Development of an oral once-weekly drug delivery system for HIV antiretroviral therapy

The efficacy of antiretroviral therapy is significantly compromised by medication non-adherence. Long-acting enteral systems that can ease the burden of daily adherence have not yet been developed. Here we describe an oral dosage form composed of distinct drug–polymer matrices which achieved week-long systemic drug levels of the antiretrovirals dolutegravir, rilpivirine and cabotegravir in a pig. Simulations of viral dynamics and patient adherence patterns indicate that such systems would significantly reduce therapeutic failures and epidemiological modelling suggests that using such an intervention prophylactically could avert hundreds of thousands of new HIV cases. In sum, weekly administration of long-acting antiretrovirals via a novel oral dosage form is a promising intervention to help control the HIV epidemic worldwide.

Dosage forms consisting of PLA arms failed in ~2000 cycles while those consisting of Elastollan ® R6000 arms showed no failure for up to 10000 cycles. (D) illustrates the dosage form consisting of Elastollan ® 1185 central elastomer and arms made predominantly of PLA before dosing to pigs (Day 0) and after retrieval from pig stomach (Day 7) (E) shows the dosage form consisting of Elastollan ® 1185 central elastomer and Elastollan ® R6000 arms before dosing to pigs (Day 0) and after retrieval from pig stomach (Day 7) (F) is an image of the Elastollan ® R6000-Elastollan ® 1185 structure in the pig stomach during retrieval on Day 7.

Supplementary Figure 3: Gastric and intestinal residence of dosage forms
Dosage forms equipped with an enteric linker were deployed in the stomach and intestine of the same pig. Using serial X-rays we determined the intactness of the dosage form. (A) shows representative X-rays of dosage form deployed in the stomach (blue circle) and that deployed in the intestine (red). The red arrow indicates an arm detached from the structure deployed in the intestine. The random-effects (RE) model accounting for both study effect size and between-study variance estimates 15% (95% CI: 14-17%) increase in fraction adherent (i.e. MPR or PDR above 80%) when switching from daily to weekly formulation.

Supplementary Note 1 SUMMARY
To predict how individual patients respond to long-acting drug formulations, we adapted an established model of within-host HIV dynamics in the presence of antiretroviral therapy 21 . The backbone of this model tracks the virus and its target cells throughout the plasma and lymph tissue and is calibrated to clinical data. We model the impact of treatment by allowing viral infectivity to depend on drug levels, which fluctuate over time (termed pharmacokinetics), and drug efficacy, determined by the IC50 and slope of the dose-response curve of the particular drug (pharmacodynamics). Both drug-sensitive and drug-resistant viral strains are tracked and interact via mutation and selection. Our model is similar to those that have been used previously to predict the dynamics of HIV 22-24 , hepatitis B virus 25 , hepatitis C virus 26,27 , herpes simplex virus 2 28 , and various bacterial infections 29,30 .

