Rounds of TPE diminish DNA damage and senescence of PBMCs
To investigate the effect of TPE, we examined blood samples before and after rounds of this clinical procedure (Fig. 1A and Supplementary Fig. 1). The samples were de-identified and used as per the approved IRB (see “Materials and methods”). Each sample was separated into plasma/serum and cells, as published . Our focus was on assaying the effects of TPE rounds on such hallmarks of aging, as DNA damage and cellular senescence. Samples 1, 2, 4, 6, 7, and 8 were from old individuals (77, 67, 72, 68, 60, and 72 years of age), while samples 3 and 5 were from middle-aged people (46 and 52 years of age) (Supplementary Table 1).
DNA damage can be caused by exogenous and endogenous sources, exemplified by X-rays, UV, and ROS [24, 25]. Accumulated DNA damage triggers genetic aberrations, senescence , and loss of cell function and leads to age-related diseases .
Oxidative DNA damage was assayed in PBMCs with the 8OH-dg kit (IT7974, G-Biosciences) (Fig. 1B). The relative level of 8-OHdG was high before TPE, and although there is a difference in the fold change among the samples, the DNA damage was significantly decreased by the rounds of TPE (Fig. 1B).
For assaying senescence, messenger RNA (mRNA) was isolated from the PBMCs and profiled with qRT-PCR for p16, using GAPDH as the control (Fig. 1C). Expression of the p16 senescence marker was high in the PBMCs before TPE and was reduced by subsequent rounds of TPE (Fig. 1C).
Tumor-related genes were also profiled by qRT-PCR on PBMCs. There was no significant change in the expression of c-Myc, a known oncogene, while the expression of tumor suppressor genes such as P21 and p53 increased after TPE treatment in most patients (Supplementary Fig. 2).
These results demonstrate that repetitive plasmapheresis reduces the markers of senescence and DNA damage in human PBMCs.
Rounds of TPE gradually restore youthful lymphoid/myeloid markers to old PBMCs
To determine the effects of repeated TPE on the age-imposed myeloid skewing, we performed qRT-PCR on the gene expression pattern that reflects the lymphoid cell fates and macrophage-linked inflammaging (Fig. 1D). Lymphoid and natural killer cells decrease with aging, while macrophages, and particularly inflammatory CD68 + macrophages, increase [11, 26,27,28], explaining the age-related deficiency in combatting viral infections and the tendency to easily develop hyper-inflammation .
As shown in Fig. 1D and E, rounds of TPE increased the CD3, CD4, and CD8 markers in PBMCs. For the B cell markers, there was a large sample-specific range, but CD19 and B220 were generally induced by the rounds of TPE (Fig. 1D, E). CD94, a NK cell–specific marker, was low in the old PBMCs, but rebounded after the rounds of TPE (Fig. 1D, E).
Further suggesting a rejuvenation of the leukocyte subsets, the expression of macrophage-specific markers, CD11b and CD68, was generally reduced by the rounds of TPE (Fig. 1D, E). Interestingly, these transitions were clearly longitudinal and gradual, becoming more statistically significant with more rounds of the procedure and stably improving overall for several months. And importantly, the positive effects of TPE were maintained for at least 1 month, the time from one round to another. In agreement with the increase in lymphoid gene markers and diminished inflammatory markers, the ratios of lymphoid to myeloid markers showed a sharp increase through the rounds of TPE, suggesting that this procedure re-balances adaptive immunity and that it diminishes the cellular signatures of inflammation (Fig. 1E). In addition to the net change in the lymphoid/myeloid markers, which is shown in Fig. 1D and E, the direct comparison between the R0 and Rlast rounds for each marker is shown in Supplementary Fig. 3.
These results demonstrate that repetitive TPE diminishes the pro-inflammatory leukocyte skewing and upregulates the lymphoid markers of old and middle-aged human PBMCs.
