We have found familial encephalopathy with neuroserpin inclusion bodies (FENIB) in two Caucasian families living in the USA; both are affected by pre-senile dementia. In the larger family, affected individuals present clinically around the fifth decade of life with cognitive decline including deficits in attention and concentration, response regulation difficulties and impaired visuospatial skills. Memory is also impaired, but to a lesser degree than is typically seen in patients with Alzheimer's disease. After several years of disease progression, most affected individuals are unable to work and eventually require nursing-home care. The second, much smaller family shows an earlier clinical onset, during the second or third decades of life, and presents with both epilepsy and progressive cognitive decline ending in institutionalization9. The principal neuropathologic finding in both families is the presence of eosinophilic neuronal inclusions distributed throughout the deeper layers of the cerebral cortex and in many subcortical nuclei, especially the substantia nigra (Fig. 1a). These inclusions, which we have named Collins bodies, are round, 5–50 µm in diameter and strongly periodic acid-Schiff (PAS) positive, but diastase resistant. They are distinctly different from numerous previously described entities, including Lewy bodies, Pick bodies and Lafora bodies10. The inclusions were obtained as an enriched fraction from the postmortem brain of an affected individual (Fig. 1b), and SDS-polyacrylamide gel electrophoresis (SDS–PAGE) showed them to be composed primarily of a single major protein (Fig. 1c), identified as neuroserpin from amino-acid sequencing (data not shown) and confirmed by western blotting ( Fig. 1c). The neuroserpin protein consists of 410 amino acids and is glycosylated to a final gene product of relative molecular mass 55,000 ( Mr 55K) (ref. 11). The neuroserpin gene (PI12) has been mapped to 3q26 (ref. 11). DNA sequencing of polymerase chain reaction (PCR) products designed to flank all eight coding exons of two affected individuals from the large family revealed only a single point mutation: a T to C transition at nucleotide 226 (Fig. 2a). This mutation, which we identify as PI12Syracuse, results in the substitution of Ser 49 in neuroserpin by proline (S49P). The T to C transition creates an MnlI restriction site, which makes screening for this mutation in the family straightforward. Therefore, primers were designed to generate a 344-base-pair (bp) PCR product covering this selected region of exon 2. MnlI digests of the PCR product generated from the wild-type neuroserpin gene give two fragments, of 254 and 90 bp. However, the T to C transition in PI12Syracuse creates an MnlI restriction site within the 254-bp product, giving additional products of 158 and 96 bp. Because affected individuals are heterozygotes, MnlI restriction digests of the PCR product of exon 2 will give fragments of 254, 158, 96 and 90 bp for individuals carrying this mutation. This mutation occurs in 14 individuals out of the 38 family members screened (Fig. 2b, c); conversely, it was not detected upon screening 120 chromosomes from 60 unrelated Caucasian individuals.

Figure 1: Identification, purification and composition of Collins bodies.
figure 1

a, Haematoxylin and eosin (H and E) stained section of cerebral cortex showing presence of Collins bodies (arrows). b, Purified fraction of Collins bodies stained with H and E. Scale bar, 50 µm. c, SDS-PAGE of purified Collins bodies solubilized by heating at 75 °C in 4% SDS for 2 h: lane 1, Coomassie blue stain; lane 2, western blot with antineuroserpin antibody.

Figure 2: Identification and genotyping of PI12Syracuse mutation.
figure 2

a, Electropherogram showing heterozygosity in PI12 exon 2 sequence corresponding to nucleotide 226. b, MnlI digestion of a 344-bp PCR product generated from exon 2 from genomic DNA. Restriction digests of the wild type give two fragments of 254 and 90 bp. The T to C transition in the PI12Syracuse mutation creates an additional MnlI restriction site within the 254-bp product giving additional fragments of 158 and 96 bp. M indicates the presence of the mutation. The data are representative results obtained from some of the 40 family members screened. c, FENIB pedigree with S49P mutation. Roman numerals represent the generation, and arabic numbers represent the individuals within that generation. The phenotype is represented by coloured diamonds and the genotype by bars. The phenotype (FENIB) is defined either by history (red), by neuropathological examination (black) or by neurological examination (blue). Family members who were clinically suggestive of expressing the FENIB phenotypes are indicated by yellow. Asymptomatic individuals are indicated by (white). The wild-type allele is represented by open bars and S49P mutation by filled bars. The pedigree has been disguised to protect the confidentiality of the family.