Viral dynamics model
We use a slightly modified version of the well-established viral dynamics model to describe the dynamics of multiple strains of HIV within an individual patient 21,31 . This model tracks uninfected target cells of the virus (CD4 + T cell , x), actively infected cells (y), and free virus (v). Described as a system of differential equations, the model is Uninfected target cells are produced at a rate λ and die at a rate dx. Infection occurs at a rate βi per contact between uninfected cell and free virus of strain i. Mutations can occur during the reverse-transcription phase, and so infection with virus of strain i can lead to an infected cell with strain j with rate Qij, where Q is the mutation rate matrix 21,31 . Here we will track wildtype drugsusceptible virus, a suite of drug-resistant strains, and intermediate genotypes along the pathway to resistance. To account for the observation that only a small fraction of CD4 + T cell are infected even in advanced infection, and yet levels of these cells are dramatically reduced by infection, we assume that only a fraction α of target cells contacted by the virus will proceed to productive infection, while the remainder will be rapidly killed by the virus (e.g. by pyroptosis or another means of "bystander" cell death) [32][33][34] . Infected cells produce free virus at rate k and die at rate dy. Free virus is cleared at rate dv. Infected cells can also be produced by reactivation of latent infection at rate Ai.
This model reproduces many of the important features of HIV infection. During acute infection, the virus starts at low levels and grows approximately exponentially to a peak level, before declining to a lower "set-point" viral load where it remains approximately constant for many years 35, 36 . Total CD4 levels decline substantially during infection, even if only a small fraction of cells are productively infected 37,38 . If viral replication is blocked by effective treatment, then viral load levels decay until they reach a low level sustained by reactivation of latent infection 39,40 . CD4 T cells eventually recover 39 . If treatment is interrupted at any point, viral levels rapidly rebound 41,42 .
The fitness of any particular viral strain can be summarized in a single composite parameter known as the basic reproductive ratio (R0): This value describes the average number of secondary infections caused by a single infected cell over its entire lifespan. If R0 > 1, this strain can undergo continual self-sustaining rounds of replication and reach high levels. When R0 < 1, the strain will be unable to sustain infection, and will only be maintained due to reactivation from latency or production via mutation of another strain. If multiple strains all have R0i > 1, the one with the highest value will outcompete the others to reach high levels, and the others will persist only at low levels.
Here we assume, for simplicity, that strains differ only in their infection rates, β, and their frequency in the latent reservoir, A. We assume that the latent reservoir maintains a constant size, instead of explicitly modeling its dynamics. This is a very good approximation when we are modeling treatment initiated after acute infection, when the majority of reservoir seeding occurs, and after which size is relatively constant 43,44 .
This model can also be implemented stochastically to properly account for the finite size of infection and rare events such as mutation and reactivation from latency (see below).

Treatment with antiretroviral therapy
Antiretroviral therapy acts by preventing infection at a particular stage of the viral lifecycle.
The action of all current antiretroviral drugsincluding fusion inhibitors, reverse transcription inhibitors, integrase inhibitors, and protease inhibitors (which lead to production of non-infectious virus)can be described by reductions in the infection parameter β. The efficacy of a drug as a function of the drug concentration (the pharmacodynamics) is well-described empirically by a Hill-type sigmoidal dose-response curve 21,45,46 : where β is the infectivity in the absence of drug, D is the drug concentration, IC50 is drug level at which infectivity is reduced by half, and m is the slope parameter which describes the steepness of the curve.
If multiple drugs are given in combination, the efficacy equation can be adjusted to account for the combined effect of all drugs, including any drug interactions 47 .
HIV treatment is always taken as discrete doses and consequently drug levels D(t) vary over time. We use a simplified description of these pharmacokinetics, in which drug concentrations instantaneously increases after a dose to a peak concentration and then decay exponentially (with half-life τ1/2) to a trough concentration before the subsequent dose: where t is the time since the last dose.
When a drug is taken continuously with the prescribed dose interval ∆td, it reaches a steadystate where the peak level is termed Cmax and the trough level is Cmin = Cmax 2 −∆td/τ1/2 . The difference between these concentrations, ∆C = Cmax − Cmin, describes the amount by which each subsequent dose increases the drug level, even when doses are missed or before steady state is achieved.
We model imperfect adherence to antiretroviral therapy by assuming each prescribed dose is taken randomly and independently, at the prescribed time, with a probability given by the average adherence level (pA).

Long-acting antiretroviral therapy
To model a hypothetical long-acting formulation of antiretroviral therapy 48-51 we assumed a dosage form existed with identical pharmacodynamics parameters but with a longer half-life and less frequent administration. As a baseline case, we assumed the peak concentration Cmax was the same, but also explored lower values. Otherwise the parameters for long-acting antiretrovirals were assumed to be the same as for the daily formulation.
For hypothetical weekly long-acting formulations, we considered two different potential adherence scenarios. The first, weekly-choice adherence, is simply a weekly analogue of the case for existing drugs. We assume that the patient has a fixed day they are supposed to take a new pill (e.g. Monday), and each week on this day, they randomly decide, with their adherence probability pA, whether or not they will take the pill for that week. If they do not take it, they must wait until the next week to do so. This scenario may represent a particular behavioral strategy, or could occur if pills are only made available to patients on a weekly basis. We also consider a more optimistic scenario for weekly long-acting therapy adherence, which we call daily-choice adherence. In this situation, patients may decide to take their next dose on any day starting one week after their previous dose. Each day after this point, they (randomly) decide with probability pA whether or not to start the next week-long dose. In this case, for any value of pA, the chance that this individual will go a whole week without ever deciding to take their next pill is much lower. We think this scenario may be a natural consequence of human decision-making behavior and represents a possible way in which long-acting formulations may effectively increase adherence.