Stable and significant longitudinal rejuvenation of systemic proteome by TPE
Next, we profiled the serum samples from young and old control donors (young, 28–32 years of age; old 70–79 years of age, Supplementary Table 1) and the longitudinal samples from TPE patients, using RayBiotech antibody arrays (Fig. 2A and Supplementary Fig. 4A). Out of 507 proteins, 72 proteins were significantly different in their levels between the old and the young groups (> 1.75 fold change, p < 0.05). These 72 proteins were analyzed further in the longitudinal TPE datasets by heat mapping, which revealed a gradual rejuvenation of the age-specific systemic proteome by subsequent rounds of TPE (Fig. 2B and Supplementary Fig. 4B). PCA confirmed that each R0 proteome (before the first TPE) was closer to the old control group than to the young control group and shifted from the old toward the young group with the rounds of TPE (Fig. 2C).
The Gene Ontology (GO) analysis through DAVID (version 6.8, https://david.ncifcrf.gov) with the 72 selected proteins revealed that 226 terms were in biological processes (BPs), 29 terms were related to cellular components, and 40 groups belonged to molecular functions (Fig. 2 and Supplementary Fig. 5).
With respect to BPs, the top 30 GO terms are presented in Fig. 2D and Supplementary Fig. 5, and many of these participate in signal transduction (p = 1.50E − 17); regulate the inflammatory response (p = 4.10E − 14); positively regulate the ERK/ERK2 cascade (p = 5.50E − 14) and the cytokine-mediated signaling pathways (p = 4.00E − 13); positively regulate the cell proliferation (p = 5.20E − 10), chemokine-mediated signaling pathways (p = 5.40E − 09), immune response (p = 1.20E − 08), and chemotaxis (p = 5.00E − 06); and negatively regulate the apoptotic process (p = 3.60E − 07).
The inflammatory response is both an essential defensive system and a hallmark of age-related dysfunctions . We identified 19 proteins which were upregulated before the first TPE treatment or in the control old group, as compared to the young group (Fig. 2E). Notably, many of these proteins, including CCL20, CCL25, MIF, TLR3, TLR4, IL-2RA, and IL-16, were gradually downregulated by the rounds of TPE, with the expression levels measured just before the final round being close to the young levels (Fig. 2E).
Some difference in the levels of specific proteins was observed within the old cohort and between the old and TPE R0 cohorts, which could be explained, for example, by the differences in the ages, lifestyles, genetics, etc., parameters of the donors of commercial blood samples and those who were undergoing the TPE (Fig. 2E).
We also determined whether repetitive TPE may regulate the complement system including C3 and C1q, which play a key role in immune responses and also participate in non-immune crosstalks of cell–cell signaling pathways [30,31,32]. When we examined 15 commercial human serum samples across young (21–26), middle (46–52), and old (70–74) ages, we found that C1q and C3 levels were significantly elevated with age (Supplementary Fig. 6A). Mean levels of C3 and C1q complement proteins were significantly reduced by roughly half immediately following a TPE procedure, as expected from the immediate dilution of plasma, but on average, they returned to the initial levels by the next round ~ 1 month later. Even three TPE rounds had no significant lasting effects on the serum levels of C3 or C1q (Supplementary Fig. 6B and C).
Lastly, we analyzed the systemic regulators of apoptosis. Apoptosis plays an important role in immune responses and tissue homeostasis [33, 34]. However, with advancing age, resistance to apoptosis is increased through an enhanced negative loop of anti-apoptotic signaling, leading to senescence, inflammation, fibrosis, and a propensity for cancer [35,36,37]. In our BP analysis, 13 proteins were related to the negative regulation of apoptosis and the levels of these proteins were higher in the old group than in the young (Fig. 2F) [35,36,37]. Consistent with better tissue homeostasis, the levels of these apoptotic inhibitors diminished over rounds of TPE, becoming closer to the young cohort (Fig. 2F).
Therefore, rounds of TPE gradually reset the circulating protein markers of key cellular responses to their younger levels.
