP43 pathology.
It is promising, though, that studies including
such a heterogeneous group of subjects and
controls have yielded putative risk genes.
If we continue down this road, what will
come next? A further increase in sample size
allows the detection of polymorphisms with
even smaller effects, as foreshadowed by meta-
analyses in other complex traits15. Analysis of
epistasis and pathophysiological pathways has
the potential to uncover additional small genetic
effects. But will these advances translate into
improved patient care? The first genome-wide
single nucleotide polymorphism (SNP)-based
pathway analyses on Alzheimer’s disease16,17
(presented at the International Conference on
Alzheimer’s Disease 2010) did not reveal new
pathomechanisms but rather reiterated long-
standing hypotheses about the involvement of
lipid metabolism and the immune response
in the pathogenesis of Alzheimer’s disease.
Nevertheless, earlier attempts to target these
pathways pharmacologically proved unsuc-
cessful18,19. Arguably, genetic profiling on the
basis of underlying patterns of susceptibility
might help to streamline the design of future
commentary
clinical trials by including subjects who would
benefit most from the compound under study,
but in reality we have not advanced sufficiently
to implement these tools in such a way.
The complexity of genetic risk
Geneticists often argue that finding a genetic
risk factor of small effect size does not neces-
sarily have to translate directly into an advance
in patient care. Rather, it may bring to light
important pathways for further study through
cell biology and to incorporate into new animal
models of disease.
There are several issues with this notion. First,
at what point are we sufficiently convinced that
a newly described genetic risk factor or pathway
warrants further investigation? Most research-
ers are reluctant ‘believers’ in genes that confer
a relatively small increase in risk observed in
association studies, in part because the high
odds ratios obtained in genetic association
studies of APOE ε4 unrealistically raised our
expectations, and because too many attempts
to find additional risk genes have failed. This
is true even when the association is observed
consistently across studies and populations. For
instance, genetic variability in sortilin-related
receptor, L(DLR class) A repeats-containing
(SORL1) has been repeatedly associated with
Alzheimer’s disease in populations of European,
African, Latino and Asian ancestry (http://www.
alzgene.org/). This strongly suggests that SORL1
is involved in the disease. Nevertheless, SORL1 is
not widely acknowledged as a genuine risk gene
for Alzheimer’s disease. As a consequence of this
hesitation to trust new associations, few genes
are carried forward into in-depth molecular
genetic analyses aimed at identifying function-
ally relevant sequence variants that can serve as
starting points for in vivo studies.
Second, the pathways that are brought to
light, although interesting, are difficult to
study through reductionist molecular biol-
ogy techniques. For example, take clusterin, a
chaperone molecule that is involved in a wide
range of physiological processes, any number of
which could be disturbed in Alzheimer’s disease.
Moreover, these processes are not always linear;
depending on the ratio of clusterin to amyloid-β
for instance, clusterin can have either proamy-
loidogenic or neuroprotective effects20. For now,
it remains undecided whether physiological
effects of risk alleles can be observed in a sim-
ple cellular model or whether they require more
complex modeling in a more natural context.
Third, the situation is complicated fur-
ther when (as is often the case) the strongest
association is observed for a SNP without
an obvious effect on the protein function or
expression, in genes of unknown function
(such as TMEM106B) or in intergenic regions.
2 advance online publication nature medicine
According to the Catalog of Published Genome-
Wide Association Studies (http://www.genome.
gov/gwastudies/), more than 50 genome-wide
significant hits for a wide range of traits have
been found at intergenic SNPs, substantiating
the notion that these regions may harbor as yet
unknown functional elements. Experimental
confirmation of small genetic effects is chal-
lenging enough in and of itself. It becomes even
more daunting in the absence of any obvious
testable hypothesis. Experimental modeling of
epistatic networks and long-range effects adds
another layer of complexity. At this rate, transla-
tion into meaningful biomarkers or therapeutic
approaches will still be a long way off.
The absence of a clear biological interpre-
tation does not diminish the relevance of a
finding. There are several opportunities for
geneticists to validate their findings and provide
more depth to their observations. Apart from
an obvious need for better annotation of the
genome in a multitude of cell types (an ongoing
effort at the RIKEN Institute in Japan), genetic
data can be combined with data generated from
proteomics or transcriptomics experiments in
samples from the same individuals. Additional
data can also be collected on other markers of
disease progression, including cerebrospinal
fluid (CSF) biomarkers, structural measures
such as hippocampal atrophy or ventricular
enlargement and measures of cognition, to
identify underlying pathways or pathomecha-
nisms and to increase confidence in the associa-
tion of genetic variants in these risk genes with
disease21–23. This calls for longitudinal collec-
tion of broad phenotypic data (including envi-
ronmental exposure, early-life history of disease
and medication), as well as repeated systematic
biosampling, not only of DNA and RNA, but
also of body fluids, skin (for induced pluri-
potent stem cell studies) and brain at autopsy.
This might require studies to be scaled down in
terms of sample size, but if the design includes
enrichment for genetic factors (for instance, an
early-onset cohort or a genetically isolated pop-
ulation), and a willingness to collaborate among
the researchers, the chance of identifying robust
genetic associations increases. The availability of
biosamples will also enable a more rapid trans-
lation of findings back to the subject.
A related point of consideration in the design
of genetic studies derives from the growing
awareness that the various neurodegenerative
diseases show marked overlap at the patho-
logical level, which suggests that they share
pathological mechanisms. Pathogenic muta-
tions in genes that are generally considered to
cause one type of neurodegenerative disease
have been identified in people clinically diag-
nosed with another type of neurodegeneration.
A mutation in the FTLD gene tau (MAPT) is
repeatedly detected in individuals with a clinical
phenotype of Alzheimer’s disease24, a preseni-
lin-1 (PS1) mutation has been detected in an
individual with autopsy-confirmed Pick’s dis-
ease25 and, for mutations in the gene encoding
progranulin (GRN), clinical heterogeneity is
frequently reported26. It is not inconceivable
that intermediate to low penetrant susceptibil-
ity alleles also contribute to more than one of
the clinically distinct syndromes, perhaps by
accelerating the neurodegenerative cascade
downstream of aberrant protein aggregation.
This raises the question of whether we should
perform our studies across the clinical diag-
nostic boundaries. The need to define an indi-
vidual’s phenotype is different from a clinical
perspective (at least in the absence of therapeu-
tics targeting the underlying causes) than from
a research perspective. By studying endopheno-
types (for instance, TDP-43 protein abundance,
cortical thickness or age of onset) regardless of
the clinical diagnosis, disease modifiers may
be discovered that can ultimately provide suit-
able targets for therapeutic intervention for a
broader range of patients.
Reexamining early-onset forms of disease
Undescribed genetic loci could also be identi-
fied by refocusing attention and resources on
unexplained early-onset forms of Alzheimer’s
disease. Perhaps we are not putting all the avail-
able resources for gene finding to use. Although
genetic research on Alzheimer’s disease in the
past decade has largely concentrated on the
complexity of late-onset disease and its low-
penetrant risk factors, much of the familial
clustering in early-onset Alzheimer’s disease
is still enigmatic. Prevalence estimates sug-
gest that more than 10% of individuals with
Alzheimer’s disease have early onset disease, but
only a limited proportion can be explained by
known mutations27, indicating that additional
studies of early-onset Alzheimer’s disease using
current high-throughput genetic tools might
prove fruitful. By analogy, recent progress in the
genetics of FTLD and ALS has been driven by
family-based studies28–34. These families might
also provide a suitable setting in which to dis-
cover genetic modifying factors, which perhaps
delay onset or explain intrafamilial heterogene-
ity in clinical presentation.
With technological advances in whole-
exome and genome sequencing, multiplex
families in which the disease-related mutation
is still unidentified because of broad linkage
peaks or the absence of simple coding muta-
tions in regions of interest35 can be resolved by
sequencing three or four affected relatives. This
approach might also be applied to less infor-
mative families (in which most relatives are
no longer available for linkage analysis) or to

identify the presence of one or more variants
of intermediate penetrance. When applied to
risk genes identified from GWASs, clustering
of trait-associated rare and common variants
in functional domains in these genes could
suggest new disease mechanisms. By analogy,
the observation of a predominant location of
pathogenic mutations in the C-terminal regions
of TARDBP and FUS suggests a role of aberrant
RNA processing in neurodegeneration36.
The possibility that rare variants of intermedi-
ate penetrance might be involved in Alzheimer’s
disease has received limited attention but war-
rants further investigation. For example, null
mutations in GRN are a highly penetrant cause
of FTLD28,29, but GRN missense mutations are
also observed at an increased frequency in indi-
viduals with FTLD and Alzheimer’s disease37,38.
The precise relevance of these GRN missense
mutations to neurodegeneration is still an area
of active investigation, but in silico and in vitro
evidence suggests that these mutations may
impair progranulin function and affect suscep-
tibility to neurodegeneration26. Interestingly,
serum progranulin concentrations in some
people carrying a GRN missense mutation are
intermediate between null mutation carriers
and nonmutated control samples39, underscor-
ing the idea that these variants might act as risk
factors of intermediate effect. This emphasizes
the pathogenic relevance of rare sequence
variants that do not fit in the picture of what is
considered pathogenic (in the case of GRN-null
mutations).
The first wave of next-generation sequencing
efforts is ongoing40. It is imperative to carefully
design these studies and not to move forward
simply because the technique and the money are
available. Such studies require careful thought
on which subjects to select, which source of
DNA to use and which technique to employ
(that is, do we expect only amino acid substi-
tutions or nonsense mutations?). And given the
many sequence variations that will be detected,
how do we go about separating the wheat from
the chaff? The challenges that follow sequenc-
ing should not be underestimated. We should
be prepared for the possibility that these tech-
niques uncover a different variety of genetic risk
variants that cannot easily be modeled with the
current approaches. Proving causality could be a
bottleneck. The availability of extended (endo-)
phenotypic data and biosamples may facilitate
the process of establishing the true pathogenic
variants, by allowing genotype-phenotype cor-
relation and functional characterization of the
detected variants directly in biosamples from
affected individuals.
The possibility of recessive mutations in
Alzheimer’s disease is basically unexplored. Two
recent studies on homozygous mutations in the
nature medicine advance online publication 3
gene encoding amyloid precursor protein (APP)
in people with Alzheimer’s disease41,42 support
this contention and suggest that rare, recessive
mutations in other genes may contribute to
‘sporadic’ Alzheimer’s disease. These mutations
can be traced in GWAS genotype data43,44 or by
whole-genome sequencing of selected individu-
als with the disease, possibly in genetically iso-
lated populations. In addition, other sources of
genetic variation remain to be explored. Copy
number variation and mutations in regula-
tory regions are more readily investigated in
Parkinson’s disease45, but there is evidence
that gene dosage effects are involved in other
neurodegenerative conditions as well, includ-
ing APP duplications in Alzheimer’s disease46
and MAPT duplication47 and GRN deletions
in FTLD48, as well as promoter mutations and
mutations that affect miRNAs or their binding
sites49. The first genome-wide studies on copy-
number variation in Alzheimer’s disease are
ongoing, and they may reveal more common
copy number polymorphisms that affect risk
(for example, ref. 50).
Additional questions remain. What are the
implications of alternative splicing, distant
regulatory elements or antisense transcription
for disease susceptibility? How important is
epigenetics in neurodegeneration? What is the
contribution of somatic mutations to the occur-
rence of disease? Is it sufficient to sequence DNA
extracted from lymphocytes? What could we
learn from sequence analysis or RNA profiling
in the aging brain?
These are just some of the questions we can
ask ourselves, and for many we have no answer
yet. But whether setting out to explore these
less well-defined concepts, or following more
conventional approaches, we should try to stay
close enough to the affected individual and
his or her disease, by incorporating strategies
for phenotyping and biobanking in the study
design in close collaboration with clinicians.
This allows a rapid initial assessment of the
biological relevance of the genetic findings in
biosamples, as well as an exploration of gen-
otype-phenotype correlations, putative bio-
markers or the effect of compounds on human
material, in a first attempt to translate the find-
ings to patient care.
ACKNOWLEDGMENTS
A number of the issues identified in this commentary
were discussed in a session on genetics at the
Herrenhausen Symposium on Neurodegeneration
with P. St. George-Hyslop, P. Heutink and
B. De Strooper, with added discussion points and
commentary provided by the attendees at the
conference. The research in my group has been
funded by the Flemish and Federal governments of
Belgium, the VIB, the Institute Born-Bunge and the
University of Antwerp, as well as by many national
and international governmental research funding
agencies and charity organizations. The main
commentary
contributors, however, are the many people with
neurodegenerative disease and their families that
generously participate in research and the efforts of
researchers to pave the way toward new therapeutics.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests
1. St. George-Hyslop, P. et al. Nat. Genet. 2, 330–334
(1992).
2. St. George-Hyslop, P. et al. Science 235, 885–890
(1987).
3. Goate, A. et al. Nature 349, 704–706 (1991).
4. Sherrington, R. et al. Nature 375, 754–760 (1995).
5. Van Broeckhoven, C. et al. Nat. Genet. 2, 335–339
(1992).
6. Levy-Lahad, E. et al. Science 269, 973–977 (1995).
7. Rogaev, E.I. et al. Nature 376, 775–778 (1995).
8. Van Broeckhoven, C. et al. Science 248, 1120–1122
(1990).
9. Bertram, L. & Tanzi, R.E. Hum. Mol. Genet. 13 Spec No
1, R135–R141 (2004).
10. Corder, E.H. et al. Science 261, 921–923 (1993).
11. Sleegers, K. et al. Trends Genet. 26, 84–93 (2010).
12. Harold, D. et al. Nat. Genet. 41, 1088–1093 (2009).
13. Lambert, J.C. et al. Nat. Genet. 41, 1094–1099
(2009).
14. Van Deerlin, V.M. et al. Nat. Genet. 42, 234–239
(2010).
15. Teslovich, T.M. et al. Nature 466, 707–713 (2010).
16. Lambert, J.C. et al. J. Alzheimers Dis. 20, 1107–1118
(2010).
17. Jones, L. et al. Alzheimers Dement. 6, S113 (2010).
18. Aisen, P.S. Lancet Neurol. 1, 279–284 (2002).
19. Trompet, S. et al. J. Neurol. 257, 85–90 (2010).
20. Yerbury, J.J. et al. FASEB J. 21, 2312–2322 (2007).
21. Kauwe, J.S. et al. J. Alzheimers Dis. published online,
doi:10.3233/JAD-2010-091711 (15 July 2010).
22. Potkin, S.G. et al. PLoS One 4, e6501 (2009).
23. Shulman, J.M. et al. PLoS One 5, e11244 (2010).
24. Rademakers, R. et al. Hum. Mutat. 22, 409–411
(2003).
25. Dermaut, B. et al. Ann. Neurol. 55, 617–626 (2004).
26. Sleegers, K., Cruts, M. & Van Broeckhoven, C. Annu.
Rev. Neurosci. 33, 71–88 (2010).
27. Bettens, K., Sleegers, K. & Van Broeckhoven, C. Hum.
Mol. Genet. 19, R4–R11 (2010).
28. Baker, M. et al. Nature 442, 916–919 (2006).
29. Cruts, M. et al. Nature 442, 920–924 (2006).
30. Van Deerlin, V.M. et al. Lancet Neurol. 7, 409–416
(2008).
31. Sreedharan, J. et al. Science 319, 1668–1672
(2008).
32. Kabashi, E. et al. Nat. Genet. 40, 572–574 (2008).
33. Kwiatkowski, T.J. Jr. et al. Science 323, 1205–1208
(2009).
34. Vance, C. et al. Science 323, 1208–1211 (2009).
35. Rademakers, R. et al. Am. J. Hum. Genet. 77, 643–652
(2005).
36. Sleegers, K. & Van Broeckhoven, C. Nature 458, 415–
417 (2009).
37. Brouwers, N. et al. Neurology 71, 656–664 (2008).
38. van der Zee, J. et al. Hum. Mutat. 28, 416 (2007).
39. Sleegers, K. et al. Ann. Neurol. 65, 603–609 (2009).
40. Zuchner, S., Beecham, G., Haines, J. & Pericak-Vance,
M.A. Alzheimers Dement. 6, S74 (2010).
41. Di Fede, G. et al. Science 323, 1473–1477 (2009).
42. Tomiyama, T. et al. Ann. Neurol. 63, 377–387 (2008).
43. Clarimón, J. et al. Neurobiol. Aging 30, 1986–1991
(2009).
44. Nalls, M.A. et al. Neurogenetics 10, 183–190 (2009).
45. Nuytemans, K., Theuns, J., Cruts, M. & Van Broeckhoven,
C. Hum. Mutat. 31, 763–780 (2010).
46. Rovelet-Lecrux, A. et al. Nat. Genet. 38, 24–26
(2006).
47. Rovelet-Lecrux, A. et al. J. Alzheimers Dis. published
online, doi:10.3233/JAD-2010-100441 (15 July
2010).
48. Gijselinck, I. et al. Hum. Mutat. 29, 53–58 (2008).
49. Theuns, J. et al. Am. J. Hum. Genet. 78, 936–946
(2006).
50. Beecham, G.W. et al. Alzheimers Dement. 6, S113
(2010).