The disease phenotype was ascertained either by observing Collins bodies in brain tissue or by obtaining a history of pre-senile dementia together with evidence of cerebral and cognitive dysfunction on neurological examination. Individuals without these findings were considered asymptomatic. As shown in Fig. 2c, all clearly affected individuals are heterozygous for the mutation. Conversely, some heterozygotes are asymptomatic (Fig. 2c); however, all of these individuals are less than 50 years of age. The pattern of inheritance showing affected individuals in each generation and in both genders is consistent with an autosomal dominant mode of transmission. Five age-related classes of disease penetrances were defined: 5% if younger than 45; 42% if 45 or older, but younger than 50; 70% if 50 or older, but younger than 53; 83% if 53 or older, but younger than 56; and 95% if 56 or older. Using these age-related classes of disease penetrance, a maximum lod (logarithm of the odds favouring linkage) score of 3.40 at zero recombination between the disease and PI12Syracuse was calculated. Thus, the co-segregation of the mutation and the disease is consistent with the interpretation that PI12Syracuse is the disease gene. The conclusion that a mutant neuroserpin can result in disease is strengthened by our discovery of a second familial occurrence of FENIB; two family members in two generations revealed a single base change in the neuroserpin gene which translates into a missense mutation to an arginine at Ser 52 (S52R). This mutation, designated as PI12Portland, is found in both the mother and her older child, but not in two unaffected family members (data not shown).

The identity and significance of these mutations are put beyond doubt by their predictable consequences12. Neuroserpin is a typical member of the serpins, a family of serine proteinase inhibitors that share a tightly conserved tertiary structure. The inhibitory function of the serpins depends on their ability to undergo a profound conformational transition, involving an opening of the main five-stranded sheet of the molecule, and allowing the insertion of the reactive centre loop as an extra middle strand. The control of this sheet-opening mechanism is crucial to the stability of the molecule, and mutations that favour the ready opening of the sheet facilitate pathological intermolecular linkages, with the reactive loop of one molecule inserting into the opened sheet of the next to give loop–sheet polymers13. Such loop–sheet polymerization of conformationally unstable mutants of α1-antitrypsin results in inclusion-body formation in hepatocytes, with the subsequent development of liver cirrhosis14. Many dysfunctional variants of the serpins are now recognized, and an overall analysis of these variants shows a clustering of mutations in the domain centred on Ser 53 at the commencement of the B-helix15; Ser 53 is the serpin template number for Ser 49 in neuroserpin, which when substituted by proline will destabilize the sheet opening mechanism, probably leading to formation of loop–sheet polymers. A direct precedent for this comes from a mutation at the same site in α1-antitrypsin Siiyama8, with replacement of the serine by a phenylalanine resulting in spontaneous polymerization of the variant16 and consequent formation of intracellular inclusions.

Confirmation that the same mechanism applies to the S49P variant of neuroserpin is provided by electron microscopy of the intraneuronal inclusions isolated from the FENIB family (Fig. 3). These show that the whole inclusions are formed by entangled fibrils which immunogold label for neuroserpin. To look more closely at the morphology of individual fibrils, the inclusions were sonically fragmented for 5 min to give short-chain filaments, and these again have the typical appearance of other serpin loop–sheet polymers (Fig. 3). Supporting evidence for the loop–sheet basis of the linkage is provided by the biochemical properties of the fibrils, which were non-dissociable in 1 M potassium chloride but dissociated to a single band on SDS-PAGE in the absence of thiol-reducing agents. As expected for such polymers, there was a complete absence of both inhibitory activity and loop cleavage, even in the presence of a 10-fold excess of the cognate proteinase, tissue plasminogen activator. Furthermore, the histological appearance of the inclusions, as PAS-positive diastase-resistant bodies, emphasizes the identity of the general pathophysiology with the late-stage blockage of processing and secretion which results in nearly identical histology with the closely homologous variant of α1-antitrypsin. Finally, the clinical presentation of the two FENIB families closely matches the severity of their individual mutations. Although the domain controlling the opening of the main sheet of the molecule is centred on Ser 49 in neuroserpin, another adjacent conserved serine (56 in the template, 52 in neuroserpin) forms an even more critical interaction with the sheet12. In keeping with the predicted gradation in severity of the molecular lesions, the neurodegeneration presents some 20 years earlier with S52R than it does in the family with the S49P mutation.

