Accelerated Approval or Risk Reduction? How Response Biomarkers Advance Therapeutics through Clinical Trials in Cystic Fibrosis

Abstract

Progress in the development of new therapies for cystic fibrosis (CF) has benefitted from therapeutically responsive biomarkers to streamline drug development. Paradoxically, these response biomarkers have proven to be essential even in the presence of data demonstrating a lack of correlation with clinical outcomes across individuals with CF. This finding is not surprising, particularly in the setting of a rare disease given complex disease processes and an often-limited pool of clinically effective therapies by which to link biomarker and clinical responsiveness. While many response biomarkers will be unable to progress from their status as markers of biologic efficacy to either established correlates of clinical efficacy or surrogate endpoints, they remain critical to the overall success of therapeutic development.

Types of Response Biomarkers and their Complicated Relationship to Clinical Outcomes in the Rare Disease Setting

The advancement of new therapeutics in the rare disease setting is critically dependent on informative biomarkers that can reduce the burden of clinical development efforts in small patient populations. Response or pharmacodynamic biomarkers are indicators of mechanism of action or intended drug activity and include those which may have the additional property of being correlated with measures of clinical benefit or harm. A subset of these response biomarkers known to be clinical correlates include those which can reliably and accurately predict clinical response and serve as surrogate endpoints that may be able to replace traditional clinical efficacy outcomes in pivotal clinical trials (Figure 1)[1, 2]. Evaluation of response biomarkers through the clinical trial process progressively aims to improve confidence in the relationship between an intervention, a biomarker, and a clinical endpoint[3]. Qualification of response biomarkers for use in later phase clinical development has been well-detailed by regulators and is motivated by the FDA’s critical path initiative[4], offering the opportunity to streamline development efforts through the use of surrogate endpoints in an accelerated approval pathway[1].

Key figure, Figure 1. Progression of response biomarkers as their association with clinical outcomes strengthens.

Response biomarkers are by definition markers of biological efficacy, but only some progress to become established clinical correlates and surrogate endpoints (center). The right panel depicts the continuum of confidence that can exist in the association between a response biomarker and clinical outcomes, with the lower panel exhibiting a parallel continuum reflecting the context for use of the biomarker which is dependent on confidence in association with clinical outcomes.

Although qualification of a biomarker as a surrogate endpoint for a clinical outcome is a worthy goal, only a minority of response biomarkers progress beyond being biological efficacy measures capturing treatment-associated perturbation of biological systems believed to be integral to a disease process (Figure 1). There are a multitude of reasons for this, many of which are amplified in the rare disease setting. Beyond the hurdles of biomarker discovery and quantitation, there is a challenge of resolving mutually-dependent conditions: response biomarkers are by definition sensitive and specific to therapeutic interventions targeting a disease-relevant biological pathway. Thus, the progression of a response biomarker towards increased utility in drug development is dependent on the availability of biomarker data from completed clinical trials of a clinically effective therapy targeting the same pathway. In 2019, there were 21 novel drug approvals to treat orphan diseases affecting 200,000 or fewer Americans. Although this represents great progress in the advancement of rare disease therapies, many of these programs involved first-in-class therapies and studies with small sample sizes that limit the characterization of treatment-related associations between biomarkers and clinical efficacy endpoints. It is often the case that orphan drug developers must utilize biomarkers to inform clinical development without the benefit of confirmatory data establishing the responsiveness of the biomarker to similar interventions, let alone establishing the associations between biomarker response and changes in clinical efficacy.

Despite these challenges, there are advantages in using response biomarkers to advance novel therapeutics through clinical development in parallel with the very efforts to evaluate them. This reflects the drug development experience in cystic fibrosis (CF), an autosomal recessive disorder caused by mutations in the gene encoding the CF transmembrane conductance regulator (CFTR) affecting over 30,000 individuals in the United States (U.S) and approximately 70,000 worldwide[5]. Over the last two decades, significant progress has been made through the discovery, clinical trial evaluation, and ultimate regulatory approvals of therapies addressing both the complex clinical manifestations of the disease and the underlying genetic defects caused by mutations in the CFTR gene[6]. This progress has been driven by engaged industry sponsors in partnership with regulators, academic researchers, and established clinical trial networks including the CF Therapeutics Development Network, a multicenter clinical trials network supported by the CF Foundation[7]. Key to the success of therapeutic development in CF has been the use of response biomarkers to confirm mechanism of action, drive decision making, and suggest streamlined development pathways. This perspective highlights the important role of specific response biomarkers in advancing CF therapeutic development despite complexities confirming their clinical utility.