Biomarkers in Alzheimer’s disease drug
development
Kaj Blennow
Biomarkers may be of great value in
Alzheimer’s disease drug development to
select the most optimal drug candidates
for large and expensive phase 3 clinical
trials. Biomarkers will also be important
to provide evidence that a drug affects
the underlying pathophysiology of the
disease, which, together with a beneficial
effect on the clinical course, will be
essential for labeling the drug as having
a disease-modifying effect.
Twenty-five years ago, our knowledge of the
molecular pathogenesis of Alzheimer’s disease
was very limited. In the mid-1980s, researchers
succeeded in identifying amyloid-β (Aβ) and
tau as the main components of plaques and
tangles, which are the hallmarks of the dis-
ease. Together, these important achievements
marked the start of modern Alzheimer’s dis-
ease research, and subsequent research efforts
have led to a detailed knowledge of amyloid
precursor protein (APP) metabolism, Aβ gen-
eration and the pathogenic events involved in
plaque development and of tau homeostasis
and tangle formation1. The prevailing hypoth-
esis for Alzheimer’s disease pathogenesis is the
amyloid cascade hypothesis, which states that
Aβ aggregation with plaque formation is the
central event, ultimately leading to neuronal
degeneration and clinical symptoms2.
These research advances have led to a large
number of drug candidates that are now in
Kaj Blennow is at the Clinical Neurochemistry
Laboratory, Institute of Neuroscience and
Physiology, Department of Psychiatry and
Neurochemistry, The Sahlgrenska Academy at
University of Gothenburg, Mölndal, Sweden.
e-mail: kaj.blennow@neuro.gu.se
Published online 21 September 2010;
doi: 10.1038/nm.2221
nature medicine advance online publication 1
C O M M E N TA RY
clinical trials and may have disease-modifying
effects. The vast majority of these new treat-
ments aim to inhibit Aβ toxicity and include
secretase inhibitors inhibiting the production
of Aβ from APP, immunotherapies aimed at
increasing the clearance of Aβ from the brain
and inhibitors of Aβ aggregation1. If this type
of anti-Aβ treatment proves able to reduce
plaque pathology and have a beneficial effect
on cognition in people with Alzheimer’s dis-
ease, then the recent years of Alzheimer’s dis-
ease research will be a success story.
Unfortunately, reports of clinical trials
showing negative results have begun to accu-
mulate. In recent years, there have been sev-
eral clinical trials where drug candidates with
an Aβ-targeting mode of action have failed to
show any substantial effects on primary cog-
nitive outcome measures, such as tramipro-
sate (that binds soluble Aβ, thus maintaining
it in a nonfibrillar form) and tarenflurbil (a
γ-secretase modulator)3. These disappoint-
ing results are causing concern that the amy-
loid cascade hypothesis will be proved wrong
and that the potential success will turn into
a failure. However, there are several other
possible explanations for this. These include
that the trials have been done in subjects with
Alzheimer’s disease with dementia (that is, at
a far too advanced stage of the disease), that
clinical diagnostic procedures have been too
unspecific to allow identification of a homo­
genous group of patients with Alzheimer’s dis-
ease pathology and that the drug candidates
tested have been ineffective despite promising,
but misleading, data from preclinical develop-
ment.
Are we too far down the road in clinical
trials?
It is logical to assume that disease-modify-
ing drugs targeting Aβ will have only minor
effects on the clinical course of the disease in
Alzheimer’s disease cases with advanced neu-
rodegeneration. A potential example of this
is the follow-up study of cases in the Elan
AN1792 phase 1 Aβ immunotherapy trial. In
subjects who underwent post-mortem assess-
ment, there was evidence of marked plaque
removal, despite their clinical deterioration to
severe dementia before death4. Thus, arrest-
ing Aβ aggregation or even plaque removal
may have limited effect in Alzheimer’s disease
cases with dementia, owing to severe neuronal
and synaptic loss and heavy tangle load at this
stage of disease.
It is likely that we need to design trials on the
basis of diagnostic algorithms that recognize
the predementia stage (prodromal Alzheimer’s
disease) or even the asymptomatic phase of
the disease (preclinical Alzheimer’s disease)
to give promising drug candidates a chance
of showing a disease-modifying effect. This
proposal is supported by studies in transgenic
mouse models of Alzheimer’s disease, which
suggest that drugs would be most efficacious
early on in the disease process, before severe
plaque pathology is present5.
Mild cognitive impairment (MCI) may be
regarded as a transitional state between nor-
mal aging and Alzheimer’s disease. However,
MCI is an etiologically heterogeneous entity;
only ~50% of people with MCI have pro-
dromal Alzheimer’s disease, whereas others
have a benign form of MCI as part of the
normal aging process and some have other
disorders6. Given that individuals with MCI
have only mild disturbances in episodic
memory and other characteristic symptoms
of Alzheimer’s disease are absent or vague, it
is a great challenge for the clinician to identify
which MCI cases have prodromal Alzheimer’s
disease.
Testing Alzheimer’s disease drugs in clini-
cal trials in which unspecified MCI cases are
included will thus mean that around half of
commentary
the cases do not have prodromal Alzheimer’s
disease, the disorder for which the drug is
intended, which may seriously affect the pos-
sibility of identifying clinical effects. One pos-
sible example so far is the large trial on the
cholinesterase inhibitor donepezil, in which
more than 700 subjects with unspecified MCI
were treated for 3 years without any signifi-
cant effect of donepezil treatment7.
The addition of positive biomarkers as an
inclusion criterion in such MCI trials would
enrich the proportion of subjects with under-
lying Alzheimer’s disease pathology and,
thereby, increase the possibility of identifying
a positive effect of a drug. Several studies have
consistently shown that the combination of
high cerebrospinal fluid (CSF) levels of total
tau (T-tau) and phosphorylated tau (P-tau)
and low levels of the 42–amino-acid isoform
of Aβ (Aβ42) has a high predictive value for
identifying cases of prodromal Alzheimer’s
disease in people with MCI, with one study
reporting a sensitivity of 95%8. A high pre-
dictive value has been verified in large mul-
ticenter studies, including the Alzheimer’s
Disease Neuroimaging Initiative study9, the
Descripa study10 and the Swedish Brain Power
project11. Numerous studies have also shown
that magnetic resonance imaging (MRI)
measurement of hippocampal atrophy and
positron emission tomography (PET) imag-
ing with the 2-[18F]fluoro-2-deoxy-d-glucose
(FDG) ligand and β-amyloid–binding trac-
ers such as Pittsburgh Compound-B (PiB)
have high predictive values of approximately
85–90% for prodromal Alzheimer’s disease in
the MCI stage of the disease12,13. Thus, bio-
markers will be key tools for enrichment of
clinical trials with true prodromal Alzheimer’s
disease cases.
Is Alzheimer’s disease too
heterogeneous?
Trials involving clinically diagnosed sub-
jects with Alzheimer’s disease may enroll
patients with different pathologies, owing
to misdiagnoses. Apart from the fact that
the relative intensity of plaques and tangles
may differ markedly among individuals with
Alzheimer’s disease14, there is also a sizable
overlap in pathology between Alzheimer’s
disease and other dementias, such as Lewy
body dementia and vascular dementia15,16.
Current clinical criteria for Alzheimer’s dis-
ease may also be too nonspecific to identify
cases of pure Alzheimer’s disease with high
certainty. For example, recent data show that
approximately 20% of clinically diagnosed
Alzheimer’s disease cases enrolled in clini-
cal trials may have negative PiB-PET scans17.
This diagnostic uncertainty introduces a high
2 advance online publication nature medicine
Position as theragnostic
biomarker in trial type
Aβ
Aβ immuno- BACE1 γ-secretase aggregation
Biomarker Pathogenic process therapy inhibitor inhibitor inhibitor
Amyloid PET Brain Aβ load
CSF Aβ42
γ-secretase-dependent
APP and Aβ metabolism
CSF Aβ40
Primary biomarkers
CSF sAPPβ
β-secretase-dependent
CSF BACE APP and Aβ metabolism
activity
CSF Aβ1–14,
Aβ1–15, Aβ1–16 γ-secretase-independent
APP and Aβ metabolism
CSF sAPPα
CSF Aβ oligomers Aβ oligomerization
CSF total tau Intensity of neuronal
degeneration and
Downstream
biomarkers
MRI hippocampal brain atrophy rate
volume
Tau phosphorylation
CSF phospho-tau
and tangles
Brain glucose
FDG PET metabolism
Figure 1 Flow chart for hypothetical use of theragnostic biomarkers in Aβ-targeting clinical trials.
Theragnostic biomarkers can be divided into primary biomarkers (orange), used to identify and monitor
the specific biochemical effect, or mode of action, of the drug, and downstream biomarkers (blue),
used to identify and monitor effects on pathogenic processes downstream of the drug target. At right,
validated biomarkers that may be valuable to monitor specific pathogenic processes are indicated with
green boxes, whereas promising biomarkers that need further development and validation of methods
are indicated with yellow boxes. sAPP, soluble APP.
degree of noise into trials, which may mark- (the number or extent of Aβ plaques in the
edly reduce the possibility of identifying any brains of Alzheimer’s disease–transgenic
effect of a drug. mice) but have no beneficial effect in people
It is likely that the effectiveness of disease- with Alzheimer’s disease, indicating that these
modifying drugs targeting Aβ will vary mouse models have a very low predictive
between cases depending on the severity of power of treatment success in patients with
the Aβ plaque pathology. In other words, sporadic Alzheimer’s disease1. Alzheimer’s
clinically diagnosed Alzheimer’s disease cases disease–transgenic mice have a huge over-
with biomarker evidence of a disturbance in expression of Aβ and develop plaques much
Aβ metabolism—for example, low CSF Aβ42 faster (months instead of decades) than
or high PiB binding on PET—may be more people who develop Alzheimer’s disease and
responsive to anti-Aβ drugs than individuals are thus probably much more responsive to
not showing such a disturbance. Biomarkers anti-Aβ treatment than humans with sporadic
reflecting the central pathogenic processes Alzheimer’s disease. Indeed, several anti-Aβ
in Alzheimer’s disease may thus be valuable drug candidates that were shown to reduce
for post hoc data analyses of clinical trials to Aβ burden in Alzheimer’s disease–transgenic
stratify patient cohorts, enabling identifica- mice, such as tramiprosate (an Aβ antiaggre-
tion of subgroups with more homogeneous gation compound), tarenflurbil (a γ-secretase
neuropathology. modulator), rosiglitazone (a type 2 diabetes
drug that also acts as a β-secretase inhibitor),
Are we taking poor drug candidates into statins, nicotine, phenserine (a cholinest-
phase 3 clinical trials? erase inhibitor that also reduces Aβ amounts
The effect of disease-modifying, Aβ-targeting by decreasing APP mRNA translation) and
drug candidates on plaque pathology is com- estrogens, have now failed in clinical trials of
monly evaluated in APP and presenilin trans- patients with Alzheimer’s disease1,3. It is likely
genic mice. There are numerous drugs that that the promising data from the Alzheimer’s
cause a substantial reduction in ‘Aβ burden’ disease–transgenic mice models may have had

