. Human Herpesvirus 6 Detection in Alzheimer's Disease Cases and Controls across Multiple Cohorts. Neuron. 2020 Mar 18;105(6):1027-1035.e2. Epub 2020 Jan 23 PubMed.


Please login to recommend the paper.


  1. As compared to the Readhead paper, which with sophisticated bioinformatics tools proved a functional association with AD and its viral risk factors, here Jacobson and colleagues use a single bioinformatic pipeline, utilizing the same data set and same brain regions as Readhead to match viral reads in RNAseq data set to clinical outcome. This is salient and raises the need for methodological harmonization and specific guidelines for how pathogen-host interactions can be investigated in the setting of the more immuno-privileged brain. It is surprising that in the bioinformatic analysis of Allnut and colleagues, HSV-1 was not considered, or simple reference taxon was not present. Overall, a more comprehensive comparative analysis between the two methods would have been desirable, also in the discussion part of the manuscript. Nevertheless, the authors might provide a more detailed comparison.

    The wet-lab ddPCR findings clearly reveal that the prevalence of HHV6A and HHV6B is either very low or the viruses not present in the dorsolateral prefrontal cortex of the cohort, independent of the clinical diagnosis. The authors, however, do not rule out that viral DNA could be still detectable in other regions of the brain, likely areas of the olfactory circuitry, where we could find traces of HHV6A and HHV6B antigens. The authors invoke the microbiocidal activity of Aβ, discovered by Moir and colleagues, to suggest it may dampen the spread of HHV6 and limit the ability to detect the virus in this region after its initial invasion. In support of this, our analysis of the olfactory cortices showed HHV6 antigens associated either with Aβ plaques or neurofibrillary tangles.

    That said, the ddPCR method used by Allnut and colleagues was sensitive enough to detect chromosomally integrated HHV-6 in five out of 708 patients, which is comparable to the rate of HHV6 infection and integration in the human population. This is another profound discrepancy that is worth noting between the Readhead and the present paper. Redhead, by using whole-exome sequencing, reported an increased correlation between ciHHV6 and AD, which was not reproduced in a wet-lab setting.

    I anticipate that in the future metagenomic techniques such as virome sequencing (Hannigan et al., 2015) may be used on the same samples and additional brain regions to provide additional evidence about the association of Herpes viruses with AD.

    Overall, the present paper emphasizes how alternative hypotheses for the AD etiology need to be investigated using both bioinformatics but also wet-lab tools and the absolute need to run both experimental paradigms in parallel. The renewed interest in this field, at least in the U.S. and China, will allow more comprehensive and large-scale studies addressing the vector-host interactions, particularly in light of the several immunological genetic risk factors for AD (APOE, HLA, CLU, TREM2, etc.).


    . The human skin double-stranded DNA virome: topographical and temporal diversity, genetic enrichment, and dynamic associations with the host microbiome. mBio. 2015 Oct 20;6(5):e01578-15. PubMed.

    View all comments by Lavinia Alberi
  2. This work by Allnut et al. complements our analysis on whether HHV6 and 7 are elevated in AD brains compared to control. Our analysis is focused on the validity of the method used to test whether there is a statistical significance to the HHV virus level estimated from the RNAseq data. Allnut et al. took a different approach and looked at whether the quantification of virus expression level is accurate. Using Pathseq, a bioinformatics tool to quantify pathogen expression level, Allnut found no difference between AD and control. They further validated their finding using ddPCR.

    Our analysis demonstrated that the statistical test influenced the positive results in Readhead et al., but we didn't find the root cause of the problem. Allnut et al. moved one step further and discovered that the quantification method in the original publication is the cause of the false positive. Taking both analyses together, we can conclude that HHVs do not contribute significantly to the etiology of AD. The only remaining factor is whether the ROSMAP and MSSB data set represent the diverse AD population. 

