Towards a personalized approach in senolytic trials

Abstract

Attempts to translate senolytics from preclinical models to humans are gaining momentum. Early clinical trials have provided positive biological signals, but we lack clear evidence for the efficacy of senolytics in humans. Based on what we have learned in these initial trials, it may be time to aim for a more personalized approach in designing future senolytic trials, and potentially also in the eventual clinical use of these compounds.

Given the extensive evidence in preclinical models of the beneficial effects of clearing senescent cells on multiple age-related comorbidities, there is now intense interest in and a proliferation of clinical trials testing senolytic compounds for a variety of conditions in humans. As summarized here, early results from these studies have not uncovered any safety concerns and have identified some signals for biological efficacy. That said, this is probably an opportune time to assess what we have learned thus far from the clinical trials that have been completed and begin to perhaps redesign future trials on the basis of lessons learned to date.

Overview of clinical trials of senolytics

Table 1 summarizes the findings of published clinical trials of senolytics. For systemic treatment, only dasatinib plus quercetin (D + Q) has been used; the majority of trials have been single arm and involved small numbers of participants, and focused mainly on feasibility and safety. Although some of the uncontrolled trials have shown biological signals of efficacy, the clinical implications of these findings in the absence of a control group have been difficult to assess. Only two randomized controlled trials of systemic senolytic treatment have been published to date: one was in patients with idiopathic pulmonary fibrosis and involved six treated and six control participants, and no effect of D + Q was noted on measures of frailty, or pulmonary or physical function¹. The largest randomized controlled trial of systemic treatment was published by our group², in which 60 postmenopausal women were randomized to either a control group or treatment with D + Q, which was given every 4 weeks over 20 weeks. In this study, there was no effect of D + Q on a marker of bone resorption (serum cross-linked C-telopeptide of type I collagen), but a positive signal for an increase at 2 and 4 weeks in the bone formation marker procollagen type 1 N-propeptide (P1NP) was detected. In addition, exploratory analyses from this study, along with findings from an uncontrolled trial by Millar et al.³, have provided potential insights into the design of future trials of senolytics.

Table 1 |.

Summary of clinical trials of senolytics to date

Referencea	Purpose	Study participants; type of trial; and number of participants	Key findings
Systemic treatment
Justice et al.11	Safety and tolerability of D + Q in patients with IPF	Patients with IPF; open label, no control group; n = 14	Supported feasibility, possible improvement in physical function measures over 3 weeks
Hickson et al.12; Saul et al.13	Biomarker response to D +Q	Patients with diabetic kidney disease; open label, no control group; n = 9	Reduction in biomarkers of senescence, including the SenMayo gene set13, in adipose tissue biopsies
Nambiar et al.1	Safety and tolerability of D + Q, and inform future study feasibility	Patients with IPF; randomized, control trial; n = 6 per group	D + Q was well tolerated, no differences between groups in measures of frailty, or pulmonary or physical function
Gonzales et al.14; Garbarino et al.15	Assess central nervous system penetrance, safety, feasibility and efficacy in patients with early-stage Alzheimer’s disease	Patients with early-stage Alzheimer’s disease; open label, no control group; n = 5	CNS penetrance of D (but not Q) was observed, with outcomes supporting safety, tolerability and feasibility in patients with Alzheimer’s disease; changes in plasma and PBMC biomarkers noted
Farr et al.2; Farr et al.8	Assess efficacy of D + Q on markers of bone metabolism in aging women	Postmenopausal women; randomized controlled trial; n = 30 per group	D + Q did not reduce markers of bone resorption but did increase markers of bone formation; exploratory analyses indicated that women with the highest senescent cell burden had greater increases in bone formation
Millar et al.3	Safety, feasibility, and preliminary effects of D + Q in older adults at risk for Alzheimer’s disease	Patients with mild cognitive impairment; open label, no control group; n = 12	No safety signal with D + Q, trends for improvements in some cognitive scores and gait measures; exploratory analyses indicated that patients with the highest senescent cell burden had greater improvements in gait measures
Local treatment
Lane et al.6	Safety and efficacy of single-dose, intra-articular injection of UBX0101 (p53–MDM2 interaction inhibitor) in patients with knee osteoarthritis	Patients with painful knee osteoarthritis; randomized controlled trial; n = 183 patients randomized 1:1:1:1 to placebo or 3 different doses of UBX0101	Changes in knee pain scores no different in placebo compared with UBX0101 groups, treatment well tolerated
Crespo-Garcia et al.4	Ascending dose safety study of UBX1325 (a BCL-xL inhibitor) in patients with advanced diabetic macular edema	Patients with advanced diabetic macular edema; open label, no control group; n = 4 low dose, n = 4 high dose	UBX1325 was well tolerated; improvements in measures of visual acuity noted
Klier et al.5	Tested ability of UBX1325 to mitigate effect of diabetic macular edema on visual acuity	Patients with diabetic macular edema; randomized controlled trial; n = 65 UBX1325, n = 33 sham	No serious adverse events, trends suggestive of potential efficacy