a large impact on decisions to proceed with
these drugs to phase 3 clinical trials.
One example of this might be tarenflurbil,
which is the R-enantiomer derivative of flur-
biprofen, a nonsteroidal anti-inflammatory
drug (NSAID). Data from experiments in cell
culture and in Alzheimer’s disease–transgenic
mice suggest that this drug selectively lowers
Aβ42 amounts and acts as a γ-secretase mod-
ulator18. The authors of this study cautiously
concluded that “rigorous clinical testing of
R-flurbiprofen will be necessary to determine
if it has the ability to lower Aβ42 in humans
and therapeutic efficacy in Alzheimer’s dis-
ease.”18. However, despite data showing
that the drug, even at high doses, does not
affect CSF Aβ42 concentrations or cause a
shift toward shorter Aβ isoforms in CSF in
humans19, and a phase 2 trial showing no
overall effect on primary outcome measures
in individuals with Alzheimer’s disease20,
the drug was taken to a large phase 3 clinical
trial. This 18-month trial on more than 1,600
subjects with Alzheimer’s disease showed no
significant effects on cognitive or functional
primary outcomes21.
Biomarkers used to detect and monitor
the biochemical effects of therapeutics may
be called ‘theragnostic’ biomarkers22. These
types of biomarkers may be valuable in drug
development to bridge the gap between animal
studies and large clinical trials. Biomarker evi-
dence from small-scale trials that a drug has a
true effect on the Alzheimer’s disease process
directly in humans would help in selecting
the most promising drug candidates, thereby
improving the success rate of large clinical
phase 2 and 3 trials. Such trials could be short-
term proof-of-principle studies on a limited
number of healthy volunteers in phase 1
(ref. 23), as well as proof-of-concept studies
on individuals with Alzheimer’s disease in
phase 2 (ref. 24). This type of early clinical
biomarker study would aid decision making
regarding whether to embark on large and
expensive phase 3 trials.
What constitutes a proficient
theragnostic biomarker?
A theragnostic biomarker should reflect
the central pathogenic processes of the dis-
ease. In contrast to drug development for
other neurodegenerative disorders, such as
Parkinson’s disease, Alzheimer’s disease drug
development has the benefit of several such
biomarkers readily available. CSF concentra-
tions of T-tau reflect cortical axonal degenera-
tion, whereas P-tau reflects tangle pathology
and Aβ42 reflects brain amyloid pathology22.
Further, MRI imaging of hippocampal atro-
phy gauges progression of neurodegenera-
nature medicine advance online publication 3
tion in a manner that correlates well with
both neuropathological measures of tangle
load and cognitive symptoms12. Last, PET
imaging with the FDG ligand allows for the
assessment of the glucose metabolism rate
in specific brain regions, whereas fibrillar Aβ
deposits and plaques in the brain can be visu-
alized with β-amyloid–binding tracers such as
PiB13. These biomarkers may serve as valuable
theragnostic tools in Alzheimer’s disease drug
development.
Treatment with symptomatic drugs (such
as cholinesterase inhibitors) is usually allowed
as background medication in Alzheimer’s dis-
ease clinical trials. Therefore, it is important
that theragnostic biomarkers used to monitor
the effect of an anti-Aβ drug are not affected
by this type of treatment. Several studies
have shown that the CSF biomarkers Aβ42,
T-tau and P-tau do not change during cho-
linesterase inhibitor treatment and also have
a very low intraindividual variability, both in
short-term and long-term trials25,26. These
results indicate that even minor changes in
biomarker amounts can be monitored with
CSF biomarkers. It is noteworthy that stud-
ies have shown that treatment with donepezil
may retard both the decline in brain glucose
metabolism evaluated by FDG-PET27,28 and
the progression rate of hippocampal atrophy
measured by MRI29,30. Thus, cholinesterase
treatment might be a potential confounder
when evaluating the effect of disease-modify-
ing therapies with these types of biomarkers.
Unresolved issues for Alzheimer’s
disease biomarkers
Both imaging and CSF biomarkers show
promise as diagnostic and theragnostic tools
in clinical trials, but they have in common a
need for standardization. Although biological
and within-laboratory variability for CSF bio-
markers is low, there is a variation in biomarker
levels between laboratories. Furthermore, there
are no available calibrators (certified reference
standards) for these biomarkers and no consen-
sus on what assays should be used. This com-
plicates multicenter trials and also precludes
the introduction of generally applicable cutoff
levels. In response to this problem, a global
quality-control program for CSF biomark-
ers, supported by the Alzheimer’s Association,
has recently been launched22. The aim of this
program is to serve as the basis for efforts to
minimize variation caused by differences in
preanalytical and laboratory procedures. A
substantial part of the variability is probably
also caused by assay-related factors, espe-
cially batch-to-batch variation of assays. This
involves a need for assay vendors to implement
improved standards for quality control in assay
commentary
production. Further information on the qual-
ity control program is available at http://www.
neurochem.gu.se/TheAlzAssQCprogram/.
For structural MRI imaging of medial tem-
poral lobe atrophy, there is no consensus on
which structure is the region of interest (hip-
pocampus, amygdala, entorhinal cortex or
parahippocampal gyrus), whether to use visual
rating or computerized algorithms for evalu-
ation of atrophy and whether age-dependent
cutoffs should be applied for hippocampal
atrophy to correct for age-related changes12.
The utility of structural MRI as a biomarker
in clinical trials will thus be increased by stan-
dardization of acquisition and analysis meth-
ods12. A task force to develop a harmonized
procedure for evaluation of hippocampal
atrophy is ongoing under the auspices of the
Alzheimer’s Association.
Similar standardization issues also apply
to PET imaging. For amyloid PET, there is
no consensus on the optimal brain region
to quantify amyloid retention and very lim-
ited data comparing the performance of the
various radioligands (PiB, FDDNP, AV-45 and
AZD2184)13. For FDG-PET, further data is
needed on which brain regions to evaluate in
each stage of the disease, that is, hypometabo-
lism in the parietotemporal cortex to diagnose
Alzheimer’s disease, in the medial temporal
lobe to predict MCI and in the posterior cin-
gulate cortex to predict progression from MCI
to Alzheimer’s disease13.
Implementing theragnostic biomarkers
in clinical trials
Conceptually, it may be useful to divide bio-
markers into primary (specific) and second-
ary (downstream) biomarkers (Fig. 1). A
primary biomarker can be used to identify
and monitor the specific biochemical effect,
or mode of action, of a drug. In trials with
Aβ-targeting compounds such as secretase
inhibitors or Aβ immunotherapy, examples
of primary biomarkers include CSF Aβ42 and
Aβ40, β-secretase (BACE1) activity and sol-
uble APP-β and amyloid-PET. Downstream
biomarkers are used to identify and monitor
effects on pathogenic processes downstream
of the drug target, for example, CSF T-tau
and MRI measurements of brain atrophy to
identify and monitor an effect on the rate of
neuronal degeneration in an anti-Aβ trial.
A key question will also be which of the
biomarkers to include in a trial. For example,
several studies have shown a strong correla-
tion between low CSF Aβ42 concentration
and high binding of the PiB ligand on PET,
suggesting that these biomarkers will give
similar information on the amount of Aβ
plaque pathology31,32. For a clinical trial, fac-
commentary
tors such as the willingness among clinicians
to perform lumbar puncture on the one hand,
and the availability of cyclotrons and PET
scanners and financial consideration on the
other, will form a basis for a decision on which
of these biomarkers to choose.
Can we find new biomarkers?
A surrogate biomarker can be used as a sub-
stitute for a clinical end point in a clinical
trial33. In the case of Alzheimer’s disease, such
a biomarker should correlate with cognitive
or functional clinical outcomes. Synapses
are the primary functional unit for neuronal
communication, and the degree of synaptic
degeneration has also been found to show the
best correlation with severity of dementia in
Alzheimer’s disease34. Synaptic proteins might
thus serve as valuable surrogate CSF biomark-
ers, as the abundance of these molecules is
likely to correlate with cognitive function and
disease progression. Although synaptic pro-
teins, such as synaptotagmin and rab3a, have
been shown to be present in human CSF35,
assay development for this class of proteins
has been difficult, owing to their low concen-
trations in CSF.
Probably for the same reason, assay devel-
opment has been difficult for Aβ oligomers
in CSF. Aβ oligomers would be a valuable
biomarker in Alzheimer’s disease drug
development for several reasons. First, recent
experimental data suggest that soluble Aβ oli-
gomers inhibit long-term potentiation and
have synaptotoxic properties, and thereby
have a central role in Alzheimer’s disease
pathogenesis36. Second, it is difficult to pre-
dict in which direction CSF Aβ42 may change
in an Aβ immunotherapy trial. Thus, CSF Aβ
oligomers might be an important biomarker
for Alzheimer’s disease. Some preliminary
studies on Aβ oligomers in CSF have been
published; a recent study with a previously
undescribed ELISA assay found a clear
increase in CSF Aβ oligomers in Alzheimer’s
disease37.
Targeted proteomics studies have shown
that, in addition to Aβ1–42 and Aβ1–40, there
are several shorter Aβ isoforms truncated at
the carboxy terminus. By a technique based
on immunoprecipitation with an Aβ-specific
monoclonal antibody combined with matrix-
assisted laser desorption–ionization time-of-
flight (MALDI-TOF) mass spectrometry, all
Aβ isoforms can be quantified simultane-
ously38. Cell culture studies have shown that
the longer isoforms (Aβ1–17 up to Aβ1–42)
are produced by the amyloidogenic pathway,
by the concerted action of β-secretase and
γ-secretase, whereas the short Aβ isoforms
(Aβ1–14, Aβ1–15 and Aβ1–16) are pro-
4 advance online publication nature medicine
duced by a previously unknown pathway of
APP processing38. A series of studies in dogs,
monkeys and Alzheimer’s disease–transgenic
mice have shown that γ-secretase inhibitor
treatment results in a marked increase in
Aβ1–14, Aβ1–15 and Aβ1–16 abundance,
probably as a result of increased substrate
availability of the C99 APP stub induced by
γ-secretase inhibition39–41. A recent clinical
trial on the γ-secretase inhibitor LY450139
in Alzheimer’s disease found no effect on the
CSF concentrations of Aβ1–40 or Aβ1–42, as
measured by ELISA42. In another study in the
same trial, a marked dose-dependent increase
in Aβ1–14, Aβ1–15 and Aβ1–16 was found by
the immunoprecipitation–MALDI-TOF tech-
nique24. Thus, these short Aβ isoforms may be
valuable and sensitive biomarkers to monitor
the biochemical effect of γ-secretase inhibitor
treatment in clinical trials.
Use of biomarkers to label a drug
‘disease modifying’
Ideally, a change in the slope of cognitive
decline may identify differences between
symptomatic and disease-modifying drugs.
Clinical trials designed with a delayed start
(placing patients on drug or placebo for a
period at the start of the trial and later ini-
tiating treatment also in the placebo group)
or staggered withdrawal (placing patients on
active drug at the trial start and later random-
izing them to drug or placebo) have been
suggested to allow this differentiation. This
approach may be theoretically attractive, but it
has not been successful, probably owing to the
very large variability in the rate of cognitive
change between cases, both in the dementia
and the MCI stages of Alzheimer’s disease.
Academic institutions, the pharmaceuti-
cal industry and regulatory organizations
agree that biomarkers have a crucial role in
the drug development process and that evi-
dence obtained from biomarker studies show-
ing that a drug candidate affects the central
disease processes in Alzheimer’s disease will,
together with a beneficial effect on cognition,
be essential for a drug to be labeled disease
modifying33.
How far have we come today?
Apart from animal studies, there is only pre-
liminary evidence that biomarkers may be
useful as theragnostic markers in humans.
For CSF biomarkers, a phase 2 study of the Aβ
clearance-enhancing compound PBT2 showed
a dose-dependent reduction in the amount
of the primary biomarker Aβ42 in CSF43.
Furthermore, in the AN1792 immunother-
apy trial, there was a decrease toward normal
amounts of the downstream biomarker T-tau
in CSF, suggesting that the treatment may have
reduced the rate of neuronal degeneration44.
For MRI measurements of atrophy, a decrease
in atrophy rate would be the logical expecta-
tion for a drug with a disease-modifying
effect. However, the AN1792 trial showed an
unexpected increase in the brain atrophy rate
in antibody responders, possibly as a result
of removal of Aβ plaques45. A recent report
on the bapineuzumab immunotherapy trial
showed a reduced cortical PiB retention during
treatment compared with both baseline and
placebo17. These results bring hope that amy-
loid PET will be valuable to assess the effect of
Aβ-targeting disease-modifying therapies on
cortical fibrillar Aβ load in clinical trials.
Thus, there are only limited data from
Alzheimer’s disease trials suggesting that bio-
markers have a place as theragnostic markers
in phase 1–2 studies or surrogate end points
in phase 3 studies. A challenge in this field of
research is the lack of drugs with an estab-
lished disease-modifying effect that can be
used to validate a biomarker and, at the same
time, that biomarker evidence that a drug
candidate affects the central disease processes
will be required to label a drug as disease
modifying33. This means that Alzheimer’s
disease drug development and biomarker
research need to go forward hand in hand.
The increasing number of clinical trials on
disease-modifying drug candidates that
include biomarkers for enrichment of pure
Alzheimer’s disease cases or as end points
will hopefully keep providing accumulative
evidence for the usefulness of biomarkers in
Alzheimer’s disease clinical trials and serve
as the basis for approval by the regulatory
authorities of disease-modifying drugs for
this devastating disease.
COMPETING FINANCIAL INTERESTS
The author declares competing financial interests:
details accompany the full-text HTML version of the
paper at http://www.nature.com/naturemedicine/.
1. Blennow, K., de Leon, M.J. & Zetterberg, H. Lancet
368, 387–403 (2006).
2. Hardy, J. & Selkoe, D.J. Science 297, 353–356
(2002).
3. Mangialasche, F. et al. Lancet Neurol. 9, 702–716
(2010).
4. Holmes, C. et al. Lancet 372, 216–223 (2008).
5. Garcia-Alloza, M. et al. Mol. Neurodegener. 4, 19
(2009).
6. DeCarli, C. Lancet Neurol. 2, 15–21 (2003).
7. Petersen, R.C. et al. N. Engl. J. Med. 352, 2379–2388
(2005).
8. Hansson, O. et al. Lancet Neurol. 5, 228–234
(2006).
9. Shaw, L.M. et al. Ann. Neurol. 65, 403–413 (2009).
10. Visser, P.J. et al. Lancet Neurol. 8, 619–627 (2009).
11. Mattsson, N. et al. J. Am. Med. Assoc. 302, 385–393
(2009).
12. Frisoni, G.B., Fox, N.C., Jack, C.R. Jr., Scheltens, P. &
Thompson, P.M. Nat. Rev. Neurol. 6, 67–77 (2010).
13. Nordberg, A., Rinne, J.O., Kadir, A. & Langstrom, B.
Nat. Rev. Neurol. 6, 78–87 (2010).

