Just as genetic variants can be tied to altered gene expression, they can also be pegged to protein changes. Researchers led by Carlos Cruchaga and Oscar Harari, both at Washington University School of Medicine, St. Louis, linked Alzheimer's disease-associated loci to altered protein amounts in brain tissue, cerebrospinal fluid, and plasma from people with AD. Using these protein quantitative trait loci (pQTLs), they identified risk genes from GWAS hits—and, while they were at it, perhaps even some candidate drugs.

  • Loci pegged to protein changes in AD brain, CSF, plasma.
  • More than 70 percent of brain loci also turned up in the fluids.
  • Could some of 25 drugs known to target the altered proteins be repurposed?

Alzforum previously covered a medRxiv preprint of this work (Feb 2021 news). Now published in the July 8 Nature Neuroscience, the study includes more data about how pQTLs overlap in the brain, CSF, and plasma, and how they relate to other QTLs or disease risk genes. The researchers also earmarked some altered proteins as potential targets for repurposing FDA-approved drugs.

As reported in their preprint, first author Chengran Yang and colleagues searched for pQTLs by correlating more than 14 million SNPs with protein levels in parietal lobe tissue, CSF, and plasma from 428, 971, and 636 older donors, respectively. Of those, 297, 249, and 230 participants had AD; the rest were cognitively normal. The researchers found 32 pQTLs in brain tissue, 274 in the CSF, and 127 in the plasma.

In the peer-reviewed paper, the researchers reported how many loci overlapped between the tissue types. More than 70 percent of all brain tissue loci also popped up in the fluids, once again confirming these fluids as proxies for what is happening in the brain. The CSF shared more pQTLs with the brain than did the plasma. Some loci tweaked levels of more than one protein, with the most extreme being an APOE variant that altered 13 different CSF proteins (see image below).

One Locus, Many Proteins. An APOE-TOMM40 locus altered the concentration of 13 CSF proteins (left), while an ABO variant affected seven plasma proteins (middle) and an SPCS3-VEGFC variant, five brain proteins (right). Thicker lines reflect a stronger effect size. Red indicates higher, blue lower protein concentration in the respective tissue. [Courtesy of Yang et al., Nature Neuroscience, 2021.]

To better understand exactly how these pQTLs might control protein levels, Yang and colleagues determined whether they overlapped with expression, splicing, or epigenetic QTLs. The assumption is that if two QTLs co-localize, then that variant alters protein levels through that mechanism. Within the three sample types, 16 to 28 percent of pQTLs altered gene expression, 3 to 17 percent changed splicing, 1 to 10 percent altered DNA methylation, and less than 2 percent modified histone acetylation. On the other hand, 48 to 77 percent of all pQTLs did not overlap with any of the other QTLs, indicating that those variants alter protein levels in other ways. “They may modify protein secretion, cleavage, or receptor expression,” Cruchaga told Alzforum.

How did these variants relate to known risk alleles? Using Mendelian randomization to link GWAS data to the pQTLs, the geneticists identified which protein-altering variants also increase risk of Parkinson’s disease, frontotemporal dementia, amyotrophic lateral sclerosis, and stroke, from within respective GWAS loci found for these diseases. These included the plasma myeloid cell receptor CD33 for AD, the plasma lysosomal hydrolase IDUA for PD, CSF carbonic anhydrase IV for ALS, and CSF or plasma E-selectin for stroke. The scientists confirmed these findings with co-localization, showing that 42.5 percent of the loci flagged as pQTLs were also GWAS risk loci.

Last but not least, the authors searched for drugs known to interact with these disease-associated proteins. Of the 25 they found, one is AVE9633, an anti-CD33 antibody that was discontinued after a Phase 1 trial for acute myeloid leukemia (Maakaron et al., 2021). Another CD33 antibody, AL003, is currently in Phase 1 for AD (May 2019 conference news). When expressed on microglia, CD33 hampers TREM2 signaling (Jul 2019 news; Apr 2019 conference news). These new proteomic data validate the premise of therapeutically suppressing CD33, Rudy Tanzi, Massachusetts General Hospital, Boston, wrote to Alzforum (full comment below).

Another candidate drug is chondroitin sulfate, a glycosaminoglycan and component of cartilage that is sold as a supplement. The authors place chondroitin sulfate into the α-L-iduronidase (IDUA) pathway, which degrades glycosaminoglycans in the lysosome. IDUA deficiency causes the recessive lysosomal storage disorder mucopolysaccharidosis type I, and as such is a target of experimental enzyme replacement gene therapy (Maccarana et al., 2021; Hordeaux, 2019). 

Two FDA-approved drugs are acetazolamide, a diuretic sold as Diamox, and the beta-blocker carvedilol; they inhibit carbonic anhydrase and E-selectin, respectively. “We are considering what compounds to move into clinical trials,” Cruchaga said.

The researchers made their summary statistics and data freely available for download.—Chelsea Weidman Burke


  1. The new data from the Cruchaga lab are very intriguing and shine a particular spotlight on the role of CD33 in AD pathogenesis. We first described the genetic association of CD33 with AD risk in a family-based GWAS in 2008 (Bertram et al., 2008) and later reported increased expression of CD33 in microglia in AD brain (Griciuc et al., 2013).