    View all comments by Zhandong Liu
  3. The study by Allnutt et al. reports the application of a computational approach for the detection of microbial sequences to two existing RNA-sequence data sets (MSBB and ROSMAP) to examine potential microbial associations with Alzheimer’s disease. The authors also present results from digital droplet Polymerase Chain Reaction (ddPCR)-based detection of HHV-6A and HHV-6B DNA within stored DNA samples from the ROSMAP and JHBRC cohorts. Overall, Allnutt et al. report extremely low detection frequencies of any viruses within the RNA-sequence data, and for HHV-6A/B within the ddPCR samples in particular. The authors conclude that this does not support an association between HHV-6 and Alzheimer’s disease. Although this finding is discordant with our previously published findings (Readhead et al., 2018), we are encouraged to see the increased scientific attention that is being afforded to this relatively underexplored area.

    We respectfully note that the detection rates for HHV-6A/B reported by Allnutt et al. are approximately an order of magnitude lower than might be expected from previous investigations of HHV-6A/B prevalence within brain tissue. Without a better understanding of this discrepancy, it is difficult to evaluate the impact of claims made about associations with clinical populations. In the analysis of the RNA-sequence data sets, the authors observed HHV-6A/B sequences in 0 percent of the MSBB control samples, and 0.63 percent of the ROSMAP control samples. Among the 166 normal control subjects that underwent ddPCR, the authors observed HHV-6A DNA in 0.6 percent of samples, and HHV-6B DNA in 3 percent of samples. Previous reports of HHV-6A/B frequencies within control populations include 40 percent of temporal and frontal cortex samples (Lin et al., 2002), 26.8 percent of white-matter samples (Cermelli et al., 2003), and 10 percent of temporal cortex samples (Wipfler et al., 2018). Several studies looking across multiple brain and CNS sites have reported rates of 24.3 percent (Chan et al., 2001), 68.6 percent (Chapenko et al., 2016), and 88.2 percent (Hemling et al., 2003). In our own study (Readhead et al., 2018), observed frequencies varied across data sets and brain regions, but across the four brain regions RNA-sequenced in the MSBB, we observed HHV-6A frequencies ranging from 18–35 percent, and HHV-6B frequencies ranging from 15–31 percent. Within the ROSMAP RNA-sequence data, we observed HHV-6A frequency of 41 percent and HHV-6B frequency of 44 percent. Further work is needed to understand the variability of prevalence observed for HHV-6A/B within the brain. In addition to technical issues around assay detection sensitivity, whether a virus is uniformly present throughout infected tissue or distributed focally will inform under what sampling conditions, and what quantities of input material are needed to reliably detect its presence.

    Allnutt et al. also utilized a metagenomics classifier called PathSeq (Kostic et al., 2011), and cite the recent application of this tool to detect JC virus and dengue virus in two cases of viral encephalitis (Johnson et al., 2019; Reoma et al., 2019). While these examples are clinically striking, they are characterized by productive, fulminant viral infection (with associated PathSeq scores > 5,000), and thus importantly different from the modes of activity that species such as HHV-6A/B demonstrate in a non-encephalitic brain, where the virus is likely to be in a latent or quiescent state and present at very low abundance. The maximum HHV-6A/B PathSeq score reported by Allnutt et al.  was 33.5, despite the presence of several chromosomally integrated HHV-6 samples (which contain a viral genome within every nucleated cell). This can be contrasted with the more frequent occurrence of an exogenous HHV-6A/B infection, where as few as one in 10,000 cells may be infected. This suggests that although this computational approach can identify acute, clinically relevant infections, it may be less well suited to the detection of lower abundance, indolent infections. This is also consistent with the absence of detection of other very common, neurotropic viruses within the analysis of the RNA-sequence data sets. For example, in the analysis of 901 RNA-seq samples, the authors did not observe any instances of Herpes-simplex virus-1 (HSV-1), HSV-2, HHV-7, or Kaposi’s sarcoma-associated herpesvirus (KSHV) infection.