14. Nelson, P.T., Kukull, W.A. & Frosch, M.P. J. Neuropathol.
Exp. Neurol. 69, 449–454 (2010).
15. Kotzbauer, P.T., Trojanowsk, J.Q. & Lee, V.M. J. Mol.
Neurosci. 17, 225–232 (2001).
16. Schneider, J.A., Arvanitakis, Z., Leurgans, S.E. &
Bennett, D.A. Ann. Neurol. 66, 200–208 (2009).
17. Rinne, J.O. et al. Lancet Neurol. 9, 363–372 (2010).
18. Eriksen, J.L. et al. J. Clin. Invest. 112, 440–449
(2003).
19. Galasko, D.R. et al. Alzheimer Dis. Assoc. Disord. 21,
292–299 (2007).
20. Wilcock, G.K. et al. Lancet Neurol. 7, 483–493
(2008).
21. Green, R.C. et al. J. Am. Med. Assoc. 302, 2557–2564
(2009).
22. Blennow, K., Hampel, H., Weiner, M. & Zetterberg, H.
Nat. Rev. Neurol. 6, 131–144 (2010).
23. Bateman, R.J. et al. Ann. Neurol. 66, 48–54 (2009).
24. Portelius, E. et al. Alzheimers. Res. Ther. 2, 7
(2010).
nature medicine advance online publication 5
25. Blennow, K. et al. Neurosci. Lett. 419, 18–22
(2007).
26. Zetterberg, H. et al. J. Alzheimers Dis. 12, 255–260
(2007).
27. Chen, X., Magnotta, V.A., Duff, K., Boles Ponto, L.L.
& Schultz, S.K. J. Neuropsychiatry Clin. Neurosci. 18,
178–185 (2006).
28. Tune, L. et al. Am. J. Geriatr. Psychiatry 11, 169–177
(2003).
29. Krishnan, K.R. et al. Am. J. Psychiatry 160, 2003–
2011 (2003).
30. Hashimoto, M. et al. Am. J. Psychiatry 162, 676–682
(2005).
31. Fagan, A.M. et al. Ann. Neurol. 59, 512–519 (2006).
32. Degerman Gunnarsson, M. et al. Dement. Geriatr. Cogn.
Disord. 29, 204–212 (2010).
33. Hampel, H. et al. Nat. Rev. Drug Discov. 9, 560–574
(2010).
34. Davidsson, P. & Blennow, K. Int. Pychogeriatr. 10,
11–23 (1998).
commentary
35. Davidsson, P., Puchades, M. & Blennow, K.
Electrophoresis 20, 431–437 (1999).
36. Walsh, D.M. & Selkoe, D.J. J. Neurochem. 101, 1172–
1184 (2007).
37. Fukumoto, H. et al. FASEB J. published online,
doi:10.1096/fj.09-150359 (25 March 2010).
38. Portelius, E. et al. Neurobiol. Aging published online,
doi:10.1016/j.neurobiolaging.2009.06.002 (14 July
2009).
39. Portelius, E. et al. Neurodegener. Dis. 6, 258–262
(2009).
40. Cook, J.J. et al. J. Neurosci. 30, 6743–6750 (2010).
41. Portelius, E. et al. J. Alzheimers Dis. published online,
doi:10.3233/JAlzheimer’s disease-2010-100573 (15
July 2010).
42. Fleisher, A.S. et al. Arch. Neurol. 65, 1031–1038
(2008).
43. Lannfelt, L. et al. Lancet Neurol. 7, 779–786 (2008).
44. Gilman, S. et al. Neurology 64, 1553–1562 (2005).
45. Fox, N.C. et al. Neurology 64, 1563–1572 (2005).

Clinical trials of disease-modifying
therapies for neurodegenerative diseases:
the challenges and the future
Anthony E Lang
Neurodegenerative diseases such as
Parkinson’s disease and Alzheimer’s
disease represent a crucial and
exponentially increasing challenge to
health care systems throughout the
world. There is an urgent need for
effective treatments that will both delay
their onset and slow their inexorable
progression. Many obstacles stand
in the way of realizing these goals.
It is expected that future advances
will have a major impact on how and
when the diagnosis will be made. It is
hoped that these will eventually make
it possible to initiate effective disease-
modifying therapies long before the
neurodegenerative process becomes
established and symptomatic.
Neurodegenerative diseases of the elderly
currently represent a major challenge to the
health care system. With the increasing lon-
gevity of the general population, the predicted
figures for the prevalence of these disorders
and their global financial impact are truly
staggering. For example, it has been estimated
that by 2030 as many as 7.7 million people in
the US will have Alzheimer’s disease, and by
2050 this number will reach approximately
13.5 million, with total annual costs for care
rising from $172 billion in 2010 to $1.08 tril-
lion in 2050 (ref. 1). This analysis does not
take into account the value of unpaid care
Anthony E. Lang is in the Department of Medicine
(Neurology) at the University of Toronto, Toronto,
Canada.
e-mail: lang@uhnresearch.ca
Published online 21 September 2010;
doi: 10.1038/nm.2220
nature medicine advance online publication 1
C O M M E N TA RY
provided by families and others, estimated
to have been $144 billion in 2009 (ref. 2).
In the UK, the cost of dementia is now esti-
mated to exceed the combined cost of cancer,
heart disease and stroke3. The prevalence of
Parkinson’s disease is expected to double by
the year 2030 (ref. 4), and, given the greater
risk of Parkinson’s disease dementia in
elderly individuals with Parkinson’s disease,
it is expected that the health care costs for
this disorder will increase even more rela-
tive to the prevalence of the disease. These
concerns highlight the urgent need for the
development of effective disease-modifying
therapies. Effective treatment would have a
major influence on the economic and social
burden of these age-related disorders. For
example, in the case of Alzheimer’s disease,
it has been estimated that a delay in onset by
5 years would translate into a 50% decrease
in disease prevalence, and a delay of 10 years
would result in a virtual disappearance of the
disease5,6. Unfortunately, despite considerable
investment, to date all attempts at developing
such treatments have failed.
Terminology
The ‘Holy Grail’ of neurodegenerative dis-
eases is the development of effective ‘neuro-
protective’ therapy. By definition, this implies
that the treatment has a direct impact on the
biology of the disease, as confirmed by a
reliable measure of this effect. Alternatively,
‘disease modification’ implies a modification
of the clinical course of the disease without
implicating mechanism. Other interventions
could be classified as ‘neurorestorative’ (for
example, cell-based and some gene thera-
pies) and ‘neuroregenerative’ (for example,
trophic factors). Although these experimen-
tal approaches hold promise for changing the
course of neurodegenerative diseases, I will
largely restrict my comments to small mol-
ecule treatments directed at the underlying
neurodegenerative processes, but many of the
important issues apply broadly to all of these
types of therapies.
Although the term ‘neuroprotection’ has
been used widely with respect to early clinical
trials in Parkinson’s disease and Alzheimer’s
disease7,8, a positive result in such a clinical
trial does not mean that the drug is acting on
the underlying biological mechanisms that
cause the disease. Although the future ideal
may be to confirm true neuroprotection, in
the immediate future the best we can hope
for is disease modification, and, if we succeed
in truly altering the inexorably progressive
clinical course of these diseases, it shouldn’t
matter how this is accomplished. Presentation
with signs and symptoms of these disorders
probably represents a combination of the
disease reaching a critical threshold of neu-
ronal dysfunction along with the failure of a
number of central nervous system compen-
satory mechanisms that have been activated
in response to the underlying neurodegen-
eration. For example, it is well known that
individuals with Parkinson’s disease have pro-
found nigrostriatal dopamine cell loss before
developing clinical motor features9 (this is
particularly extreme in people with auto-
somal recessive forms of Parkinsonism, such
as those with Parkin mutations10). Their final
presentation with motor symptoms probably
represents a failure of a variety of compensa-
tory mechanisms. If it were possible to even
partially restore some of these mechanisms
before the individual reaches an irretriev-
able degree of degeneration, the clinical
course of the disease might be considerably
altered without even changing the underly-
commentary
ing neurodegenerative process. Being able
to identify people at the appropriate point in
their disease pathogenesis would, therefore,
be extremely important.
Problems in planning, designing and
conducting clinical trials
One of the obvious challenges to changing the
course of neurodegenerative diseases relates
to our incomplete or inadequate understand-
ing of the underlying disease pathogeneses.
Age remains the single most important risk
factor for Alzheimer’s disease and Parkinson’s
disease11, yet its influence on the underlying
biological mechanisms is not understood.
Particularly in Parkinson’s disease, age also
has a strong influence on clinical manifesta-
tions (that is, there is greater cognitive and
axial motor disturbances in those with later
age of onset)12,13, and this will have a major
impact on many aspects of clinical trial
design, including subject selection and treat-
ment outcomes.
To date, animal models for both Alzheimer’s
disease and Parkinson’s disease have failed to
reflect many of the key aspects of the human
diseases14–19. Such models have been espe-
cially ineffective in predicting the response
of affected individuals to putative disease-
modifying therapies20. If more reliable, more
representative animal models can be devel-
oped, success in establishing effective disease-
modifying therapies for humans will require
robust and easily replicated effects, preferably
in several animal species21.
Therapeutic approaches that are directed at
single biological mechanisms or targets may
be inadequate, given the complexity of these
multifaceted neurodegenerative disorders.
Successful treatment may require a cocktail
approach, such as that applied in cancer. These
treatments may vary depending on the stage of
the disorder being treated. For example, very
early treatment in at-risk individuals might be
exclusively directed at selected primary disease
mechanisms, whereas later treatment may need
to address mechanisms that are further down-
stream and additional secondary mechanisms,
including neuroinflammation22. Furthermore,
we have absolutely no understanding of the
basis of the selective vulnerability of various
neuronal populations in neurodegenerative
diseases. A better understanding of how and
why certain regions of the brain are more
affected than others (for example, through
sharing of common biological functions or
being connected in common neuronal net-
works) may assist in the choice of agents to
include in future neuroprotective cocktails.
Additionally, if therapeutic efficacy requires
modulating neural networks rather than indi-
2 advance online publication nature medicine
Hypothetical course
of neuronal loss
Progressive neuronal degeneration
3
2
1
Time Premotor period
Figure 1 Theoretical effects of disease modifying treatments in Parkinson’s disease. Effects of
initiation of two disease-modifying treatments at different stages of Parkinson’s disease on the clinical
features and course of the illness. 1, after onset of classical clinical features of the disease; 2, after the
onset of premotor features (before classical motor); 3, before the onset of any clinical features in at-risk
individuals or with the use of a sensitive disease trait biomarker.
vidual molecular targets, this may require
innovative drug discovery strategies that fall
outside the current single-target approach
that is the paradigm within the pharmaceuti-
cal industry. Clearly, advances in this regard
will require preclinical animal models that are
more translatable to the human disease than
the current ones and that can be well character-
ized by multimodal micro–positron emission
tomography (PET) and functional magnetic
resonance imaging procedures.
Neurodegenerative diseases such as
Alzheimer’s disease and Parkinson’s disease
are quite heterogeneous at multiple levels,
including their genetic, clinical and neuro-
pathological bases, so disease pathogenesis is
probably equally heterogeneous. This hetero-
geneity may therefore translate into a variety
of responses to putative disease-modifying
drugs in each patient subgroup. Despite wide-
spread recognition of this important feature of
neurodegenerative diseases, to date, treatment
trials have largely lumped all patients together
in the expectation of a uniform therapeu-
tic response. However, a number of ongoing
large-scale studies designed to evaluate clini-
cal, imaging and other biological markers and
their natural histories promise to have a major
impact on the understanding of this hetero-
geneity and therefore on the design of future
trials of disease-modifying therapies. These
include the Alzheimer’s Disease Neuroimaging
Initiative in the US and similar projects in
other countries23–25 and the Michael J. Fox
Foundation–supported Parkinson Progression
Marker Initiative (PPMI), as well as the pro-
posed Longitudinal and Biomarker Studies in
Mildly effective drug
Moderately effective drug
Threshold for
premotor features
Threshold for
classical motor features
Overt clinical symptoms
Parkinson Disease (LABS-PD), directed by the
Parkinson Study Group26.
A preeminent goal of these and other studies
is the development of various types of biomark-
ers that will be essential components in realiz-
ing effective disease-modifying therapies27–29.
Markers that accurately track the progression
of the underlying disease (progression marker)
and that can serve as surrogates of the desired
therapeutic effect (response marker) will be
crucial for use in phase 2 clinical trials allow-
ing prioritization of compounds for subsequent
study in phase 3 trials. One major stumbling
block to the development of disease-modifying
therapy is the inability to demonstrate and mea-
sure the central effects of the study drug, which
also compromises the selection of drug dos-
age in clinical trials. Pharmacodynamic mark-
ers would allow the direct measurement of the
biological effects on the proposed drug targets,
thus supporting drug development decisions.
‘Theragnostic markers’ is another term used to
describe biomarkers designed to identify and
monitor the biochemical effect of therapeutic
interventions30. Trials employing such a bio-
marker could involve a relatively small num-
ber of patients for short treatment periods, thus
allowing faster decisions of whether to proceed
with large and expensive phase 2 or 3 clinical tri-
als. A diagnostic marker would improve accu-
racy in diagnosing patients that enter phase 3
trials, increasing the power to detect disease
modification and reducing costs. A practical
example of the need for this type of biomarker
is already well recognized from experience in
trials in early Parkinson’s disease, in which a
proportion of enrolled subjects did not have