    We also showed that inactivating or reducing levels of CD33 leads to increased phagocytosis of and clearance of Aβ by microglia, while TREM2 has opposite effects (Griciuc et al., 2013Griciuc et al., 2019).

    Based on these findings, we have proposed that reducing CD33 levels or inactivating CD33 would be a potentially promising therapeutic for the prevention and treatment of AD. The new proteomic data from the Cruchaga lab strongly validates this premise.

    Along these lines, we have recently shown that gene therapy aimed at reducing microglial expression in FAD mice significantly ameliorates AD pathology (Griciuc et al., 2020). For purposes of disclosure, we have two issued patents on using gene therapy and immunotherapy to reduce CD33 levels and inactivate CD33, respectively, for the prevention and treatment of AD and other neurodegenerative disorders.


    . Genome-wide association analysis reveals putative Alzheimer's disease susceptibility loci in addition to APOE. Am J Hum Genet. 2008 Nov;83(5):623-32. PubMed.

    . Alzheimer's disease risk gene CD33 inhibits microglial uptake of amyloid beta. Neuron. 2013 May 22;78(4):631-43. PubMed.

    . TREM2 Acts Downstream of CD33 in Modulating Microglial Pathology in Alzheimer's Disease. Neuron. 2019 Sep 4;103(5):820-835.e7. Epub 2019 Jul 10 PubMed.

    . Gene therapy for Alzheimer's disease targeting CD33 reduces amyloid beta accumulation and neuroinflammation. Hum Mol Genet. 2020 Oct 10;29(17):2920-2935. PubMed.

  2. This is a well-done study that extends prior GWA pQTL studies by adding samples from AD subjects. These multi-omics approaches are changing the landscape of genetic-based targets. The comments with regard to CD33 are spot-on because this protein is a target for an FDA-approved antibody-mediated chemotherapy.

    These types of studies are useful to the field, providing a wealth of publicly accessible information. For example, we recently combined data from a prior plasma proteome study with AD GWAS data to show that pQTLs for ITIM and ITAM-related proteins are over-represented as nominally significant AD risk factors. This analysis included the SIG14 locus as described by Yang et al., noting that SIG14 is deleted in a portion of the human population. See Shaw et al., 2021


    . Analysis of Genetic Variants Associated with Levels of Immune Modulating Proteins for Impact on Alzheimer's Disease Risk Reveal a Potential Role for SIGLEC14. Genes (Basel). 2021 Jun 30;12(7) PubMed.

  3. Genome-wide association studies (GWAS) were made possible by the completion of the Human Genome Project in 2003 and the International Haplotype Map Project, which sought to find genes and genetic variations that affect health and disease, in 2005. GWAS connect inherited genetic variations with higher prevalence of a particular human trait or illness, and enable better strategies to detect, treat, and prevent diseases. However, many genetic variants may not directly cause diseases, but instead reside in proximity to, and therefore co-occur in nonrandom association with, the disease-causing variant. Thus, it is sometimes difficult to mechanistically interpret genetic variations or utilize them as drug targets or as disease biomarkers.

    To assess the mechanistic and functional consequences of genetic variants, expression quantitative trait loci (eQTL) analysis, which links genetic variations to changes in expression levels of mRNAs, is being conducted. eQTL studies were made possible by the advent of gene-expression microarrays and then RNA-Seq, which query the whole transcriptome. However, the levels of proteins, the cellular and organism molecular workhorses that support most functions, are only partially predicted by the levels of their mRNA precursors, with concordance hovering at 40-50 percent.

    Harnessing a new aptamer-based technology to simultaneously query over 1,300 proteins, Yang et al. undertook protein quantitative trait loci (pQTL) analysis to directly determine how genetic variations impact protein expression in both healthy individuals and Alzheimer’s disease subjects. Yang et al. identified genetic variations that impact expression levels of 274 proteins in the cerebrospinal fluid, 127 proteins in the plasma, and 32 proteins in the parietal lobe cortex.

    Between 48.0 and 76.6 percent of the genetic variations leading to changes in protein expression (pQTL) could not be explained by effects on mRNA expression, since they did not co-localize with promoter/enhancer regions, splicing, DNA methylation, or histone acetylation. The pQTLs that are not associated with mRNA expression are likely affecting protein stability, cleavage, secretion, receptor binding, post-transcriptional modifications, or clearance. Consistent with this idea, the majority of the pQTLs, which were found to act either in cis or trans, are in coding regions. The cis acting pQTLs likely modify protein stability, cleavage sites, and post-transcriptional modifications sites, while trans acting pQTLs likely affect receptors, proteases, co-factors or subunits that interact with the investigated protein.

    Trans pQTLs, which are largely tissue-specific, enable additional drugging opportunities for non-druggable targets. For example, loss-of-function mutations in one of the two gene copies for the secreted protein progranulin (PGRN) cause frontotemporal dementia (FTD). Likewise, regulatory mutations that decrease PGRN expression by ~15 percent are associated with an increased risk to develop Alzheimer’s disease (AD), Parkinson’s disease (PD), and limbic-predominate age-related TDP-43 encephalopathy (LATE). In addition, such mutations contribute to a decreased age of onset, and faster disease progression, in amyotrophic lateral sclerosis (ALS).