Benefits of Response Biomarkers in Early Phase Studies Irrespective of their Correlation with Clinical Response

An inherent goal of many biomarker development efforts is demonstration of a robust association between changes in a biomarker and a clinical outcome, which can lead to elevation of a response biomarker’s status to a clinical correlate (Figure 1). A mechanism towards establishing this in the setting of response biomarkers is to use data from prior completed trials to demonstrate that treatment-associated biomarker responses correlate with changes in clinical outcome (positive or negative) across individuals in the treated population. In CF, there are several treatment-associated response biomarkers that play an essential role in advancing clinical development programs[8], but which share a perplexing lack of correlation with clinical response at the individual level. Evidence collected to date suggests that these biomarkers may never become clinical correlates, yet developers and regulators have grown to rely on their use and CF drug development would be severely hampered in their absence.

People with CF are prone to chronic microbial airway infection, with the species Pseudomonas aeruginosa (Pa) being a particularly problematic opportunist associated with morbidity and mortality and thus a target of several antimicrobial drug development efforts[9–11]. These topical antimicrobial treatments are delivered as inhaled aerosols directly to the airway, and Pa density in the expectorated sputum of study subjects (as log₁₀ colony forming units [CFU]/g) has proven to be a very useful biomarker in early phase development of CF anti-Pa therapies[12]. Demonstration that a given antimicrobial treatment has produced bacterial killing and appears to lie within the linear portion of a Pa sputum density change dose response curve has helped rationalize Phase 3 and subsequent commercial drug dosage regimens[13]. Importantly, Pa density change can be obtained during relatively short (14 to 28 days) studies and employing substantially fewer subjects than would be required to demonstrate clinical efficacy[14].

Maturation of sputum Pa density as an important response biomarker in CF has benefitted from the experiences of several inhaled antimicrobial development programs that have ultimately led to regulatory approval[15]. Data from large Phase 3 studies in which both Pa density changes and clinical efficacy were collected have reinforced the observation that agents capable of reducing Pa density in the CF airway are also capable of providing clinical benefit at the population level[16, 17]. However, realization of a population-level relationship between biomarker change and clinical response is not demonstration that these two measures correlate at the individual level, and in the case of Pa density, evidence demonstrates only a weak correlation at best (Figure 2). A CF clinician cannot be confident therefore that a treatment-associated Pa reduction in his or her patient’s sputum Pa bacterial density will be associated with observable clinical benefit, or conversely that treatment will not be associated with clinical benefit in the absence of detectable biomarker change. This reality justifies a valid reluctance to consider the further qualification of Pa density as a surrogate endpoint substituting for a clinical outcome in future CF anti-pseudomonal trials.

Figure 2. Association between a CF response biomarker and clinical outcome: bacterial density and lung function.

Weak association was observed between changes in Pseudomonas aeruginosa (Pa) density and absolute changes in lung function as measured by forced expiratory volume in one second (FEV₁) (as a percentage of that predicted based on reference equations derived in healthy populations) in a study investigating response after an average 14-day course of intravenous (IV) anti-pseudomonal antibiotics (Abx). This is despite meaningful and significant average reductions in bacterial density (−1.53 log10 CFU/g, 95% CI −2.97 −0.11) and improvements in FEV₁ % predicted (9.8%, 95% CI: 6.4 13.2) observed across the study population[44, 45]. This lack of association has been observed across other clinical outcomes[46], and is further evidenced by a lack of association between antimicrobial susceptibility testing and clinical response in CF[47].

Reasons for the disconnect between clinical response and reduction in Pa sputum density at the individual level extend beyond the potential for “off target” efficacy with inhaled antimicrobials. CF sputum is extremely heterogeneous in terms of both content and airway origin, and a single sample of Pa sputum at limited time points in a clinical trial is likely inadequate to fully capture the changing lung microbial environment[18]. Second, CF is a multi-organ disease with a variety of complications; as Pa has been associated with multiple poor clinical outcomes ranging from decreased lung function to increased risk of pulmonary exacerbation (a respiratory event requiring acute antibiotic intervention), it is unclear how to optimally capture clinical benefit. The potential for an individual to exhibit “benefit” with respect to a given outcome measure is likely linked to disease stage and state thus complicating the choice of optimal outcome measure across trials enrolling a broad patient population, as often the need in a rare disease setting[14]. Individuals with CF at an early stage of lung disease and with essentially normal lung function have less room for lung function improvement than those who have experienced reversible loss, and in prior antimicrobial trials have demonstrated reductions in risk of exacerbation without a corresponding improvement in lung function[19–21]. On the other hand, adolescents with CF appear to have more reversibility of lung function loss than do older adults and have proven to have the greatest improvements in lung function in prior clinical trials[16]. Finally, inhaled antimicrobials with Pa activity can also have off target activity against other bacterial opportunists in the CF airway. In a given individual, antimicrobial treatment benefit may result as much or more from suppression of one of these other species and thus not correlate well with observed Pa density changes (Figure 3).