Viral dynamic parameters
Generally viral dynamics models track infection only in the blood, since that is the compartment most readily sampled during routine clinical care. However, the majority of the infection takes place in the lymph tissues, which has been shown in many studies to be well-mixed with the blood 52,53 . Because the study of resistance requires tracking total body levels of infection, we derived scaling factors to relate the full-body infection size to levels in the blood. To relate the blood concentration of CD4 + T cell (x) to the total body level (X), we use the following scaling: where CD4 =10 6 Vplasma /fbp fcb. Similarly, to scale virus levels measured as viral RNA copies per mL of plasma (v) to total body virions (V), we assume that virus distributes between the periphery and lymph with the same proportion as cells and use: where v =500 Vplasma /fcb. Supplementary Table 2 gives the values we used for these scaling values and their sources. Supplementary Table 3 gives the parameters used in the viral dynamics model for the wildtype virus. Alterations to viral fitness caused by the drug and resistance mutations are described by the parameters given in the next section.
With these parameter values, pre-treatment total body levels of infected cells and free virus are of order 10 8 and 10 10 , respectively, and healthy levels of CD4 + T cells are of order 10 11 (in agreement with experimental data 54 ).

Pharmacodynamic parameters
For all drugs, IC50 and slope values for wild-type virus were determined by ex vivo laboratory experiments conducted in the Siliciano laboratory at Johns Hopkins School of Medicine 46,55-57 (Supplementary Table 4). We consider the methods of this group to be the gold standard, because a) they use a single-round infectivity assay, b) they use a highly quantitative fluorescent-cellsorting readout, c) they use primary CD4 + T cells and plasma from donors, d) they generate resistance mutations in a constant genetic background. They additionally characterized a suite of resistance strains, which is any viral variant that is less inhibited than the wild type for some drug concentration. Each resistant strain is characterized by three parameters that describe the alterations to the dose response curve of viral infectivity relative to the wild type (β): Resistance mutations increase the IC50 of the dose response curve by a factor ρ and also in general confer a fitness cost, meaning that in the absence of the drug, infectivity is reduced compared to the wildtype (0 < s < 1). This fitness cost is often alternatively described by the replication capacity, 1 − s. Many mutations also reduce the slope (m) of the dose-response curve by a fraction σ < 0 55 .
We also considered other sources of data on resistant strains that were not characterized by the Siliciano group. These other sources do not always meet all the assay standards outlined above, and often only measure the fold-change in IC50 (ρ) of resistant strains, omitting the fitness cost (s) or changes in the slope (σ). When these values were missing, we assumed σ = 0 and s = 0.05 (Supplementary Tables 6, 7).

Pharmacokinetic parameters
For existing immediate-release formulations, the maximal plasma concentration (Cmax) and the half-life (τ1/2) were taken from the FDA drug labels 58 (Supplementary Table 4). For long-acting formulations, we consider a hypothetical scenario. We believe that the pharmacokinetic profiles observed in the swine model in this preliminary study are likely to be different than those observed in eventual human testing, both because drug absorption and elimination are species-specific and because the current dosage forms are hand-made in the laboratory but will undergo extensive improvements and industrial-grade manufacturing before administration to humans. Therefore, we did not focus on characterizing the swine pharmacokinetics in detail and exactly reproducing them in our model. Instead, we assumed a relatively simple profile for a hypothetical clinical long-acting formulation. Since the form will be given once a week, we assumed that the half-life was sevenfold longer than the existing formulation. As a baseline case, we assumed the peak concentration Cmax was the same, but also explored lower values. Otherwise the parameters for long-acting antiretrovirals were assumed to be the same as for the daily formulation.