    We agree with the authors’ conclusion that although their findings do not support an association of HHV-6A/B with Alzheimer’s disease, neither do they rule it out. Taken together, we would suggest that without a clearer understanding of how to reconcile the unusually low frequencies of detected HHV-6A/B with a diverse scientific literature that supports a much higher prevalence, it is difficult to properly evaluate potential associations with Alzheimer’s disease. Although our previous study (Readhead et al., 2018) detected HHV-6A/B at much higher rates than Allnutt et al. (though in accordance with previously published estimates), we expect that mere detection of a particular virus in a single dimension (e.g., RNA or DNA sequences) will convey only limited understanding of how viruses might contribute to the complex pathobiology of Alzheimer’s disease. This influenced our adoption of diverse, integrative analyses throughout our study (Readhead et al., 2018), which were aimed at reducing reliance on any single observation, and building a broad perspective of the multiscale pathology of Alzheimer’s disease. We expect that an expansion of viral detection assays to include searches for viral proteins and noncoding RNAs will contribute valuable additional context to these investigations. Despite differences in findings between Allnutt et al. and our own study (Readhead et al., 2018), we are energized by the surge of investigations into potential roles for microbes in Alzheimer’s disease, and are excited about their potential to help advance our understanding of this devastating disease.


    . Multiscale Analysis of Independent Alzheimer's Cohorts Finds Disruption of Molecular, Genetic, and Clinical Networks by Human Herpesvirus. Neuron. 2018 Jul 11;99(1):64-82.e7. Epub 2018 Jun 21 PubMed.

    . Herpesviruses in brain and Alzheimer's disease. J Pathol. 2002 Jul;197(3):395-402. PubMed.

    . High frequency of human herpesvirus 6 DNA in multiple sclerosis plaques isolated by laser microdissection. J Infect Dis. 2003 May 1;187(9):1377-87. Epub 2003 Apr 9 PubMed.

    . The Viral Hypothesis of Mesial Temporal Lobe Epilepsy - Is Human Herpes Virus-6 the Missing Link? A systematic review and meta-analysis. Seizure. 2018 Jan;54:33-40. Epub 2017 Nov 24 PubMed.

    . Prevalence and distribution of human herpesvirus 6 variants A and B in adult human brain. J Med Virol. 2001 May;64(1):42-6. PubMed.

    . Detection frequency of human herpesviruses-6A, -6B, and -7 genomic sequences in central nervous system DNA samples from post-mortem individuals with unspecified encephalopathy. J Neurovirol. 2016 Aug;22(4):488-97. Epub 2016 Jan 4 PubMed.

    . Herpesviruses in brains in Alzheimer's and Parkinson's diseases. Ann Neurol. 2003 Aug;54(2):267-71. PubMed.

    . PathSeq: software to identify or discover microbes by deep sequencing of human tissue. Nat Biotechnol. 2011 May;29(5):393-6. PubMed.

    . Chronic Dengue Virus Panencephalitis in a Patient with Progressive Dementia with Extrapyramidal Features. Ann Neurol. 2019 Nov;86(5):695-703. Epub 2019 Sep 11 PubMed.

    . Fatal encephalopathy with wild-type JC virus and ruxolitinib therapy. Ann Neurol. 2019 Dec;86(6):878-884. Epub 2019 Oct 16 PubMed.

    View all comments by Joel Dudley
  4. The value of a method seeking to detect a specific pathogen depends on its sensitivity and on the usage of checks such as exclusion of false negatives and positives, e.g., by recovery experiments. In the case of PCR, the sensitivity can depend on the target gene, the primers, the source of DNA polymerase, etc. Checks for false positives were mentioned in this study by Allnutt et al., but it would be very informative to know if checks for false negatives were used also—as their findings were almost wholly negative.