IPF, idiopathic pulmonary fibrosis; PBMC, peripheral blood mononuclear cells.

^aWhen multiple publications analyze data from the same trial, they are listed together.

Although D + Q has generally been well tolerated when administered systemically, there are concerns about possible toxicities of systemic administration of other senolytic drugs, such as the BCL-xL inhibitor, UBX1325. As such, this senolytic was administered locally into the eye in patients with diabetic macular edema in two studies. In the first, Crespo-Garcia et al.⁴ gave a single injection of UBX1325 to 8 patients with diabetic macular edema and noted improvements over 12 and 24 weeks in some measures of visual acuity. The second⁵, larger trial was a randomized controlled trial that was also conducted in patients with diabetic macular edema, in which participants received a single injection of UBX1325 (n = 65) or sham injection (n = 33). At 48 weeks after treatment, the difference between UBX1325 and sham in mean change to week 48 in best corrected visual acuity was 5.6 more Early Treatment of Diabetic Retinopathy Study (ETDRS) letters (95% confidence interval, −1.5 to 12.7), which is suggestive of potential efficacy and the need for larger trials to further evaluate these findings. Although as yet unpublished (and not included in Table 1), there has been a press release from Unity Biotechnology that describes results from a larger randomized controlled trial — also in diabetic macular edema — that compared UBX1325 to a standard-of-care, anti-VEGF treatment (aflibercept), and demonstrated non-inferiority of UBX1325 to aflibercept at 36 weeks, but not at the average of weeks 20 and 24, which was the prespecified primary end point. In addition to UBX1325, a p53–MDM2 interaction inhibitor (UBX0101) has been tested in a randomized controlled trial for efficacy in painful knee osteoarthritis using local injection, and there were no differences observed in measures of pain severity between the sham and treated groups⁶.

Lessons learned from existing clinical trials

As is evident, the limited data from the randomized controlled trials of either systemic or local senolytic treatment have not uncovered any safety concerns and have provided some signals for biological efficacy, but clear evidence for the efficacy of senolytic interventions in humans is still lacking. As such, it is important at this point to consider the design and conduct of future trials of senolytics, where we can perhaps take stock of the data to date — specifically, what we have learned so far, and what the future directions should be for senolytic trials.