the disease of interest—presynaptic dopamine
imaging has yielded scans without evidence of
dopaminergic deficit (SWEDD) in up to 14%
of study participants31. Diagnostic markers
may also be crucial in dividing patients into
subgroups according to who will be more or
less responsive to therapies directed at specific
disease-related targets. This would reduce the
heterogeneity of enrolled subjects and increase
the likelihood of demonstrating specific target–
related impact in the trial. Pharmacogenomic
markers that predict serious adverse medica-
tion effects would also assist greatly in early
drug development.
Finally, biomarkers allowing a definitive
diagnosis at the earliest stages of disease rep-
resent one of the highest priorities in the search
for effective disease-modifying therapies. The
goal is to be able to diagnose patients before the
establishment of widespread neuronal damage,
at a time when they would presumably be more
likely to respond to the effects of the therapeu-
tic intervention. Depending on the safety and
cost of the treatment, an alternative would
be to identify individuals with a high risk of
developing neurodegeneration even before
it has begun, such as individuals inheriting
various disease-associated genetic mutations
or genetic risk factors. A widespread concern
related to the large trials conducted to date has
been that individuals with well-established
diagnoses, even in the earliest clinical stages of
the disease, may already suffer from a neuro-
degenerative process that is too far advanced to
benefit from treatment, particularly from the
more focused treatments directed at a single
pathogenetic mechanism. An analogy might
be that of ischemic heart disease, where early
treatment of various risk factors such as hyper-
tension, hypercholesterolemia, coronary artery
narrowing and so on may be quite effective but,
once congestive heart failure is established,
treating these factors may be too late to affect
the outcome.
As our understanding of neurodegenerative
diseases progresses, it is clear that there will
be a need to diagnose these disorders at the
very earliest stages. Our definitions of these
diseases will necessarily change from requir-
ing early cognitive or motor disturbances in
Alzheimer’s and Parkinson’s, respectively, to
considering even earlier clinical, and preferably
biological, disturbances. This would allow the
initiation of treatments that might completely
prevent the development of the clinical fea-
tures currently required for diagnosis (Fig. 1).
Proposals already exist for the development of
new diagnostic criteria that include biomark-
ers for the early diagnosis of Alzheimer’s dis-
ease32, and additional diagnostic categories of
mild cognitive impairment of the Alzheimer’s
nature medicine advance online publication 3
type, prodromal Alzheimer’s dementia and
preclinical Alzheimer’s disease are under active
consideration. Likewise, studies evaluating var-
ious ‘premotor’ features of Parkinson’s disease
(for example, olfactory dysfunction, rapid eye
movement behavior disorder and changes in
heart rate variability) are being combined with
central nervous system imaging (for example,
dopamine transporter single-photon emis-
sion computed tomography and transcranial
ultrasound) in an attempt to define the pres-
ence of the disease before the development of
the typical motor features33–35. In planning
disease-modifying trials in at-risk individuals
or in those with very early preclinical disease
(this term may not be truly accurate, as many
affected individuals will have clinical features
that simply don’t constitute the diagnosis as
currently conceived), establishing dependable
primary endpoints will be problematic in the
absence of a reliable biomarker of disease state.
One common consideration is the conversion
to manifest disease, but establishing an early
and accurate clinical diagnosis of manifest
Parkinson’s disease or Alzheimer’s disease may
be methodologically challenging36.
Study design issues
To date, neuroprotective trials have largely
evaluated people in the early clinical stages of
the disease (generally untreated), using clini-
cal endpoints that involve either the change in
a classical clinical measure of the disease over
time or progression to the point of reaching a
disease milestone (for example, need for dop-
aminergic therapy in Parkinson’s disease tri-
als37–39). The greatest concern in these studies
has been the potential for the study interven-
tion to cause symptomatic benefit that pre-
cludes the determination of disease-modifying
effects. Studies using a washout design, where
the drug is removed after a certain period of
treatment, have failed to adequately overcome
this problem, as the symptomatic treatment
effect may exceed the duration of the washout.
The delayed-start design40,41 was developed
in hopes of overcoming this confounding
effect. In such a study, an initial placebo con-
trol phase is followed by a second phase in
which all patients receive the active treatment.
However, such an approach is not without its
own potential problems42,43. Although simula-
tion studies have shown a clear advantage of
washout versus delayed-start designs44, the
prolonged washouts necessary may not be
practical or possible in the case of treatments
that provide substantial and needed symptom-
atic benefit. The rate of decline of a neuroim-
aging surrogate has been used in Parkinson’s
disease in the hopes of avoiding the confound-
ing effects of clinical symptomatic benefit of
commentary
treatment31,45,46. However, although imaging
of the status of the presynaptic nigrostriatal
dopamine system can readily distinguish indi-
viduals with early Parkinson’s disease from con-
trols, and abnormalities can be observed even
before motor symptoms and signs are appar-
ent, these studies have been disappointing in
monitoring both disease-modifying34 and
neurorestorative47,48 therapies. This may be
due, in part, to the potential pharmacological
modulation or regulation of presynaptic pro-
teins that may not relate to the actual disease
status. Fluorodeoxyglucose (FDG)-PET with
multivariate principal components analysis
shows promise in providing an alternative
method of monitoring disease status. However,
experience with this technique in Parkinson’s
disease is largely limited to one center49,50.
In the absence of reliable biomarkers of dis-
ease state that are not regulated by the drug that
is being tested, any assessment of disease modi-
fication becomes extremely difficult once effec-
tive symptomatic treatment is initiated. The
best example of this is the use of dopaminergic
medication for Parkinson’s disease; once this
has been started, clinical assessments become
extremely limited in their ability to reflect
disease progression. Therefore, long-term
studies in patients who are already on these
dopaminergic treatments are now emphasizing
features that tend to be resistant to dopamin-
ergic medication (so-called nondopaminergic
features, believed to be due to the extension of
the disease beyond the nigrostriatal system51).
These include ambulatory capacity (freezing,
falling, gait abnormalities and postural sta-
bility), cognitive dysfunction, disability and
quality of life. New statistical approaches to
assess these outcomes are also required. For
example, in the Neuroprotection Exploratory
Trials in Parkinson’s Disease (NET-PD) Long-
term Study 1 comparing creatine to placebo,
five measures of decline will be combined into
a single outcome and compared by a nonpara-
metric Global Statistical Test7. This novel sta-
tistical approach is especially well suited to a
multidimensional disorder such as Parkinson’s
disease where patient disability and functional
impairment are due to a combination of
disease- and treatment-related disturbances in
diverse brain systems52.
The lack of biomarkers also hampers the
choice of which promising disease-modifying
treatments to pursue. One approach taken by
the NET-PD program has been to evaluate
a variety of agents through a futility trial
design, as originally applied in oncology53.
This approach attempts to quickly screen
out agents associated with a disease course
not sufficiently different from historical
controls. Those compounds not rejected as
commentary
futile can then proceed to further study in a
long-term ‘large simple’ trial design. In this
type of study, individuals with midstage dis-
ease are randomly assigned to receive active
study drug or placebo and are followed for
long periods (5–10 years) during which they
are treated with routine symptomatic therapy
according to the judgment of the investiga-
tor. An interesting sideline challenge to this
approach in Parkinson’s disease has been the
observation that historical controls may not
adequately establish futility threshold, owing
to possible changes over the years in practice
parameters, in the use of the clinical rating
scale (the Unified Parkinson’s Disease Rating
Scale) or in poorly understood changes in the
patient population of interest53. Alternative
study designs might also be considered in the
early evaluation of possible disease-modify-
ing treatments. For example, where rates of
conversion or changes in outcomes are not
well known in advance, adaptive designs
assist in drug development by introducing
flexibility within trial design, allowing pre-
specified modifications in a seamless fashion
on the basis of ongoing accruing data (http://
biopharmnet.com/doc/doc12004.html)54.
Currently, it must be emphasized that
Alzheimer’s disease and Parkinson’s dis-
ease differ greatly in their prospects and in
approaches to planning disease-modifying
therapy. Alzheimer’s disease is considerably
further along than Parkinson’s disease with
respect to nonclinical criteria that might be
used in the evaluation of new treatments.
These include cerebrospinal fluid markers of
different components of the disease (such as
Aβ42 or tau), imaging of the underlying bio-
logical changes (amyloid imaging) and pro-
gressive metabolic and anatomic correlates
of the status of the underlying neurodegen-
eration (FDG-PET and volumetric magnetic
resonance imaging, respectively)29. The addi-
tion of a genetic risk factor (apolipoprotein E
ε4 status) can also be included; indeed, it has
already been found that the effectiveness and
adverse event profile of one form of passive
Aβ immunotherapy differs between APOE
ε4 carriers and noncarriers54. Hopefully,
planned efforts to evaluate Parkinson’s disease
4 advance online publication nature medicine
(PPMI, LABS-PD) similar to those applied to
Alzheimer’s disease over the past 5 years will
set the stage for more successful study of puta-
tive disease-modifying treatments.
As advances are made on the fronts high-
lighted above, it is expected that major changes
will be possible in the design of studies evalu-
ating potential disease-modifying therapies.
Such crucial components of clinical trials as
the primary inclusion criteria, the number of
subjects required to determine efficacy and the
primary outcome variables will be very differ-
ent from those used currently. For approval
of a treatment with an indication of disease
modification, it is likely that, initially, combi-
nations of clinical and biomarker outcomes
will be required. However, as more accurate
and reliable biomarkers become available, there
will hopefully come a time that outcomes will
rely exclusively on nonclinical evaluations, and
successful treatment will completely prevent the
onset of the clinical disorders we now recognize
as Alzheimer’s disease and Parkinson’s disease.
ACKNOWLEDGMENTS
I thank B. Greenberg for his helpful comments.
COMPETING FINANCIAL INTERESTS
The author declares no competing financial interests.
1. The Lewin Group. Saving lives, saving money: dividends
for Americans investing in Alzheimer’s Research. (Falls
Church, Virginia, USA, 2004).
2. Alzheimer’s Association. <http://www.alz.org/
documents_custom/trajectory.pdf> (2010).
3. Alzheimer’s Disease International. <http://www.alz.
co.uk/research/files/WorldAlzheimerReport.pdf>
(2009).
4. Dorsey, E.R. et al. Neurology 68, 384–386 (2007).
5. DeKosky, S.T. & Marek, K. Science 302, 830–834
(2003).
6. Brookmeyer, R., Gray, S. & Kawas, C. Am. J. Public
Health 88, 1337–1342 (1998).
7. Olanow, C.W., Kieburtz, K. & Schapira, A.H. Ann.
Neurol. 64 Suppl 2, S101–S110 (2008).
8. Francis, P.T., Norberg, A. & Arnold, S.E. Trends
Pharmacol. Sci. 26, 104–111 (2005).
9. Brooks, D.J. Ann. Neurol. 44, S10–S18 (1998).
10. Broussolle, E. et al. Neurology 55, 877–879 (2000).
11. Levy, G. Arch. Neurol. 64, 1242–1246 (2007).
12. Levy, G. et al. Arch. Neurol. 62, 467–472 (2005).
13. Aarsland, D. et al. J. Neurol. 254, 38–45 (2007).
14. Jenner, P. Ann. Neurol. 64, S16–S29 (2008).
15. Dawson, T.M., Ko, H.S. & Dawson, V.L. Neuron 66,
646–661 (2010).
16. Ashe, K.H. & Zahs, K.R. Neuron 66, 631–645
(2010).
17. Epis, R. et al. Eur. J. Pharmacol. 626, 57–63 (2010).
18. Kokjohn, T.A. & Roher, A.E. Alzheimers Dement. 5,
340–347 (2009).
19. Philipson, O. et al. FEBS J. 277, 1389–1409
(2010).
20. Lang, A.E. Lancet Neurol. 5, 990–991 (2006).
21. Jucker, M. Nat. Med. published online, doi:10.1038/
nm.2224 (21 September 2010)
22. Hirsch, E.C. & Hunot, S. Lancet Neurol. 8, 382–397
(2009).
23. Aisen, P.S. et al. Alzheimers Dement. 6, 239–246
(2010).
24. Trojanowski, J.Q. et al. Alzheimers Dement. 6, 230–
238 (2010).
25. Weiner, M.W. et al. Alzheimers Dement. 6, 202–211
(2010).
26. Ravina, B. et al. Mov. Disord. 24, 2081–2090
(2009).
27. Scherzer, C.R. Neurobiol. Dis. 35, 148–156 (2009).
28. Hampel, H. et al. Nat. Rev. Drug Discov. 9, 560–574
(2010).
29. Blennow, K. Nat. Med. published online, doi:10.1038/
nm.2221 (21 September 2010).
30. Blennow, K., Hampel, H., Weiner, M. & Zetterberg, H.
Nat. Rev. Neurol. 6, 131–144 (2010).
31. Fahn, S. et al. N. Engl. J. Med. 351, 2498–2508
(2004).
32. Dubois, B. et al. Lancet Neurol. 6, 734–746 (2007).
33. Siderowf, A. & Stern, M.B. Ann. Neurol. 64,
S139–S147 (2008).
34. Ponsen, M.M., Stoffers, D., Twisk, J.W.R., Wolters,
E.C. & Berendse, H.W. Mov. Disord. 24, 1060–1065
(2009).
35. Haehner, A. et al. Mov. Disord. 22, 839–842
(2007).
36. Kieburtz, K. Neurology 66, S50–S57 (2006).
37. Parkinson Study Group. N. Engl. J. Med. 321, 1364–
1371 (1989).
38. Parkinson Study Group PRECEPT Investigators et al.
Neurology 69, 1480–1490 (2007)
39. Olanow, C.W. et al. Lancet Neurol. 5, 1013–1020
(2006).
40. Leber, P. Alzheimer Dis. Assoc. Disord. 11 Suppl 5,
S10–S21 (1997).
41. Olanow, C.W. et al. Mov. Disord. 23, 2194–2201
(2008).
42. Clarke, C.E. Mov. Disord. 23, 784–789 (2008).
43. D’Agostino, R.B. Sr. N. Engl. J. Med. 361, 1304–
1306 (2009).
44. Ploeger, B.A. & Holford, N.H. Pharm. Stat. 8, 225–
238 (2009).
45. Parkinson Study Group et al. J. Am. Med. Assoc. 287,
1653–1661 (2002).
46. Whone, A.L. et al. Ann. Neurol. 54, 93–101 (2003).
47. Freed, C.R. et al. N. Engl. J. Med. 344, 710–719
(2001).
48. Olanow, C.W. et al. Ann. Neurol. 54, 403–414
(2003).
49. Tang, C.C., Poston, K.L., Dhawan, V. & Eidelberg, D.
J. Neurosci. 30, 1049–1056 (2010).
50. Eckert, T., Tong, C. & Eidelberg, D. Lancet Neurol. 6,
926–932 (2007).
51. Lim, S.Y., Fox, S.H. & Lang, A.E. Arch. Neurol. 66,
167–172 (2009).
52. Huang, P. et al. Mov. Disord 24, 1732–1739
(2009).
53. NINDS NET-PD Investigators et al. Neurology 68,
20–28 (2007).
54. Salloway, S. et al. Neurology 73, 2061–2070
(2009).