    Other useful information would include details of the preparation of the DNA they obtained for the study, because in certain procedures there is a high risk of loss of some DNA species if present at very low levels. Such details would be particularly important in view of the fact that the amounts of HHV6 and of HSV1 in brain are likely to be very low—otherwise, the subjects would show symptoms of encephalitis or meningitis. Nonetheless, in the case of HHV6, a number of studies, some of which are by Jacobson and colleagues, have revealed the presence of viral DNA and/or RNA in brains, including the olfactory bulb/tract—indicating the probable pathway to the brain (Lin et al., 2002; Opsahl and Kennedy, 2005; Yao et al., 2010; Harberts et al. 2011; Lin, 2016). A discussion of the conflict between their current and the previous data would be invaluable.

    As for viral RNA, a high sensitivity when seeking transcripts of these viruses would be especially necessary because transcription of HSV1 and HHV6 DNA during latency is very limited, and probably in most brains the viruses are latent most of the time, with reactivated virus present in only relatively few individuals on any one occasion.

    Surprisingly, the apparent absence of HSV1 transcripts is not discussed, nor, for that matter, the discrepancies with the results of Readhead et al., who detected mainly HSV1, HHV6, and HHV7 in brain. In fact HSV1 DNA presence in normal elderly brains as well as in AD brains has been well established: Data from over 200 studies, using a great variety of methods, are consistent with viral presence, virus activity, variable state of latency or reactivation, and are consistent with the concept that HSV1 in brains of APOE-e4 carriers confers a high risk of AD. Further, the finding that APOE-e4 is a risk for cold sores in the PNS—which are caused usually by HSV1—is consistent with the concept that HSV1 in the brains of APOE-e4 carriers is particularly damaging in the nervous system.  

    That Allnutt et al. conclude from the similarity of their values of frequency/prevalence, or amount/level/abundance, of a specific virus in ADs and controls, that HHV6 is probably not involved in AD, indicates that a general point needs to be made: The authors and many others seem unaware that "controls" can be infected but asymptomatic, i.e., infected does not necessarily mean affected. Host genes or other factors might well determine an individual's response to an agent. For example, at least 80 percent of people are infected with HSV1 but only about one-fifth of them suffer from cold sores: the other four-fifths, being asymptomatic, would be considered controls. Equally important, and equally unappreciated, the possibility that AD patients are more likely to be infected with HSV1 than controls is often stated—but is vitiated by the fact that the prevalence of HSV1 in brain is only slightly lower in controls than in the patients (Itzhaki et al., 1997). 


    . Herpesviruses in brain and Alzheimer's disease. J Pathol. 2002 Jul;197(3):395-402. PubMed.

    . Early and late HHV-6 gene transcripts in multiple sclerosis lesions and normal appearing white matter. Brain. 2005 Mar;128(Pt 3):516-27. Epub 2005 Jan 19 PubMed.

    . Review part 2: Human herpesvirus-6 in central nervous system diseases. J Med Virol. 2010 Oct;82(10):1669-78. PubMed.

    . Human herpesvirus-6 entry into the central nervous system through the olfactory pathway. Proc Natl Acad Sci U S A. 2011 Aug 16;108(33):13734-9. PubMed.

    . Human herpesvirus multiplex ddPCR detection in brain tissue from low- and high-grade astrocytoma cases and controls. Infect Agent Cancer. 2016;11:32. Epub 2016 Jul 26 PubMed.

    . Multiscale Analysis of Independent Alzheimer's Cohorts Finds Disruption of Molecular, Genetic, and Clinical Networks by Human Herpesvirus. Neuron. 2018 Jul 11;99(1):64-82.e7. Epub 2018 Jun 21 PubMed.

    . Herpes simplex virus type 1 in brain and risk of Alzheimer's disease. Lancet. 1997 Jan 25;349(9047):241-4. PubMed.