A potential direction comes from exploratory analyses from the studies performed by our group², with some supporting evidence from Millar et al.³. In these analyses, we found in our trialof D + Q in postmenopausal women that although the treated group overall demonstrated an increase in the bone formation marker P1NP at 2 and 4 weeks, this increase was modest relative to the control group² (approximately 16% difference between groups, Fig. 1). However, a prespecified hypothesis in that study was that the baseline senescent cell burden would determine the clinical response to the senolytic intervention. To assess senescent cell burden, we used a modification of an assay that was previously developed and validated in the Sharpless laboratory: assessment of T cell expression levels of p16 (also known as CDKN2A) mRNA⁷. Using this assay, we found that the increase in P1NP levels in the overall group of women treated with D + Q was driven largely by the women in the highest tertile (T3) for T cell expression of p16 mRNA² (Fig. 1). In addition, because the T cell p16 assay is technically challenging, we also measured a panel of 36 senescence-associated secretory phenotype (SASP) factors² and identified a panel of six circulating SASP factors (sclerostin, MMP2, FAS, PARC (also known as CCL18), osteoactivin (also known as glycoprotein non-metastatic melanoma protein B (GPNMB)) and TNFR1) that were higher at baseline in the women in the T3 group for T cell expression of p16 as compared with those in the lower two tertiles (T1 and T2)⁸. We then developed a SASP score, which was simply the geometric mean of these six SASP factors for each participant and tested whether this senescent cell burden index was also predictive of a biological response to D + Q in postmenopausal women⁸. As shown in Fig. 1, this was indeed the case: the SASP-score T3 group also demonstrated increases in P1NP similar to those seen in the T cell p16 T3 group. Of interest, the subset of participants who were in the T3 group for both T cell p16 mRNA levels and the SASP score — perhaps identifying those participants with the highest burden of senescent cells — had even more robust increases in serum P1NP levels following D + Q treatment than those who were in only one T3 group (Fig. 1). Collectively, these findings are consistent with the hypothesis that the underlying senescent cell burden dictates the clinical response to a senolytic intervention.

Fig. 1 |. Schematic summary of findings on the effects of D + Q on a serum marker of bone formation (P1NP) from the clinical trial by Farr et al.

Specific details of the data can be found in the original publications^2,8. Differences are shown in median per cent changes (D + Q above control) in serum P1NP levels 2 weeks following D + Q administration in all women; those in the highest tertile for T cell p16 mRNA levels (T3); those in the T3 tertile for a SASP score; or those participants who were in the T3 tertile for both measures.

Further support for this hypothesis was provided by a subsequent study by Millar et al.³ in older adults with mild cognitive impairment. In this study, 12 patients with mild cognitive impairment were administered D + Q every 2 weeks for 12 weeks. These investigators also assessed T cell p16 mRNA levels and, despite the relatively small number of participants, they found that those within the high senescent-cell-burden subgroup (levels above the median) had the largest change from baseline in dual-task gait speed and dual-task cost of stride length.

Both our study² and that of Millar et al.³ used the T cell p16 mRNA assay as validated by the Sharpless laboratory for assessing senescent cell burden⁷. It is important to note, however, that although T cell p16 mRNA levels may be predictive of a biological response to a senolytic intervention, further studies are needed to define whether T cells with elevated p16 mRNA levels are themselves truly senescent or whether an increase in senescent cell burden across tissues simply leads to an increase (perhaps via the SASP) in the expression of p16 in a specific subpopulation of Tcells. Nonetheless, regardless of the underlying biology, T cell p16 expression does appear to have utility in stratifying participants in clinical trials. However, additional studies need to be done to assess the variability over time in T cell p16 mRNA levels, as well as correlated SASP measures in a given individual, to define the robustness of these measures in stratifying study participants for senescent cell burden.

In addition to identifying potential responders in future senolytic trials, our analyses also revealed another issue that warrants further study. Specifically, we found that despite treatment with D + Q every 4 weeks for 20 weeks, the increase in the bone formation marker serum P1NP peaked at week 4 and then returned to baseline by 20 weeks. There are several possible explanations for this; the first has to do with the underlying bone biology and the second is potentially directly relevant to the design of future senolytic trials. In terms of the bone biology, the time course of changes in P1NP we observed is almost identical to that observed with the FDA-approved bone-anabolic agent romosozumab⁹. Thus, it is plausible and perhaps even likely that following the initial increase in bone formation with either D + Q or romosozumab, there are compensatory mechanisms that are upregulated in the bone micro-environment that limit the anabolic effects of either intervention over time. The second possibility, which is more directly relevant to senolytic trials, is that after the clearance of a subset of susceptible senescent cells by a given senolytic (in this case, D + Q), the remaining senescent cells are either resistant or develop resistance with continued administration (similar to the emergence of drug-resistant cancer cells). Although, to our knowledge, this issue has not thus far been documented in any of the preclinical studies of senolytics, the time course of the skeletal response to D + Q in our trial does raise this possibility as one that warrants consideration. If this were true, then future trials may benefit from prolonging the duration between dosing to prevent the development of resistance or use sequential treatment with another senolytic that targets different survival pathways in senescent cells to more broadly optimize their clearance.