Adherence
For some results, we present predicted patient outcome as a function of the input adherence level, where adherence was varied from 0 to 100% in increments of 5%. In other results, we present outcomes for a hypothetical cohort of treated individuals with a distribution of adherence levels. We use an empirically-determined distribution of adherence levels to antiretroviral therapy taken from a study using unannounced pill counts 59 .

Mutation rates
The entry Qi,j in the mutation rate matrix Q describes the probability that entry of viral strain i into a host cell will lead to infection with strain j, which is just the probability that the viral genotype i is mutated to produce strain j during reverse transcription. Working out this rate for any particular mutation of interest requires multiple steps. Mutations that confer drug resistance to HIV are described by their genotype at the amino acid level: for example, M184V is a mutation in the reverse transcriptase gene in which the naturally-occurring amino acid methionine (M) at position 184 is changed to a valine (V). To determine the rate at which such a mutation occurs, we need to know which triplet of nucleotides code for the M and V amino acids (redundancy in the genetic code means there will be multiple codons for each), at which frequency each possible codon is expected to exist, which nucleotide could be changed to produce the amino acid change, and at what rate each of those nucleotide mutations could occur. To do this, we use the well-known human codon table, the overall nucleotide composition of the HIV genome, and the nucleotide-tonucleotide mutation matrix for HIV 21,[60][61][62] (Supplementary Table 5). We then apply the following algorithm: • For each mutation requiring multiple amino acid changes, find all strains with all but one of these mutations, and enter the rate to get the final genotype as the same rate at which that final genotype alone would occur from the wild type strain (e.g. M184V→E138K/M184V occurs at the same rate as wild type →E138K: 5.7×10 −5 ) • For all strains, calculate the probability of this strain directly mutating to or from any other strain. If only a single nucleotide change is needed, we use the method above, and if multiple nucleotide changes are needed, we calculate the rate of each single-nucleotide change and take their product (e.g. wild type →E138K/M184V at rate 5.7×10 −5 × 1.15×10 −5 ) • For each strain, calculate the probability that a mutation does not occur, by requiring that the sum of the values across a row is 1. Note that this last step assumes that any mutations not accounted for in the mutation matrix are neutral and equivalent to the wild type, including mutations which may include combinations of the resistance mutations listed that have not yet been observed or phenotyped, and therefore not included in our analysis. In this way, we apply a closure approximation to the mutation matrix instead of modeling a full matrix of 2 n possible genotypes that could be constructed out of n possible mutant loci.
These calculations assume that the mutation rate is independent of the genetic background (other than depending on the initial and final nucleotide) and that synonymous mutations are neutral and their relative frequencies are described by the overall base composition of HIV. In the absence of drug treatment, drug-resistant mutants will still be produced by mutation, but will be selected against due to their fitness cost. The balance between these two processes results in their persistence at an average level described by mutation-selection balance 31,63 . This persistence level can be calculated by constructing a new matrix M where Mi,j = Qi,jβi, and then the frequency fi of each strain i is given by the i th entry in the (normalized) dominant left eigenvector of M. We assume that mutant viral strains exist at mutation-selection balance both in actively infected cells at the time of treatment initiation, and in the latent reservoir (pre and post treatment).