Bridging the Valley of Death of therapeutics
for neurodegeneration
Steven Finkbeiner
Neurodegenerative diseases are the
sixth leading cause of death in the
US. The market for disease-modifying
drugs is enormous, but no drug exists.
Academic scientists are increasingly
pursuing the discovery and development
of therapeutics. Their progress could
potentially reduce the risk of failure
sufficiently to warrant greater industry
investment and movement of leads
into clinical trials. Here we consider
the many obstacles to the development
of therapeutics for neurodegenerative
disease within academia, with a special
focus on organizational issues.
The imperative
It is difficult to discuss the ramifications of
neurodegenerative diseases for society with-
out sounding alarmist. In 2010, Alzheimer’s
disease, the most common cause of dementia,
afflicted 5.3 million people and cost $172 bil-
lion in the United States1. Aging is a risk factor
for Alzheimer’s disease, and the US population
over 65 will double by 2030 and quadruple
by 2050 (ref. 2). Consequently, the number
of Americans with Alzheimer’s disease will
rise to 11–15 million by 2050, unless there is a
therapeutic breakthrough1,3. Europe and parts
of Asia face even more daunting challenges
because, on average, their populations are
older than those in the United States. By 2040,
Steven Finkbeiner is at the Gladstone Institute
of Neurological Disease, Taube-Koret Center
for Huntington’s Disease Research, Consortium
for Fronto-temporal Dementia Research, and
Departments of Neurology and Physiology, University
of California, San Francisco, California, USA.
e-mail: sfinkbeiner@gladstone.ucsf.edu
Published online 21 September 2010;
doi: 10.1038/nm.2222
nature medicine advance online publication 1
C O M M E N TA RY
China is predicted to have more citizens with
dementia than the rest of the developed world
combined4. Parkinson’s disease affects 1.0–1.5
million people in the United States and is also
expected to increase two- to threefold by 2050.
Against this stark backdrop is an even starker
reality: not a single disease-modifying thera-
peutic exists for the major neurodegenerative
diseases, including Alzheimer’s, Parkinson’s and
Huntington’s diseases and amyotrophic lateral
sclerosis (ALS). The need for a drug is urgent.
The market for a drug for Alzheimer’s and
Parkinson’s diseases is enormous and grow-
ing rapidly. Even a drug for orphan diseases,
such as Huntington’s disease or ALS, would be
highly profitable. Why isn’t there one? And why
is investment in drug development in this area
so small5?
The limitations of preclinical models of
neurodegenerative diseases, the importance
of biomarkers and the difficulties of clinical
trials are authoritatively reviewed in this issue.
Hopefully, the proposed strategies will lead to
breakthroughs. Here I will focus on the orga-
nizational obstacles to developing neurothera-
peutics and steps that can be taken today with
existing tools to reduce the risks (and costs)
of drug development for neurodegenerative
diseases.
Risky business
The risk of drug development is higher for
neurodegeneration than for other diseases. One
reason is that the research and development
(R&D) tools are mostly unproven. Without
clinically proven therapies, the simpler disease
models for drug discovery cannot be validated,
and the predictive value of any research data
is unclear. Defining mechanisms of disease is
usually a crucial step in identifying drug tar-
gets. Drug developers therefore must make big
bets on the validity of putative drug targets and
perform clinical trials before knowing whether
their faith was well placed.
In addition, clinical trials for neurodegen-
erative diseases are extraordinarily expensive.
The Food and Drug Administration (FDA)
insists that drug trial designs rely heavily on
clinical measures of efficacy. Unfortunately,
clinical measures for neurodegenerative dis-
eases can be relatively insensitive and variable.
Worse still, placebo responses6 measured clini-
cally have been increasing over time7, making it
more difficult to detect a beneficial drug effect.
Large numbers of patients must therefore be
followed for long periods to detect treatment
effects. For example, a therapeutic trial for
ALS could take 2–3 years because survival is
the accepted outcome measure, whereas a trial
for rheumatoid or infectious disease might last
only 3–6 months8.
Fortunately, quantitative surrogate measures
often detect disease initiation or progression
more closely than clinical measures. They reveal
effects earlier and with fewer patients, thereby
reducing the cost of phase 2 trials to establish
proof of concept or dosing. Biomarkers custom-
designed to detect activity of a specific drug
candidate on its target can determine whether
a putative drug target was engaged during a
clinical trial. That way, even a trial that fails
to demonstrate efficacy will provide valuable
information about the disease mechanism.
Biomarkers for major neurodegenerative dis-
eases are badly needed but are poorly estab-
lished and often lacking altogether.
The Valley of Death
Traditionally, industry and academia have
worked together to combat disease. In the
United States, the National Institutes of Health
(NIH) spends about $28 billion per year to gen-
erate biomedical research discoveries. Industry
spends $15–40 billion translating those dis-
coveries into clinical medicines9. Sometimes
industry directly sponsors academic research.
In other cases, industry licenses research tools,
technologies or models from academia to
commentary
facilitate internal drug development programs.
The collaboration has been productive: of the
21 most important drugs introduced between
1965 and 1992, only five were developed solely
within the private sector10.
The last decade has wrought changes in
this relationship11. Major drug companies are
under tremendous financial pressure. The NIH
refocused its efforts to encourage academic
researchers to produce more tangible benefits
for human health12. Increased faculty interest
and funding for disease-related research natu-
rally led to increases in the generation of intel-
lectual property for academic institutions.
With this confluence of events—drug com-
panies with funds to invest and the increasing
productivity of academic research programs
devoted to finding therapies—why have we
failed to see an explosion in new leads going
from academia to industry to the clinic? Myriad
obstacles have constipated the flow. The path is
so fraught it has been given a name—the Valley
of Death—because many therapeutic strategies
start the journey but few finish (Fig. 1).
What can be done?
Despite appearances, the situation is far from
hopeless. There are initiatives that can be
pursued today. For example, Congress could
offer sufficiently improved patent protection
to disease-modifying therapies specifically
for Alzheimer’s and Parkinson’s diseases. To
qualify, a provision could be that the new drug
must produce overall heath care savings13. The
approach would not cost taxpayers anything
now, and any new drug that qualifies would
save taxpayers by lowering the cost of caring
for patients with Alzheimer’s or Parkinson’s
disease. By changing the risk/benefit ratio to
encourage greater investment from the private
sector14, we would enable industry to pursue
earlier-stage leads and thereby build bridges
from industry to academia.
The focus of this article, however, is on ini-
tiatives within academia to bridge the Valley of
Death to industry. It is the thesis here that key
components needed to discover and develop
effective neurotherapeutics already exist in
academia. Revising and reorganizing these
components could create more effective pro-
grams with the potential to survive the Valley
of Death.
Reduce the cultural differences between aca-
demia and industry. Academia and industry
are separated by a cultural gulf that severely
limits the ability of academic researchers to
do translational research. For example, drug
discovery and development are time intensive,
but the more intellectual property developed
around a lead program before it becomes pub-
lic knowledge, the more valuable it is to indus-
2 advance online publication nature medicine
try. By contrast, academic career advancement
almost universally depends on scholarship,
commonly measured as the frequency and
impact of publications. Academics jeopardize
their careers if they wait to publish a discovery
until it is sufficiently developed for industry
and protected by a patent application. If aca-
demic institutions are serious about foster-
ing translational research, they need to create
appropriate performance metrics for research-
ers, such as invention disclosures, patents and
licenses, and perhaps even consider a new
‘translational’ faculty track.
The cultural differences are also reflected in
the way research is pursued. Academics are lion-
ized for the tenacity, creativity and originality
with which they conceive and test hypotheses.
In the best case, breakthroughs occur, but often
only after focused effort requiring many cycles
of grant funding. In the worst case, substantial
academic careers can be spent on misguided
hypotheses that should have been abandoned
long ago. Neither outcome is acceptable in
industry, where R&D investments must be
justified in terms of more immediate returns.
As a consequence, industry has developed a
culture driven by milestones and punctuated
by performance-based GO/NO-GO decision
points. These cultural differences can create
problems for academics attempting to support
drug discovery efforts in their laboratories with
industry-sponsored research agreements15.
The differences in incentives, needs and
resources for researchers in academia and
industry undoubtedly contribute to addi-
tional culture gaps. The emphasis on hypoth-
esis-driven research and the scarce resources
available to purse them in academia necessar-
ily lead to small-scale experiments and limited
replicates. Unfortunately, unbiased analyses
of academic research findings have shown
a disappointing degree of reproducibility16.
Industry, on the other hand, must have tightly
reproducible assays to effectively pursue the
high-throughput screens (HTS) that are the
lynchpin of early-stage drug discovery. In
academia, an observant scientist can quickly
change course in response to a serendipitous
discovery. “In industry,” one expert told me,
“serendipity is a failure of quality assurance.”
Provide academic researchers with the right
resources for translational research. Academic
researchers often lack drug discovery resources,
especially expertise, tools and equipment. A
particular molecule or pathway might seem to
be a therapeutic target in a neurodegenerative
disease, but many academics would not know
whether and how to move these discoveries
forward. This deficiency can lead to signifi-
cant investment by academics in therapeutic
strategies that industry knows are difficult to
target pharmacologically. Arguably, the indus-
try definition of a ‘druggable’ target may need
revision, given the decreasing productivity of
industry R&D. But as a purely practical mat-
ter, industry remains the major option for a
partnership that would lead to a clinical trial.
Industry’s opinion is thus still important.
Even when academics identify and validate
a target that industry considers druggable,
they often lack the capacity to find hits and
develop them into therapeutic leads. Some
institutions have created facilities to perform
HTS, but assay options, chemical libraries and
medicinal chemistry expertise are limited.
Predictably, the value to industry of leads
from academic drug discovery programs can
be limited. To improve the options for aca-
demic researchers, the NIH has created the
Molecular Libraries Program (http://mli.nih.
gov/) and the National Chemical Genomics
Center (NCGC) with a screening capacity
equivalent to that of a major pharmaceutical
company17. Academic researchers, however,
must still evaluate leads that the NCGC devel-
ops in secondary assays. Unfortunately, those
researchers often lack the capacity to keep the
program moving.
Manage intellectual property. In 1980, the
Bayh-Dole Act gave ownership of intellectual
property created from US government–funded
research to the academic institutions where the
inventors worked18. Many other countries have
similar laws. As a result, academic institutions
established technology transfer offices (TTOs)
to decide which inventions to patent, and to
formulate terms of licensing agreements.
Intellectual property must be protected and
licensed under terms that assure industry part-
ners of a reasonable return on their substantial
investment for commercializing an invention.
Several common problems afflict TTOs.
A general lack of industry expertise within
TTOs often leads to erroneous valuations of
academic intellectual property19. A positive
incentive structure to conclude deals is often
missing. Fearing deals perceived to be too gen-
erous to industry, reluctant TTOs may take
an overly cautious approach that can kill the
technology transfer process. The TTO may lack
significant flexibility or creativity to structure
agreements to cover patent costs with start-ups
or small biotechnology companies, the major-
ity of licensees15. Unsurprisingly, the licensing
rates for intellectual property can be remark-
ably low—in the low single digits at some insti-
tutions20.
In academia, the inventor frequently grasps
the significance of the invention and can
articulate it much more effectively and with
greater credibility than TTO staff. Faculty
also may have better contacts with potential
 commentary