Figure 3: Electron microscopy of Collins bodies and serpin polymers.
figure 3

a, Intact Collins bodies. b, Sonicated neuroserpin filaments. c, Sonicated α1-antitrypsin Siiyama polymers. Scale bars, 100 nm.

The detailed models provided from similar lesions in other serpins immediately give a clear overall molecular mechanism for FENIB that is as yet incomplete for the other familial dementias. The models from other serpins includes diseases arising from mutations in antithrombin17 and C1-inhibitor14,18,19 (Fig. 4), but most pertinent of all to the neuroserpin abnormality is the model provided by the various mutations of α1-antitrypsin resulting in liver disease. These answer the question posed by other neurodegenerative disorders, such as the spongiform encephalopathies20 and Alzheimer's disease21, as to whether protein deposition and accumulation is in itself sufficient to explain the late-onset dementia. The answer from α1-antitrypsin-associated liver disease is clear; hepatocyte loss and eventual cirrhosis is a consequence of variant deposition and not loss of function, as cirrhosis only develops with a conformationally unstable variant and not with those mutations giving complete ‘null’ suppression of synthesis. Where the model provided by α1-antitrypsin differs from that of neuroserpin is that, although in both cases a heterozygous abnormality is sufficient to produce large intracellular inclusions, it is usually only in homozygotes for α1-antitrypsin variants that the accumulation is severe enough to cause liver damage22. The two variants of neuroserpin, however, result in an autosomal dominant disease, consistently presenting as a dementia in young to middle-aged adults. This may reflect the grossness of the mutations with the replacement of Ser 49 by the imino acid proline or of the critical Ser 52 by arginine. The serious consequences of neuroserpin instability are relevant to the dementias as a whole because of the much greater risk that protein accumulation poses to long-lived and non-dividing cells such as neurons; the damage to neurons from such protein deposition will be cumulative and irreversible, in keeping with the observed consistency and late onset of presentation of the dementias in general. FENIB thus conforms to the general pattern of neurodegeneration caused by aberrant protein processing and tissue deposition.

Figure 4: Serpin polymerization.
figure 4

a, Active serpin showing the initial entry of the reactive loop (red) into the partially opened A-sheet12 (blue). The broken circle outlines the domain controlling further opening of the sheet with the critical Phe-Ser-Pro sequence (black) at the start of the B-helix. As indicated by the arrow, the loop can further insert or, in the presence of mutations, be replaced by the loop of another molecule to give a loop–sheet intermolecular linkage. b, The homologous sequences of the domain in neuroserpin and α1-antitrypsin with replacements resulting in disease (red for neuroserpins Syracuse and Portland; blue for other polymerogenic serpins). The S53P mutation in α1-antitrypsin Siiyama is bold blue. c, Model of sequential loop–sheet linkage in serpin polymers (reproduced with permission26).


Collins bodies

The inclusions were identified in sections of cerebral cortex by staining with haematoxylin and eosin (H and E). Brain tissue was fixed in 10% formalin, embedded in a paraffin, sectioned at 8 µm and stained with H and E by standard procedures. The Collins bodies were isolated from 2 g of unfixed, frozen cortical tissue obtained at autopsy from an affected individual (provided by the Harvard Brain Tissue Resource Center). The isolation scheme consisted of homogenization in 10 ml 250 mM sucrose, 10 mM EDTA, 10 mM HEPES, pH 7.4, containing 10 µl of protease inhibitor cocktail (Sigma). The bodies were collected by centrifugation at 45,000 g for 20 min, and subjected to 2 cycles of detergent washes (1% n-lauryl sarcosine), collagenase digestion (1 mg ml-1), and a final wash in the homogenization medium. After each step, the bodies were collected by centrifugation and viewed microscopically after H and E staining. The bodies were solubilized by heating at 75 °C for 2 h in 4% SDS, 125 mM Tris-HCl, pH 6.8, 20% glycerol, 10% mercaptoethanol. After solubilization, the supernatant fraction was separated by SDS-PAGE in a 7.5% gel and stained for either protein with Coomassie blue or for neuroserpin using an anti-neuroserpin antibody diluted 1/2,500 in 10% blotto.