    Previous pQTL-like analysis revealed that the level of progranulin in humans is determined by a single transmembrane receptor, sortilin, which brings progranulin to the lysosome for degradation. An antibody drug has been developed that functionally mimics genetic variants associated with decreased expression of sortilin in humans and genetic ablation of sortilin in mice. This antibody, designated AL001, pharmacologically blocks sortilin. As predicted by the human and mouse genetics, AL001 was shown to increase the level of PGRN two- to threefold in people with FTD, persistently restoring PGRN back to normal levels in these individuals. AL001 is now in Phase 3 clinical trials for FTD with PGRN mutations (Mar 2021 news). 

    Yang et al. found multiple genetic variations that increase the risk for AD, PD, FTD, ALS, and stroke that impact protein but not mRNA expression levels. Approximately 42 percent of the proteins whose expression levels changed in neurodegeneration are associated with corresponding genetic variation. The findings support the idea that changes in protein expression caused by familial genetic variations lead to increased disease risk. The findings further point to both novel drug targets and to the appropriate therapeutic intervention. It is possible that restoring the expression of these proteins or their associated activities back to physiological levels will have therapeutic benefit.

    A case in point is the Siglec 3/CD33 gene. The authors found that a haplotype at the CD33 locus, which is associated with increased plasma and CSF CD33 protein levels, is also known to be associated with increased AD risk. CD33 is an inhibitory immune check point receptor for microglia. Their findings validate the idea that overexpression of CD33 may suppress microglia function, thereby increasing AD risk. Therapeutically, the findings support the hypothesis that drugs which suppress CD33 expression or function could be effective AD therapies. Such a drug is currently in clinical trials for AD (see AL003).

    Yang et al. revealed multiple additional relations between GWAS and pQTLs. Specifically, the work discovered tissue-specific cis, trans, and pleiotropic regulators of protein expression. Since proteins that are regulated by the same genetic factors are likely part of the same signaling cascade, or are otherwise functionally linked, these findings may lead to the discovery of novel pathogenic pathways.

    Likewise, the work found that the associations of the genetic variants with protein levels are largely not disease- or age-specific, suggesting that both the genetic mechanism and potential therapies may be studied in a young, healthy population. Remarkably, the authors found high correlation between changes in protein expression in the CSF and blood, suggesting that in many cases, less-invasive blood tests could serve to monitor disease biomarkers. Finally, when the authors combined their data with an external dataset, increasing the total sample size by 25 percent, they noted a nearly twofold increase in the number of pQTL signals detected. This suggests that many more pQTLs are to be identified with larger sample set analysis.

    Overall, the manuscript by Yang et al. is a technical and biological tour de force that will enhance our genetic and mechanistic understanding of protein expression and human diseases.

    A co-author of Yang et al, Herve Rhinn, is an employee of Alector, which co-funded the study. 

    View all comments by Jenna Pappalardo

Make a Comment

To make a comment you must login or register.


News Citations

  1. PWAS x GWAS? Proteome Analysis Nets 10 New Alzheimer’s Genes
  2. Antibodies Against Microglial Receptors TREM2 and CD33 Head to Trials
  3. Deleting CD33 Benefits Mice—If Their Microglia Express TREM2
  4. Could CD33 Be the Microglial Target for Stimulating Phagocytosis?

Therapeutics Citations

  1. AL003

Paper Citations

  1. . CD33-Targeted Therapies: Beating the Disease or Beaten to Death?. J Clin Pharmacol. 2021 Jan;61(1):7-17. Epub 2020 Sep 1 PubMed.
  2. . Inhibition of iduronic acid biosynthesis by ebselen reduces GAG accumulation in MPS-I fibroblasts. Glycobiology. 2021 Jun 29; PubMed.
  3. . Safe and Sustained Expression of Human Iduronidase After Intrathecal Administration of Adeno-Associated Virus Serotype 9 in Infant Rhesus Monkeys. Hum Gene Ther. 2019 Aug;30(8):957-966. Epub 2019 Jun 10 PubMed.

External Citations

  1. summary statistics
  2. data

Further Reading


  1. . Genetic control of the human brain proteome. Am J Hum Genet. 2021 Mar 4;108(3):400-410. Epub 2021 Feb 10 PubMed.
  2. . Analysis of Genetic Variants Associated with Levels of Immune Modulating Proteins for Impact on Alzheimer's Disease Risk Reveal a Potential Role for SIGLEC14. Genes (Basel). 2021 Jun 30;12(7) PubMed.
  3. . Trans-pQTL study identifies immune crosstalk between Parkinson and Alzheimer loci. Neurol Genet. 2016 Aug;2(4):e90. Epub 2016 Jul 26 PubMed.

Primary Papers

  1. . Genomic atlas of the proteome from brain, CSF and plasma prioritizes proteins implicated in neurological disorders. Nat Neurosci. 2021 Sep;24(9):1302-1312. Epub 2021 Jul 8 PubMed.