Drug discovery & development

Traditional purview Traditional purview

Valley of Death
of academia of industry

Assay development Drug development: Preclinical candidate

Target & small molecule hit to lead and (ADME and safety/ First in
identification Target validation screening lead optimization toxicology) human
Traditional NIH basic
tools available to academia
Therapeutics development

research laboratory

Academic and NIH

screening centers

Contract research organizations

Academic non-profit biotechnology incubator

Consortia of
clinicians and researchers

Traditional NIH
& foundation-supported research

Philanthropy and new foundation programs

mechanisms
Funding

New NIH Blueprint programs

Industry-sponsored research and venture capital

Drug development flow

Figure 1 The Valley of Death. Drug discovery and development is an orderly but fraught step-wise process. Traditionally, it has begun with NIH-supported
basic research and the identification of a therapeutic target. Ultimately, it leads to the development of a small molecule whose efficacy is tested in people
by industry. The steps in between are known as the Valley of Death, partly because of the high failure rate and partly because the expertise and resources
available to execute them are critically lacking. Consequently, the vast majority of leads that start the journey fail somewhere along the way. Recently,
individuals and institutions have begun to create new mechanisms to bridge the valley. ADME, absorption, distribution, metabolism and excretion.

partners11,19,21. After licensing, the indus-
try partner might want the inventor to be
involved in the development process, and the
time commitment for faculty can be consid-
erable. Participation is strongly influenced by
the extent to which the institution fosters an
entrepreneurial culture and allocates royal-
ties to the inventor22–25. Still, the career risk is
real for faculty doing substantial translational
research, especially those at early stages in their
career.
Focus each partner on what they do best.
Translating research discoveries to neurothera-
peutics involves a lot of moving parts, and the
mechanisms to fund the process have yet to
catch up (Fig. 1). There are three important
issues. First, what needs to be done to test the
hypothesis underlying the therapeutic strat-
egy? Second, can funding be found to reduce
risk sufficiently to attract an industry partner
with the resources to develop a discovery into
a clinical lead? Third, can a license be obtained
on financial terms that are consistent with the
long-term success of a for-profit company?
Ambitious, high-quality academic scien-
tists can often attract support from the NIH
to conduct a vigorous basic science program
on the pathogenesis of a particular disease.
Many patient-based foundations also offer
fellowships and grant support to academics
pursuing research into the causes of a particu-
lar neurodegenerative disease. Unfortunately,
once these research programs identify potential
therapeutic targets, financial support mecha-
nisms for further development and preclinical
testing thin out quickly.
The NIH has recognized this crucial prob-
lem, especially for orphan diseases. In response,
the National Institute for Neurological Diseases
and Stroke has created funding mechanisms to
cover every step from basic research and tar-
get identification to first-in-human clinical
trials. This development potentially provides
a complete path for academic drug discovery
and development with federal funds, which do
not dilute the value of the intellectual prop-
erty and therefore do make it more attractive
to partners.
The NIH still has work to do to realize the
potential of this blueprint for translational
research. Funds are limited: only a few pro-
grams can be supported at once, and those
tend to involve NIH program staff much more
intimately than investigator-initiated research
grants. For example, research programs are
often co-directed by a joint research commit-
tee of NIH staff and the applicant. Different
notions about drug development from NIH
staff and industry create strains and friction
as the program is executed. Funding rates are
relatively low, and a lot of time can be spent
fruitlessly on the application process. Some
of the programs require considerable drug-
development expertise to construct a compet-
itive application—expertise, as noted earlier,
that academics generally lack. Navigating
NIH offerings and determining appropriate
programs can be daunting. The NIH should
consider hiring consultants with industry
expertise to help academics assess the thera-
peutic potential of their discoveries, identify
appropriate funding mechanisms to support
development, and construct sensible develop-
ment programs for neurotherapeutics.
Patient-based foundations increasingly
want to support drug-discovery programs.
They have a unique role in encouraging their
members to participate in clinical studies, and
they have had an outsized effect on science by
using relatively modest fellowship and grant
support to recruit talented young scientists to

nature medicine advance online publication 3

commentary

Gladstone
Institutes Foundations National
Institutes
of Health
Taube-Koret Center
(Non-profit biotech
incubator) Contract Major
Academia • Biology & chemistry research pharma Therapeutics
• Drug development organizations companies
expertise
• Business development
& technology transfer
Biotechnology
Venture
capital

Figure 2 A model nonprofit biotechnology incubator to facilitate academic drug discovery and development. The Taube-Koret Center acts as a bidirectional
bridge between academia and industry to reduce the risk in the development of therapeutics. Discoveries with therapeutic potential are made in academic
laboratories and are transferred to the Taube-Koret Center within the same institution. The Center maintains the necessary expertise in drug discovery and
medicinal chemistry to validate targets and discover and optimize small molecules to modulate these targets. Contract research organizations are used by
the Center as needed to carry out specific steps in drug development. Leads that fail to meet performance targets are returned to the academic labs that
discovered them for further optimization. Leads that meet milestones are advanced until partners can be found to co-develop them further. The potential
paths to therapeutics via partners (black arrows) can involve foundations, the NIH, existing biotechnology companies, new companies created by venture
capitalists, or major pharmaceutical (pharma) companies. Since the Center’s overarching goal is to find effective therapies for neurodegenerative diseases,
rather than to make a profit from proprietary programs, it can make its new technologies, assays and disease models available to external entities that want to
evaluate the efficacy of their lead programs (gray arrows).

devote their careers to specific diseases. Some
have also invested in basic research, without
which many academics could not contemplate
a rational drug-discovery program. With a few
notable exceptions, they generally lack the
resources and expertise to support and over-
see stand-alone drug-discovery programs8.
One option for foundations is to collaborate
and back nonprofit academic drug-discovery
programs at whatever level is feasible. A single
foundation might have limited impact, but
some mechanisms of neurodegeneration, for
example, might be common. Collaboration
among foundations might enable their par-
ticipation in serious drug discovery efforts
and lead to synergies that help everyone and
leverage their investment.
In this context, major philanthropists might
have the most important role in enabling drug
discovery in academic settings. Generally, they
are motivated because they or a family mem-
ber have been affected by or are at risk for
a particular disease. The total cost of devel-
oping a new drug in 2000 was estimated to
be $800 million26,27, but only one-third of
the estimated early-stage costs of preclinical
development were direct costs. The remaining
two-thirds were ‘opportunity costs’ of tying
up capital for an extended drug-development
campaign. Used strategically, philanthropic
support can be leveraged five- to tenfold
to attract support from the NIH or private
sources to create a substantial drug-discovery
and development infrastructure. This unique
source of support can also be used to make the
academic drug-discovery process much more
robust and able to sustain programs through
potentially devastating funding gaps. Similar
to federal support, philanthropy can fuel the
development of therapeutic leads in a way
that adds substantial value and increases the
potential for partnerships8.
Venture capitalists (VCs) and ‘angel inves-
tors’ have been an important source of finan-
cial support for commercializing academic
research28. VCs supply substantial capital,
business development expertise, and indus-
try contacts that catalyze the development of
an academic discovery at a pace that would
otherwise be impossible. Unfortunately, the
recent global financial crisis and the reduced
opportunities for initial public offerings
have affected VC investment in biotech-
nology. Today, VCs are making fewer new
investments, and they tend to target discov-
eries at a later stage of development. Large
pharmaceutical companies have become the
main customers for VCs, which also affects
the disease indications that VCs pursue. VCs
should explore new, less risky investment
models. For example, rather than founding
new companies with expensive facilities and
management structures, VCs should explore
smaller, targeted investments to back devel-
opment within the inventor’s laboratory. The
creation of commercial incubators might also
be a lower-risk mechanism to rapidly advance
academic discoveries to a point where they
can, alone or in combination, become the
basis for an independent entity.
Regardless of the funding source, trans-
parency in constructing research budgets is
badly needed. The common NIH accounting
nomenclature of ‘direct’ and ‘indirect’ costs
has led foundations, donors and VCs to worry
that part of their support is used for purposes
unrelated to their interests. Yet institutions
cannot commit to do work that incurs finan-
cial loss. One solution is to change the struc-
ture of research budgets so that they include
all project-related costs and provide enough
detail so that donors can judge for themselves
whether their support is being used properly
and is tailored to the project.
Academic institutions must be transpar-
ent and able to deliver on programs they
create. Neurodegenerative diseases unfold
slowly. Consequently, the time that is neces-
sarily required to perform experiments and
make iterative progress with animal mod-
els of these slow diseases can be frustrating
for philanthropists and foundations with-
out backgrounds in biology, who need to
be assured by external review mechanisms
that their investment is a good one. More
important, academic institutions must see
the drug-discovery effort as part of their
mission. Donations that support this pro-
cess may then result in unrestricted revenue