Pathway interaction String analysis defines the rejuvenation interactome
Although TPE treatment has been in clinic for decades, its impact on the process of aging and specifically the links to rejuvenation remain unstudied. To address this limitation, we first applied Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway bioinformatics analysis to the dataset of our comparative proteomics. Forty-three pathways were identified including cytokine–cytokine receptor interaction (p = 1.60E − 28), rheumatoid arthritis (p = 6.80E − 06), pathways in cancer (p = 9.30E − 06), intestinal immune network for IgA production (p = 3.50E − 05), JAK-STAT signaling pathway (p = 2.40E − 04), MAPK signaling pathway (p = 4.00E − 04), TGF-beta signaling pathway (p = 7.80E − 04), Toll-like receptor signaling pathway (p = 1.20E − 03), and nuclear factor (NF)-κB signaling pathway (p = 5.80E − 03) (Fig. 3A).
Interestingly, we also found KEGG terms related to cancer (Fig. 3B), e.g., a disease with a well-known age-elevated risk . There are many reasons as to why cancers occur more frequently with aging, including the decline in productive immune responses, inflammation, the overall aging of immune cells , and the accumulation of senescent cells . In agreement with the age-related etiology of cancers, most KEGG cancer terms, including pancreatic cancer, and proteoglycans in cancer were upregulated in the control old groups, as compared to the control young group (Fig. 3B). Notably, these excessive cancer pathway proteins were downregulated by the rounds of TPE (Fig. 3B).
Based on the broad multitissue rejuvenation by the old plasma dilution, we noted those pathways among the 43, which are known to be altered with aging in ways that interfere with the maintenance and repair of multiple tissues: the JAK-STAT , MAPK/ERK1/2 , TGF-beta , NF-κB , and Toll-like receptor signaling  (Fig. 3B and Supplementary Fig. 7).
To analyze the pathway interactions and the crosstalks between key proteins, we used a protein–protein interaction (PPI) network, the String  (Fig. 3C, D). The String consists of known and predicted PPIs including physical and functional associations from many data sources spanning more than 14,000 organisms [44, 45]. Moreover, String allows the display of various functional studies simultaneously with PPIs including GO, KEGG, and InterPro, facilitating the search for nodal proteins [44, 45]. With selected 72 proteins, PPI enrichment p values were less than 1.0E − 16, demonstrating that their interactions were significantly meaningful (Fig. 3C, D).
Through String analysis, we found that only one protein (TLR4) was linked to all the age-specific TPE-rejuvenated networks (Fig. 3C). Notably, the levels of TLR4 protein were significantly decreased after TPE, as compared to before this procedure (Supplementary Fig. 8). In addition, String pathway interaction analysis identified six proteins (TGFBR2, TGFA, FGF21, SMAD4, TGFBR1, and FZD3) that were at the intersection of the cancer-associated pathways and, importantly, which were restored to their younger crosstalks by the rounds of TPE (Fig. 3D).
To test and expand the significance of the findings that TLR4 is a nodal point of the age-altered and TPE-normalized pathway interactions, we analyzed the age-specific levels of TLR4 gene expression in 349 individuals (young [20–29 years] and aged [65–75 years]), using data mining of publicly available datasets [46,47,48,49,50,51,52,53,54].
TLR4 expression increases with age, in agreement with our proteomics and confirming the significance of the long-term attenuation of TLR4 by TPE (Fig. 3E).
These data demonstrate that rounds of TPE normalize the interactions between several key pathways, which, at their young signaling intensity, are known to be responsible for the homeostatic health of multiple organ systems, including the immune system. TLR4 was identified as a potential nodal point of the age-specific TPE-balanced signaling crosstalks.
Age and neurological disease–specific clustering of systemic proteomes and the reduction of a circulating biomarker of neurodegeneration, TDP43, by TPE
Aging is accompanied by various diseases with neurological disorders being a prominent class . The blood biomarkers of brain diseases remain unknown, but promisingly, the Uniform Manifold Approximation and Projection (UMAP) analysis of our comparative blood proteomics defined distinct clustering of old people with Alzheimer’s disease (AD) and AD-related diseases (ADRDs), as compared to the healthy age-matched and younger cohorts (Fig. 4A and Supplementary Table 1). In agreement with the data on rejuvenation of the systemic proteome which we described above, the rounds of TPE moved the old blood proteome cluster of this AD/ADRD dataset toward the younger age cohort (Fig. 4A). Further substantiating the robustness of our results, the healthy old proteome of the age-matched controls of the AD/ADRD samples clustered close to the pre-TPE proteome, an independent dataset of healthy old blood donors (Fig. 4A).