    View all comments by Ruth Itzhaki
  5. In one of the largest studies on the topic to date, Allnutt et al. show no association between HHV-6A/B and Alzheimer’s disease in postmortem brain samples from three independent cohorts using two methods: PCR detection of DNA across 711 AD and non-AD brains from three cohorts, and RNA-Seq data for 901 individuals from two of the cohorts. Overall the frequency of HHV-6A/B detection was very low across cases and controls. These findings contradict those of Readhead et al., who demonstrated a frequency of HHV-6A/B that was roughly an order of magnitude higher, as well as a higher prevalence among AD brains compared to controls (Readhead et al., 2018). Other earlier studies of HHV-6 in postmortem brains show similar discrepancies, both in terms of prevalence and findings (Hemling et al., 2003Lin et al., 2002; Lin et al, 2002). 

    So, what does this mean? Despite differences in their findings, the above studies highlight similar methodological challenges. Case-control studies using postmortem brains are not an optimal study design to assess risk factors for dementia, which has a long, slow trajectory occurring over decades. Hampered by potential reverse causation (which came first—virus activity in the brain or dementia?) and lack of control for potential confounding factors, such studies cannot assess the effect of immune responses to HHV-6 or other viruses on the initiation or acceleration of neurodegenerative processes that occur earlier in life.

    While mechanistic studies have identified links between HHV-6 virus infection and amyloid-β fibrillization in human neural cell cultures and mouse models (Eimer et al., 2018), or between viral abundance and modulators of amyloid precursor protein metabolism (Readhead et al., 2018), it is unclear how such relationships play out in vivo. Are these effects clinically relevant? If so, which populations are affected, for example by age, dementia risk factors, comorbidities, and geographic locations? High-quality population studies with longitudinal designs are needed to assess the effect of viruses on long-term outcomes such as dementia. These should collect valid repeated measures of herpesvirus activity as well as sociodemographic and other risk-factor data, in addition to robust outcome measures. Given the current lack of information about who would constitute an appropriate trial population, calls for trials of antiviral agents for dementia prevention seem premature.


    . Multiscale Analysis of Independent Alzheimer's Cohorts Finds Disruption of Molecular, Genetic, and Clinical Networks by Human Herpesvirus. Neuron. 2018 Jul 11;99(1):64-82.e7. Epub 2018 Jun 21 PubMed.

    . Herpesviruses in brains in Alzheimer's and Parkinson's diseases. Ann Neurol. 2003 Aug;54(2):267-71. PubMed.

    . Herpesviruses in brain and Alzheimer's disease. J Pathol. 2002 Jul;197(3):395-402. PubMed.

    . Cytomegalovirus is present in a very high proportion of brains from vascular dementia patients. Neurobiol Dis. 2002 Feb;9(1):82-7. PubMed.

    . Alzheimer's Disease-Associated β-Amyloid Is Rapidly Seeded by Herpesviridae to Protect against Brain Infection. Neuron. 2018 Jul 11;99(1):56-63.e3. PubMed.

    View all comments by Charlotte Warren-Gash
  6. I examined many AIDS, "normal," MS, and PML brain sections, as well as a few glioblastoma, Parkinson's, and AD brain sections, all by in situ PCR for HIV-1 and HHV-6B. Whereas HIV-1 was exclusively in AIDS brains, ranging from a few copies to hundreds of copies per cell depending on the case, HHV-6 was prevalent only in MS and PML brains, ranging also from several to hundreds of copies per cell. In other diseases, including AD brains and "normals," the HHV-6 count was usually on the order of one positive cell in 20. So there is a background signal of ~1/20 cells in all cases except late-term fetuses, where the HHV-6 background was essentially zero. These results would not suggest a prominent association between HHV-6 and AD, but we only had two AD sections and could easily have missed an association if there was one.

Make a Comment

To make a comment you must login or register.

This paper appears in the following:


  1. Herpes Viruses and Alzheimer’s—The Debate Continues