Figure 3. Changes in Pa density and lung function over time in an inhaled antibiotic trial of levofloxacin[48].

Off-target efficacy may have contributed to the apparently paradoxical results of an open-label CF clinical trial comparing antimicrobials inhaled tobramycin (TIS, tobramycin inhalation solution) and levofloxacin (LIS, levofloxacin inhalation solution), both of which are antipseudomonal antimicrobials: subjects with CF treated with inhaled tobramycin averaged a >1 log₁₀ reduction in Pa sputum density at 4 weeks which was accompanied by a modest lung function benefit (not significantly different from zero), while subjects treated with inhaled levofloxacin averaged a lesser Pa density reduction but a greater (non-zero) lung function benefit. Reprinted from [48] Journal of Cystic Fibrosis copyright 2015, with permission from Elsevier.

Difficulties in establishing response biomarkers as clinical correlates in the context of drug development are not limited to therapies targeting very small patient populations or those with only modest therapeutic response profiles. A case in point from the CF experience is the recent approvals of a novel class of systemic disease-modifying CF therapeutics called modulators[6], which work by restoring function of the protein responsible for the primary CF defect, the CF transmembrane conductance regulator (CFTR). CFTR protein function is responsible for chloride and bicarbonate transport across epithelial cell membranes; CFTR mutations result in reduced or absent CFTR function which is responsible for CF disease sequelae in multiple organs. Measurement of CFTR function is central to CF diagnosis and is achieved by measuring elevated (>60 mM) concentrations of chloride in CF sweat, which have been demonstrated to be relatively stable over the lifetime of people with CF[22]. Differing levels of diagnostic sweat chloride concentration have been shown to distinguish individuals with differing levels of CFTR function[23].

When assessing the activity of CFTR modulators, which target the functionality of mutated CFTR proteins, a change in sweat chloride concentration is therefore a compelling biomarker for CFTR functional improvement. In fact, treatment-associated reduction in average sweat chloride concentration has evolved to become a “requirement” early in development for systemic CFTR modulator candidates advancing to larger clinical efficacy studies, with the benefit of needing only a fraction of subjects to characterize modulator dose-response relationships in comparison to clinical outcomes including lung function due to its low variability[24, 25]. Minimum thresholds of average treatment-associated sweat chloride changes (of about 10–20 mM) at the population level have corresponded to modest but clinically significant improvements in lung function (3–5% increase in lung function), exacerbation rate, and weight change in later phase trials[26, 27]. CFTR modulators with activity far exceeding minimum thresholds with respect to average sweat chloride change (e.g., >50 mM) have been associated with remarkably greater clinical responses (10–15% improvement in lung function)[28, 29]. Although the rationale of sweat chloride change as a biomarker for increased CFTR function is clear, and CFTR modulators that decrease average sweat chloride concentrations have been shown on average to improve clinical outcomes across trial populations, multiple clinical studies have failed to demonstrate correlation between sweat chloride changes and clinical response at the individual level (Figure 4)[30, 31]. In addition to the possibilities of differences in modulator distribution and bioavailability and end organ responsiveness between the dermis (i.e., the site of sweat glands) and organs associated with CF disease sequelae (e.g., the airway), it is also likely that differences in an individual’s potential to clinically respond (e.g., potential for improvement in lung function in individuals with no lung damage) also contribute to a lack of correlation[32]. Variability of CF lung disease driven in part by modifier genes as well as possible response differences between healthy (sweat glands) and diseased (airways) tissues may further complicate associations at the individual level. Finally, it may be that the sweat chloride concentration achieved post modulator treatment (rather than a change) is a more indicative measurement of future clinical benefit[33, 34], but comprehensive assessment may require both a broader spectrum of response and longer term follow up than currently available [35].

Figure 4. Association between a CF response biomarker and clinical outcome: Sweat chloride and lung function.