4 advance online publication nature medicine


for the academic institution from the extra
licenses that are generated.
Two experimental models
Academia is working to implement these
ideas and to create innovative structures with
the capability to reduce the risk of developing
neurotherapeutics and successfully navigate
the Valley of Death. Two examples are offered
below.
The Taube-Koret Center: a nonprofit
biotechnology incubator. At the Gladstone
Institutes, we established a center that func-
tions as a nonprofit biotechnology incubator
to develop therapeutics for neurodegenerative
disease (Fig. 2). It offers a model of how drug
discovery can be conducted effectively in an
academic setting and suggests collaborating
roles for the institution and philanthropists.
The Taube-Koret Center (TKC; http://taube-
koret-center.org/) was established with phil-
anthropic support from the Koret Foundation
and Taube Philanthropies, with an initial focus
on Huntington’s disease. Several therapeutic
leads from this program have shown promise
in models of other diseases, such as Alzheimer’s
disease, Parkinson’s disease, ALS and fronto-
temporal dementia, so other donors are joining
to expand its scope.
Although the TKC operates similarly to a
biotechnology company, it resides within a
nonprofit research institute. Potential neuro-
therapeutic leads mostly come from NIH- and
foundation-supported basic research programs
of two investigators at the same institution. The
center is physically separate from the main aca-
demic labs, allowing a somewhat distinct, more
industry-like environment. It also provides aca-
demics an opportunity to contribute discoveries
to the Center for further development without
obliging them to do drug discovery themselves,
if their interests and strengths lie elsewhere. A
consultant experienced in drug-development
reviews potential targets from these research
programs and advises on milestones and per-
formance-based GO/NO-GO decision points.
Leads that fail to meet performance milestones
are replaced with the next most promising
leads. The consultant, who has the necessary
experience to negotiate agreements in the area
of neurodegeneration research, works closely
with the TTO to develop an overall intellectual
property strategy with greater value for poten-
tial industry partners by protecting later-stage
intellectual property.
TKC conducts drug discovery and devel-
opment through a combination of internal
programs and external contract research orga-
nizations (CROs). For example, target identi-
fication, validation, HTS and in vivo efficacy
trials are all done on site. TKC has medicinal
nature medicine advance online publication 5
chemistry capabilities on site, but these capa-
bilities focus on in silico screening and analysis,
on complex, small-scale small-molecule syn-
theses, and on addressing the bioavailability
challenge posed by the blood-brain barrier.
Simpler analog chemistry or scale-up chemis-
try is done through CROs with guidance from
our consulting medicinal chemists. CROs are
also used for absorption, distribution, metab-
olism, excretion, toxicology and other studies
required for later-stage preclinical drug devel-
opment.
As a nonprofit incubator, TKC’s overarch-
ing goal is to find therapeutics for neurode-
generative disease. This has important policy
implications. First, TKC is purposely struc-
tured to identify and investigate as many
therapeutic targets and leads as possible in
the belief that, as drug discovery is inherently
risky, the more leads that can be developed,
the better the chance of finding one that works
in a reasonable time. Partnerships bring cru-
cial expertise and free up internal resources.
Therefore, leads are partnered at the earliest
stage possible. Should a partnership fail after
it is initiated, the rights to develop the lead
revert to TKC. Because our goal is to find a
therapeutic, we do not restrict access to the
TKC to homegrown leads. The TKC has estab-
lished models of neurodegeneration based on
mutant huntingtin, TDP-43, SOD1, LRRK2,
synuclein and APP and has invented platform
technology to conduct drug discovery in pri-
mary cells. These resources have been made
available to biotechnology or major pharma-
ceutical companies who have small-molecule
leads to test.
What advantages for CNS drug discovery
might a nonprofit academic biotechnology
incubator have over a for-profit biotechnol-
ogy company? First, an academic incubator
is shielded from the structure imposed on
for-profit companies by investors, which is
geared to provide a timely financial return.
The financial depth and stability of major aca-
demic institutions represent a more enduring
structure than many biotechnology companies.
As a result, pursuit of optimal drug discovery
can drive decisions, and philanthropists and
foundations can invest with confidence. The
incubator avoids incurring early-stage trans-
actional costs to acquire intellectual property
normally borne by biotechnology companies
because the feeder academic laboratories are
at the same institution. These laboratories also
provide the incubator with a diverse ‘pipeline’,
ensuring that efforts can be sustained even as
specific lead programs fail. Constrained by
conventional GO/NO-GO decision points,
biotechnology companies might abandon
leads prematurely and miss a potentially cru-
commentary
cial development opportunity. Instead, the
incubator can transfer the program back to
the academic laboratory from which it came
for further optimization.
Notably, the goal of the incubator is to com-
plement, not replace, biotechnology or phar-
maceutical company activities by generating
promising leads and sufficiently ‘de-risking’
them to warrant partnerships. As such, the
incubator is freer to pursue innovative thera-
peutic targets for Alzheimer’s or Parkinson’s
disease that may be seen as too risky for early-
stage R&D by industry or to pursue disease
indications, such as Huntington’s disease or
ALS, that are sometimes seen as too small to
attract early-stage VC investment. If success-
ful, both goals should lead to more and bet-
ter therapeutic options for neurodegenerative
disease.
Alzheimer’s Disease Neuroimaging
Initiative: a model public-private partnership
focused on biomarkers. Lack of good biomark-
ers contributes to the high cost of clinical tri-
als, which limits the number of studies and
the companies pursuing neurodegenerative
disease overall. The investment needed to find
biomarkers can be so high that, for the NIH
to take it on alone, support for other research
would be severely reduced. Similarly, com-
panies have difficulty justifying biomarker
programs, given the uncertain and long-term
nature of the potential benefit.
In 2004, the Alzheimer’s Disease
Neuroimaging Initiative (ADNI) was formed
to investigate relationships among clinical,
cognitive, imaging, genetic and biochemical
biomarkers characteristic of individuals devel-
oping Alzheimer’s disease29,30. These data will
help clinicians to diagnose Alzheimer’s disease,
validate biomarkers for disease progression and
design better clinical trials30. This extremely
ambitious study has received $60 million: $40
million from the National Institutes on Aging
and $20 million from 13 pharmaceutical com-
panies and two foundations. The study now
involves academic researchers at 58 sites in the
United States and Canada.
ADNI is impressive for many reasons, not
least of which is that it was conducted by a
consortium of highly successful researchers29.
Remarkably, to accelerate research, the aca-
demics leading the cooperative project agreed
to make their unpublished data publicly avail-
able soon after it is collected. Research teams
that are not formally part of ADNI have even
analyzed and published using ADNI data.
Historically, building research consortia has
been challenging. Academic career advance-
ment is substantially based on contributions
of individuals to their chosen scientific field.
Assessing those contributions within large con-
commentary
sortia is difficult, and promotion committees
struggle to assign values to multiauthored con-
sortium publications. Indeed, some institutions
instill an anticollaborative culture, and team
science is viewed negatively. Unfortunately,
some problems in neurodegeneration are so
large it is hard to imagine solving them with-
out large teams of collaborating scientists. It is
encouraging to see that the ADNI approach has
been replicated in other countries. Hopefully,
cultural changes are under way within aca-
demic institutions to place a higher value on
team science.
Recognizing that the establishment of bio-
markers is risky but necessary for developing
therapeutics for Alzheimer’s disease, ADNI
has spread the risks and benefits among many
interested partners. More broadly, the approach
highlights the idea of identifying precompeti-
tive space in neurodegeneration research that
could be successfully pursued by international
consortia and funded with pooled resources
from public and private sources8. Perhaps the
recently announced agreement among major
pharmaceutical companies to share data from
11 failed clinical trials in Alzheimer’s disease
will also provide valuable insights that will lead
to more successful future clinical trials, even
if this initiative is more an act of desperation
than a new dawn of genuine collaboration31.
Summary
Developing therapies for neurodegenera-
tive diseases is extremely challenging. As the
main developer of therapies, the for-profit
pharmaceutical industry has done the calcula-
tion. Although a therapy for neurodegenera-
tive disease would be highly profitable, it has
determined that its R&D dollar can be spent
on lead programs for other indications that are
less risky.
In view of the looming demographic
changes, action on neurodegenerative diseases
is urgent. Every player has a role. Academic
institutions need to demonstrate their com-
mitment to translational research by changing
the cultural impediments that force academic
researchers to abandon these lines of investiga-
tion or leave academia for industry. Congress
6 advance online publication nature medicine
will face back-breaking obligations to Medicare
and Medicaid unless a cost-saving treatment is
found32. More funding is critically needed for
the National Institute for Neurological Diseases
and Stroke and the National Institute on Aging,
the NIH institutes primarily responsible for
basic and translational research in neuro-
degenerative diseases. The NIH has moved
resources into supporting drug discovery and
development within academia; yet, opportu-
nities vastly exceed resources. VCs have been
leery of early-stage drug discovery for neuro-
degenerative disease, but they have recently
been exploring new and more cost-effective
ways to fund early-stage development efforts.
Philanthropists may have the greatest opportu-
nity and ability to materially enable academic
drug discovery and development.
Neurodegenerative diseases are not going
away. The cost of ignoring them will break our
fragile healthcare systems. Our only hope is
through focused research to find therapies. With
concerted action from academia, the govern-
ment and philanthropy, the risks in developing
those therapies can be substantially lowered. In
turn, industry will respond with greater atten-
tion and investment, and therapies will result.
ACKNOWLEDGMENTS
This work was supported by the Taube-Koret
Center for Huntington’s Disease Research, by NIH -
National Institute of Neurological Disorders and
Stroke grants 2R01 NS039074 and 2R01 NS045091
and NIH - National Institute on Aging grant 2P01
AG022074, and by The J. David Gladstone Institutes.
I thank S. Williams, S. Freedman, M. Sutherland,
F. Dorey, M. Lopez, L. Bruijn, R. Blumenstein, R.
Pacifici, N. Wexler, C. Johnson, C. Austin, L. Vetter,
A. Grove, P. Lansbury, T. Reisine, G. Naeve and
members of the Finkbeiner laboratory for helpful
discussions and/or comments on an early version of
this manuscript. I thank G. Howard and S. Ordway for
editorial assistance, and K. Nelson for administrative
assistance.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
1. Alzheimer’s Association. Alzheimers Dement. 6, 158–
194 (2010).
2. Centers for Disease Control and Prevention and The
Merck Company Foundation. The state of health
and aging in America (Merck Company Foundation,
Whitehouse Station, New Jersey, USA, 2007).
3. Alzheimer’s Association. Alzheimers Dement. 5, 234–
270 (2009).
4. Ferri, C.P. et al. Lancet 366, 2112–2117 (2005).
5. Acemoglu, D. & Linn, J. Q. J. Econ. 119, 1049–1090
(2004).
6. Benedetti, F., Carlino, E. & Pollo, A.
Neuropsychopharmacology published online,
doi:10.1038/npp.2010.81 (30 June 2010).
7. Spiegel, A. The growing power of the sugar pill. National
Public Radio <http://www.npr.org/templates/story/story.
php?storyId=124367058> (2010).
8. Hanson, S., Nadig, L. & Altevogt, B. Venture philan-
thropy strategies to support translational research:
workshop summary <http://www.nap.edu/cata-
log/12558.html> (Forum on Neuroscience and Nervous
System Disorders, Institute of Medicine of the National
Academies, 2009).
9. Congressional Budget Office. Research and develop-
ment in the pharmaceutical industry (US Government
Printing Office, 2006).
10. Cockburn, I.M. & Henderson, R.M. in NBER Innovation
Policy and the Economy vol. 1 (eds. Jaffe, A.B., Lerner,
J. & Stern, S.) 20–21 (National Bureau of Economic
Research, Cambridge, Massachusetts, USA, 2000).
11. Young, T.A. Int. J. Intellect. Prop. Law Econ. Manage.
1, 13–18 (2005).
12. National Institutes of Health. Blueprint neurotherapeu-
tics network. NIH Blueprint for Neuroscience Research
<http://neuroscienceblueprint.nih.gov/bpdrugs/index.
htm> accessed 23 August 2010.
13. Duggan, M. J. Health Econ. 24, 1–31 (2005).
14. Lanjouw, J.O. & Cockburn, I.M. World Dev. 29, 265–
289 (2001).
15. Fraser, J. Tomm. Technol. Trans. 1, 9–20 (2009).
16. Ioannidis, J.P.A. PLoS Clin. Trials 2006, e36 (2006).
17. Inglese, J. et al. Proc. Natl. Acad. Sci. USA 103,
11473–11478 (2006).
18. Boettiger, S. & Bennett, A.B. Nat. Biotechnol. 24,
320–323 (2006).
19. Thursby, J. & Thursby, M. AUTM J. 12, 9–22 <http://
www.provendis.info/fileadmin/info/pdfs/1255.pdf>
(2000).
20. Thursby, J., Jensen, R. & Thursby, M. J. Technol.
Transf. 26, 59–72 (2001).
21. Lockett, A. & Wright, M. Res. Policy 34, 1043–1057
(2005).
22. Friedman, J. & Silberman, J. J. Technol. Transf. 28,
17–30 (2003).
23. Markman, G., Phan, P., Balkin, D. & Gianiodis, P.
J. Technol. Transf. 29, 353–364 (2004).
24. Lach, S. & Schankerman, M. J. Euro. Econ. Assoc. 2,
252–264 (2004).
25. Link, A.N. & Siegel, D.S. Eur. J. Finance 11, 169–181
(2005).
26. DiMasi, J.A., Hansen, R.W. & Grabowski, H.G.
J. Health Econ. 22, 151–185 (2003).
27. Adams, C.P. & Brantner, V.B. Health Aff. 25, 420–428
(2006).
28. Reinhardt, U.E. Health Aff. 20, 136–149 (2001).
29. Aisen, P.S. et al. Alzheimers Dement. 6, 239–246
(2010).
30. Weiner, M.W. et al. Alzheimers Dement. 6, 202–211
(2010).
31. Anonymous. Economist 395, 81–82 (17 June
2010).
32. Bynum, J. Characteristics, costs, and health service
use for Medicare beneficiaries with a dementia diag-
nosis. Report 1: Medicare current beneficiary survey
(Alzheimer’s Association, 2009).