Genomic DNA

Peripheral blood was used as a source of genomic DNA, which was extracted using inorganic extraction kits (Oncor). Paraffin-embeded, formalin-fixed brain biopsy material was extracted using standard procedures.


Amplification was performed in 50–100 µl reactions containing 500 ng genomic DNA, 0.2 mM each dNTP, 1.5 mM MgCl2, 0.5 µM each oligodeoxynucleotide primer, 100 µg ml-1 gelatin, 10 mM Tris-HCl, pH 8.3, 50 mM KCl2. A small wax pellet was added to each PCR reaction, which was heated at 95 °C for 5 min, and cooled on ice before the addition of 10 µl (1.5 U) Taq DNA polymerase (AmpliTaq Perkin-Elmer Cetus). PCR cycling was done on a Hybaid Omnigene thermocycler using the following parameters: one cycle of 2.5 min at 94 °C, 1 min at 55 °C, and 1 min at 72 °C; 34 cycles of 1 min at 94 °C, 1 min at 55 °C (PI12 exon 6) or 58 °C (all other PI12 exons) and 1 min at 72 °C (with the addition of 1 s per cycle); and a single cycle of 10 min at 72 °C. Five µl of PCR product diluted 1/3 with 95% formamide loading buffer plus dye (20 µl total volume loaded) was heated at 85 °C for 3 min, cooled rapidly on ice, and then separated on the gel at 50 °C for between 5,000 and 15,000 Vh depending on the fragment sizes being detected23. PCR-amplified products were detected by silver staining24.

PCR primers from the PI12 introns flanking the coding exons were designed from the human PI12 genomic sequence. exon 2: (NSE2F) CAACATATCCTTCCATGAGAC; (NSE2R) ACAACATAACTGAGTCAAAGTC 402-bp product; (NSE2S) (AGCTCGAGATTC)GCTTGAAACTGTTACAATATGGC 344-bp product. The bracketed sequence contains multiple restriction sites not present in the genomic sequence and which are not necessary if not cloning the PCR product. The use of NSE2S, rather than NSE2F, with NSE2R made the detection of the PI12Syracuse mutation easier on 2% w/v agarose gels following MnlI digest. Exon 3: (NSE3F) GCTGTGCTTTAATGCTCCTCC; (NSE3R) GATACCCAACTCAAATGCTCTC 435-bp product. Exon 4: (NSE4F) GGTCCCCCTTGATCTTCCAG; (NSE4R) GCCCAAATTAGATTGACAAGG 362-bp product. Exon 5: (NSE5F) TGATAGGCATCTTTTATGGCC; (NSE5R) TGATAGCCCATCGTCCATGC 292-bp product. Exon 6: (NSE6AF) TGTTCCAGGTAACAAGATGCTC; (NSE6AR) CAGAACTCCATAGTAAATAGAGG 245-bp product. Exon 7: (NSE7AF) CCTAGGTTTTCTTCAGTATCC; (NSE7AR) CAAATCTAGAAGAGGGGAGAAG 242-bp product. Exon 8: (NSE8F) AGGTGGATAGGCATAGATGG; (NSE8R) AGTACTGAGGAAAAAAATCTCC 261-bp product. Exon 9: (NSE9F) TTATTATTTTGTTCACCCCCTC; (NSE9R) GTTTCCAAAGTTCACCTAGAG 435-bp product. DNA sequencing of the purified PCR products was carried out by the BioResource Center, Cornell University, Ithaca, New York.

Characterization of neuroserpin isolated from patients with FENIB

SDS-PAGE, western blot analysis, the interaction of neuroserpin with tissue plasminogen activator, electron micrographs and immunogold labelling were performed as described16. Figures were produced with MOLSCRIPT25.