There was lack of evidence of any association observed between changes in sweat chloride concentration and absolute change in FEV₁ % predicted in a 6-month study of a CFTR modulator ivacaftor despite large and clinically significant average reductions in sweat chloride (−53.8 mmol/L, 95% CI: −57.7,−49.9) and improvements in FEV₁ % predicted (6.7%,95% CI: 4.9%,8.5%) across the population[31]. Adapted with permission of the American Thoracic Society. Copyright © 2020 American Thoracic Society. All rights reserved. The American Journal of Respiratory and Critical Care Medicine is an official journal of the American Thoracic Society. Readers are encouraged to read the entire article for the correct context at [https://dx.doi.org/10.1164%2Frccm.201404-0703OC]. The authors, editors, and The American Thoracic Society are not responsible for errors or omissions in adaptations.

One may ask whether the lack of data confirming a response biomarker as a clinical correlate devalues it in the setting of drug development. Although the absence of such an association may infer more risk when utilizing the biomarker for decision making in early phase development, biomarkers linked directly to therapeutic mechanism are unquestionably valuable. Our CF experience exemplifies the difficulty with correlating clinical responses of even highly effective therapies with biomarkers directly linked to the mechanism of action of those therapies, and this has largely necessitated the continued reliance on clinical efficacy endpoints in later phase trials to satisfy regulatory requirements. These response biomarkers have however remained of critical use in early phase development. It is reasonable to expect in other rare disease settings that critical and informative response biomarkers may not be able to progress from their status as markers of biologic efficacy to established correlates of clinical efficacy, and yet can remain central to the success of early phase development.

Biomarker Responsiveness Enables Risk Reduction in the Decision to Progress from Early to Late Phase Trials

A critical decision in early- to mid-phase clinical trial planning is whether trials should be designed and statistically powered for establishing biologic versus clinical efficacy. In the rare disease setting, where patient resources are more limited, developers are often forced to design more modest early phase studies prioritizing proximal biomarkers and although clinical endpoint data are often collected, the studies are usually underpowered for their evaluation. The decision of whether to advance a therapy to later stage development then becomes a more complex and holistic process weighing safety, biomarker responsiveness, and typically underpowered clinical response data. The extent to which biomarkers enable risk reduction in clinical development is dependent not only on the magnitude of observed biomarker response but also on the degree to which past biomarker experience has contributed to previous successful development programs. Considerations for making these difficult go/no go decisions are outlined in Table 1.

Table 1.

Considerations for Go/No Go decisions after completion of an early phase study, assuming safety is demonstrated.

	End of Study Clinical Data
End of Study Biomarker Data	Signal*	No Signal
Signal*	Recommendation: GO Ideal scenario for using biomarker data in combination with clinical data to both support a GO/NO decision, and additionally inform dose selection for later phase trials in order to reduce the number of doses evaluated which could greatly impact future study size.	Recommendation: GO Scenario 1: Study was also powered for clinical endpoint  • The mechanism of action of the therapy as captured through the biomarker may not be directly linked to clinical response.  • Longer time may be needed to demonstrate effects on clinical outcomes. Scenario B: Study not powered for a clinical endpoint  • Risk of moving forward is dependent on level of establishment of the response biomarker in similar therapeutic settings.
No Signal	Recommendation: GO   • If utilizing a relatively unestablished biomarker, when feasible early phase studies should ideally also be sized for identifying a signal with respect to the clinical endpoint to allow for this scenario*.   • Mechanism of action may need to be demonstrated in further studies for regulatory approval.	Recommendation: No GO   •Risk for continuing is greatest in this scenario. Consider further early phase development investigating dosing strategy, formulation, and patient population.   • Early phase protocols with amendment potential for additional dose cohorts, for example, may enable more efficient drug development progression in this scenario rather than completely halting future development.

^[*]In the rare disease setting in particular, the definition of the threshold which constitutes a signal need not coincide with strict statistical criteria. This may include evaluating the effect of the outcome within a single treatment arm only (as opposed to formal comparisons with a placebo-group which requires larger sample sizes for adequate power) or utilizing larger type 1 errors that coincide with the level of acceptable risk at this stage of development. Confidence intervals of the therapeutic effect will be important to evaluate the range of effect sizes that can or cannot be ruled out based on the results of the study.