A further implication of these findings is that there may well be a complex interaction between the existing senescent cell burden and the potency of senolytics, which will presumably differ among compounds. Thus, as depicted in Fig. 2, as senescent cell burden increases with age or earlier in the context of disease, the first-generation senolytics (for example, D + Q) with modest potency may only be effective in individuals with a very high senescent cell burden (that is, in the highest tertile). A logical hypothesis that warrants further testing is that as more potent senolytic compounds (which are currently in active development) become available, these second-generation (or later) compounds may well have efficacy not only in individuals with a high senescent cell burden but also in those with a lower burden of senescent cells (Fig. 2). Thus, as new senolytics are developed, this hypothesis can be rigorously tested both in animal and human studies.

Fig. 2 |. Working model for the possible interaction between senescent cell burden and potency of a senolytic intervention.

The solid black line depicts that notion that senescent cell burden increases with aging or in the context of specific diseases; shaded area indicates possible variation amongst individuals. The purple dashed line indicates that first-generation, modest senolytics (for example, D + Q) may be effective only in aging individuals with a very high senescent cell burden. The orange dashed line indicates that second-generation (or later) senolytics, by virtue of greater potency, may be effective not only in individuals with a high senescent cell burden but also in those individuals with lower levels of senescent cells.

Moving towards a personalized approach in future senolytic trials

On the basis of these considerations, it is clear that future clinical trials — and, indeed, the potential clinical use of senolytics — are probably going to require a much more personalized approach, as is now standard of care for patients with cancer. Specifically, in oncology, the underlying cancer mutations and disease pathways are increasingly driving the choice of therapy. In that context, given the multiplicity of aging mechanisms, the success or failure of clinical trials of senolytics may depend critically on identifying those individuals in whom senescence is a dominant mechanism that drives organismal aging, as the contribution of senescence to the aging process probably differs across individuals. Consistent with this, we have found that high T cell p16 mRNA levels are present almost exclusively in otherwise generally healthy women over the age of 70 years⁸, but there is nonetheless considerable heterogeneity in these levels in older women: a substantial proportion have relatively low levels of T cell p16 mRNA, for whom a senolytic may not be effective.

On the basis of these considerations, we offer some recommendations for future clinical trials. Specifically, there should be an ongoing focus on the continued development of better biomarkers for senescent cell burden. To date, the measurement of T cell p16 mRNA levels may be the most useful biomarker for the stratification and potential selection of participants in clinical trials. In addition, assessment of panels of SASP markers, including the six identified by our group⁸, may prove useful in evaluating senescent cell burden. Along these lines, it would be important to evaluate whether the SASP score we developed⁸ (that is, the geometric mean of the six key SASP factors noted earlier) has utility in stratifying study participants for senescent cell burden where the outcomes are unrelated to the skeleton. However, these approaches need to be considerably expanded to include larger scale proteomic analyses (and other omic analyses — for example, lipidomic analysis of extracellular vesicles) to develop biomarkers for senescent cell burden. Indeed, given the potential ability to predict organ-specific aging using proteomic analyses¹⁰, it may be possible to identify a high senescent cell burden in specific organs (for example, the brain) and select study participants with increased senescent cells in those organs that are relevant to the outcomes of the specific trial. Conversely, studies that examine a broader range of outcomes and overall healthspan may use a different set of proteomic (or other) analyses — perhaps those more relevant to aging across tissues rather than in a specific tissue — to stratify and select study participants.

In summary, we propose that the next steps in the translation of senolytics should focus on the continued development of better biomarkers for senescent cell burden and studies to test whether aging individuals identified as having a high senescent cell burden on the basis of these biomarkers also exhibit the most robust clinical responses to a senolytic intervention. At a broader level, the aging field can learn from the increasingly personalized approach to cancer treatment and customize treatments on the basis of the key mechanisms that drive aging in a particular individual, which undoubtedly will differ widely across people.