Considering these observations, we decided to analyze systemic levels of TDP43, which is the trigger of several neurological pathologies and becomes increased in the blood of patients with ALS, Parkinson’s disease (PD), frontotemporal dementia (FTD), and AD [56,57,58,59,60,61].
We analyzed the levels of this protein in young, old, pre-TPE, and after TPE serum samples. Interestingly, a TDP43-specific enzyme-linked immunosorbent assay (ELISA) demonstrated a robust age-specific increase in this determinant of neurological diseases in the serum of old, relatively healthy adults, as compared to the young cohort (Fig. 4B). Moreover, the longitudinal studies on the plasma before vs. after the rounds of TPE demonstrated that systemic TDP43 was stably attenuated by plasmapheresis (Fig. 4C). Notably, the overall levels of TDP43 in older adults were not just transiently diluted by the procedure, which is expected, but remained lower for at least a month after TPE (Fig. 4C).
These results suggest that the selected proteins enable the identification of AD/ADRD through UMAP analysis of blood plasma and confirm that TPE promotes a younger systemic proteome. Additionally, we show that TPE stably attenuates a known biomarker of neurological diseases, TDP43.
A novel biomarker of protein level noise provides a direct molecular measurement of human biological age and shows that it is reduced by TPE
Biological noise is the variation of a given biological parameter in an otherwise assumed constant or steady state. Several papers suggest that biological noise increases with age [62, 63], and noise of gene expression and protein levels is an independent age-related parameter that is relatively less studied, as compared with total levels of the genes, proteins, and other cellular and tissue metrics. An unmet need remains in identifying reliable molecular biomarkers that allow measuring biological age, in contrast to predicting it by data selection and adjustment, such as machine learning models of population statistics. We approached this problem by comparing not just young vs. old samples, as typically done, but also the rejuvenated cohort, and moreover, in the longitudinal studies of the same older individuals whose blood compartments became more like those of young people after TPE, based on the cellular and humoral blood assays.
To test if not just the protein levels but also their noise became more youthful through TPE, we compared the standard deviations (SDs) of protein levels, using our comparative proteomics datasets: the control young and old cohorts, round 0 vs. the last round of TPE, and of the old cohort with the cognitive disorders (progressive AD and ADRD). As the measurement uncertainty is similar between populations, any change in SD should reflect biological variation; here, biological variation is the levels of protein expression. A change in this variation could mean many things, from a loss of transcription regulation to stochasticity in protein quality checking to changes in turnover, but the end result is that protein levels are either more heavily controlled for lower SD or less heavily controlled for a greater SD.
To find statistically significant changes in biological noise, we looked at the difference in variance with Levene’s test. Levene’s test has various possible shortcomings in both power and significance when dealing with small sample sizes and asymmetric distributions . To deal with this, we increased significance through the Benjamini–Hochberg procedure, selecting only proteins with significant changes in multiple populations and selecting proteins with other measures of substantial change in variance, albeit only larger sample size ultimately increases power. We found only 6 proteins showing increased SD in the young-to-old transition and decreased SD in the pre-to-after TPE transition: TRAIL R1, IL-16, TIMP-1, IL-15R alpha, CD27, and APJ (Fig. 5A). Interestingly, when we considered proteins which were significantly different in their SD between the healthy old and the old with cognitive disease and between the healthy young and healthy old, there were just 4 such proteins: Smad5, uPAR, FADD, and TGFBR1 (Fig. 5B). Namely for these 4 proteins, the SD increased in a young-to-old transition and in an old-to-disease transition but decreased in the R0-to-Rlast TPE transition (Fig. 5B).
We next profiled the proteins which had not just a statistically significant change in the young-to-old transition, but also had a substantial change (greater than fivefold) in SD with TPE. We found 5 such proteins: APJ, TNFRSF27, uPAR, CCL25, and TGFBR2 (Fig. 5C). Very interestingly, when the field was further narrowed to only those proteins which also have significant SD changes in the healthy-to-disease transition, there was only one protein showing increased SD in young-to-old transition and old-to-disease transition, but decreased SD in pre-to-post-TPE transition, uPAR (Fig. 5D). And for this best noise detector protein, the middle-aged individuals lay right between the young and old individuals (Fig. 5D).