In contrast to the level of support afforded to the development of CF inhaled antimicrobials and CFTR modulators by response biomarkers Pa sputum density and sweat chloride concentration, there has yet to be a biomarker that has proven as beneficial in supporting CF anti-inflammatory development[8]. Local lung inflammation secondary to airway infection drives lung damage in CF, a primary cause of mortality. For this reason, chronic suppression of lung inflammation has been a goal of CF drug developers for decades[36]. Yet, despite recognition of literally dozens of potential response biomarkers in blood and sputum linked to inflammatory pathways, the absence of a clinically safe and effective chronic CF anti-inflammatory therapy with which to evaluate relationships between biomarker(s) and clinical response has meant that there have been no opportunities to “streamline” early development programs based upon smaller studies of biomarker responses[37]. For this reason, Phase 2 studies in this setting often require larger study designs powering for clinical endpoints to ensure that a signal is not missed (Table 1). Several late-phase studies of chronic CF anti-inflammatory therapies are currently in progress however[38], and it may be that if one or more of these are successful, the situation with respect to anti-inflammatory response biomarkers in CF may change dramatically.

Response Biomarkers Need Not Be Surrogate Endpoints to be Valuable in Late Phase Development

Response biomarkers that have been successfully established as clinical correlates have the potential, with rigorous qualification, to progress to surrogate status. However, correlation between biomarker changes and clinical outcomes are not sufficient for elevation to this status[39]. Data from multiple clinical trials of a therapy are typically needed to establish that biomarker changes reliably and accurately predict clinical benefit or harm – and that an average change in the biomarker is able to predict the total expected change in a clinical outcome. At a minimum, this implies that there are no other off target effects of the drug which impact the clinical endpoint. In CF drug development, the primary roadblock to Pa density and sweat chloride progressing towards validation as surrogate endpoints is a more proximal hurdle – a lack (rather than absence) of data even supporting their establishment as clinical correlates. As a result, late phase pivotal CF clinical trials have uniformly relied on established clinical endpoints to demonstrate efficacy. Only recently have expedited regulatory approval pathways been proposed that allow use of in vitro functional assays of CFTR modulator activity as the basis for approval, but this scenario has been strictly limited to label expansion of therapies that have previously undergone regulatory approval for other indications based on establishment of clinical efficacy and safety[40]. Thus, accelerated approval mechanisms employing validated surrogate endpoints, or more often in the setting of rare diseases non-validated surrogates reasonably likely to predict clinical benefit[1], have not been applicable in the CF setting. Considerable hope remains, however, that with greater linkage of biomarker and clinical data derived from use of clinically effective therapies, including data collected outside of interventional trials, validation of response biomarkers as surrogate endpoints will be eventually be realized[41].

For individual drug developers, it can be tempting to conserve development resources and limit biomarker assessment to early development phases, particularly when regulatory approval will ultimately be based on clinical efficacy endpoints in later phase pivotal trials. Although this approach may reap short-term financial savings for the developer, it does so at the expense of overall progress in biomarker development. Further, sponsors may underestimate the extent to which submission of consistent parallel biomarker response data obtained in late phase pivotal studies may support the review of a new drug application for a therapy demonstrating only modest clinical benefit; in fact, it may be prudent to bank clinical specimens from pivotal study subjects as insurance. Only through such systematic data collection will investigators be able overcome the mutual dependence challenge for progressing development of response biomarkers to clinical correlates or surrogates: availability of linked biomarker and clinical data from completed clinical trials of a clinically effective therapy.

Concluding Remarks

Progress in biomarker development in the rare disease space relies on partnerships between academic researchers and industry sponsors that promote data and biospecimen sharing[42, 43]. In addition, quantifying the levels of uncertainty associated with use of less established response biomarkers and understanding where these biomarkers reside on the continuum of qualification with respect to their association with clinical outcomes is essential for their use in risk reduction during therapeutic development. As the CF experience demonstrates, the ability of response biomarkers to meaningfully advance therapeutic development is independent of their progression towards surrogate endpoints.

Outstanding Questions.

• How can stronger partnerships between industry and academia be formed to promote the availability of linked clinical and biomarker data necessary for biomarker evaluation in the therapeutic setting, particularly for rare diseases?

• How can the level of confidence in a response biomarker for a given therapeutic best be quantified to aid in drug development planning, making go/no go decisions, and evaluating program risk?

Highlights.

• Therapeutic drug development in cystic fibrosis (CF) benefits from streamlined drug development programs relying on reliable early phase response biomarkers.

• While the rare disease setting remains a compelling setting for the use of surrogate endpoints to reduce the burden and/or time of late phase development efforts, the CF experience demonstrates that proven response biomarkers exhibit a complicated association with clinical response prohibiting further progression towards evaluation as surrogates.

• Quantifying the levels of uncertainty associated with use of less established response biomarkers and understanding where they reside on the continuum of qualification with respect to their association with clinical outcomes provides an optimal framework for their application throughout therapeutic drug development.