To confirm and extrapolate the significance of the identified protein determinants of age-related biological noise, their mRNA expression was analyzed in 349 individuals from publicly available datasets [46,47,48,49,50,51,52,53,54]. Interestingly and in agreement with our comparative proteomics data, the mean SD values, e.g., the noise, of all these genes increased with age, albeit to a different degree for the different transcripts (Fig. 5E).
Their mRNA levels were also age-elevated (Fig. 3E [for TLR4] and Supplementary Fig. 9 [for other transcripts]). These results agree with the proteomics data and expand the significance of TLR4 as a nodal point of the age-altered, signal transduction networks (Fig. 3B, C).
Supplementary Fig. 10 arranges the data by the size of SD of protein levels in a healthy population, which gives further credence to the idea that the changes in SD are the consequence of the age/disease-imposed biological noise. Namely, the proteins that are most tightly controlled initially were also those most likely to have a significant change in SD (deregulation) with age and disease. Supplementary Fig. 10 also shows that TPE is indeed returning the SDs to their younger lower states, e.g., the noise dampening effects of TPE are manifested more on the proteins that become less controlled or noisier with age. Of note, there is no uniform change in variance in noise with age; instead, those proteins which are least controlled change most and vice versa.
Our next goal was to apply the discovered biomarkers of proteome noise toward a novel measurement of person’s biological age. There is an unmet need for unambiguous molecular quantification of biological age based on experimentally defined biomarkers, without machine learning (ML) methods, large data statistics, and data adjustments and with improved resilience to batch effects [65, 66]. Typical ML models are trained on chronological age and predict chronological age. The connection of a correlate (DNA methylation, protein levels, etc.) to biology is not straightforward, and in fact, the experimental data is numerically adjusted (variably for different data points) to correlate with the chronological age better linearly. Large datasets are used, but each line has N = 1 with a lack of verification.
The age-increased and TPE-reduced changes in the SD of the 10 protein biomarkers (Fig. 5F) suggested the possibility to uncover a natural linearity between the chronological age and the biological age, through these noise reporters. Although the mean protein levels tend to increase with age, variance also often increases. Namely, in some old samples, the levels will be similar to the young samples, and in other old samples, they will be significantly different, which is reflected in high SD. To test if the 10 protein biomarkers would enable a direct quantification of person’s biological age, we calculated the mean of their SDs (10-protein noise) and plotted these values vs. chronological age of the young (20–30 years), middle-aged (50–60), and old (70 +) individuals (Fig. 5G). Such direct analysis yielded a linear measure of biological age, which we define as an increase in the 10-protein noise (SD) with a Pearson’s R value of 0.7269 and a p value of < 0.0001 (Fig. 5G).
Next, we tested if TPE reduces the directly quantified biological age and if brain dysfunction (mild cognitive impairment, MCI) increases the biological age of chronologically similar people. As shown in Fig. 5H, based on the 10-protein noise, the biological age of people with MCI was increased by over 50 years (the average of biological age = 130.9 years), while the biological age of 40–70 + individuals became reduced by decades after multiple rounds of TPE (Fig. 5I). The additional analysis of the relative biological noise age via TPE treatment demonstrated that it becomes reduced more sharply in individuals aged 60–70 + years, as compared to people who are 46–52 years old (Supplementary Fig. 11).
These results establish that person’s biological age can be not just predicted from large datasets, but directly and accurately measured by the combined 10-protien noise; these direct quantifications demonstrate a natural linearity of the 10-protein noise with chronological age, a reduction of such biological age by three rounds of TPE, and an increase by MCI. These results additionally suggest that age and neurologic disease increase the noise of the systemic proteome, most noticeably when looking at uPAR, TRAIL R1, IL-16, TIMP-1, IL-15R alpha, CD27, and APJ, TNFRSF27, CCL25, and TGFBR2. These conclusions are derived from comparative proteomics on over 50 samples and are confirmed through the gene expression analyses of 349 individuals.