Hybrid deterministic-stochastic method
The differential equation formulation of the model (Eq. (1)) can be easily numerically integrated but is unrealistic, as it results in all viral strains existing at low levels at all times due to continuous production of mutants, and therefore cannot accurately describe the waiting time until rare events occur nor the extinction of particular strains. However, a full stochastic simulation of realistic infection sizes (×10 8 infected cells or ×10 10 virus) and strain structure (e.g. dozens of mutants) is computationally infeasible. To overcome these issues, we implemented a hybrid deterministic-stochastic approach to simulate the model.
We divided all the possible reactions of the system into those we assumed to take place in a continuous, deterministic fashion and those that occurred stochastically, based on whether they are likely to be rare or frequent transitions. Production and death of target cells, production and clearance of free virus, death of infected cells, and infection events that occur without mutation were simulated deterministically. Reactivation of latent cells and infection events in which mutation occurs were simulated stochastically. Stochastic events were simulated in a discrete-time manner, with time steps chosen as the minimum of the time between doses or 1 day. The number of cells of type i exiting the latent reservoir in a time interval ∆t was drawn from Poiss(A fi ∆t). The number of mutants produced by infection events was simply calculated by taking the number of new infections by each strain in the last time step (calculated from the numerical integration) and applying the mutation matrix Q.

Initial conditions
We simulated maintenance therapy, in which participants have previously been on effective triple-drug antiretroviral therapy and are then switched to a simpler treatment regime (often fewer drugs) with the goal of maintaining a low viral load 64 . Therefore we initialized simulations with very low (<1 copies/ml) viral loads at levels sustained purely by release of virus from the latent reservoir (see above). Formally, we assumed that the total level of uninfected and infected cells, and free virus, were given by the steady state conditions of Equation (1) in the limit when β = 0: Note that with the parameters in Supplementary Table 3, this initial condition corresponds to a residual viral load around 0.1 copies/ml. Clinical studies with ultrasensitive assays find values of around 2 copies/ml 65 , but this value represents all virions comprising the residual viremia whereas we only track the virions and reservoir cells that are replication-competent (since our estimates of reservoir exit rate were based on the timing of viral rebound 66 , which must be caused by replication-competent virus). Differences in residual viremia values can be explained by the partially-defective nature of many provirus in the latent reservoir 67,68 .
In both cases, the number of infected cells containing mutant viral strain i was a random number drawn from a binomial distribution (approximated as Poisson-distributed) with an average given by the mutation-selection balance frequency fi (described above).
We assumed the first drug dose was taken and the post-dose peak concentration was ∆C (it will take multiple doses until steady state kinetics with peak level Cmax are reached).

Declaring treatment failure
Viral dynamics are tracked for a year-long period, and treatment failure is defined as two consecutive monthly measurements above 200 copies/ml. Treatment failure is classified as occurring due to resistance if over 20% of the viral population at failure contains resistanceassociated mutations.

Processing of drugs
There are several drugs approved for ART (Supplementary Figure 4), however, given the limitations on the size of a capsule that a patient can swallow, we could only accommodate drugs with a daily dose of <50 mg in our system. Based on this initial screening, we selected four drugs for our study. All drugs were purchased as weak acids or bases. As the commercial oral formulations contain salt forms of the drugs, we converted the parent drugs into their respective salt forms viz. dolutegravir sodium, cabotegravir sodium, rilpivirine hydrochloride and tenofovir alafenamide fumarate (TAF). Unless indicated otherwise, all experiments described in this manuscript were performed with the salt forms of the drugs. TAF was synthesized as described before 1 . Briefly, tenofovir alafenamide (10g, 0.021 mol), fumaric acid (2.44 g, 0.021 mol), was suspended in acetonitrile (100mL) and heated to reflux with stirring until all solids dissolved. The reaction mixture was allowed to stand at room temperature to crystallize TAF, which was filtered, washed with acetonitrile (30mL), and dried before further use.
Dolutegravir sodium was synthesized as described by Johns et al. 2 . The parent acid (10g, 0.024 mol) was suspended in ethanol (30mL) and heated to reflux with stirring. Sodium hydroxide (8.4mL, 2N) was added to the solution while heating. The salt crystallized out on standing. It was filtered, washed with ethanol (50mL) and dried. Cabotegravir sodium was prepared similarly.
To prepare rilpivirine hydrochloride, the parent base (10g, 0.027 mol) was suspended in methanol (100mL) and heated to reflux with stirring. Isopropanol (50mL) containing 4mL of 6N HCl was added to the reaction mixture while heating. The salt was crystallized by placing the mixture at room temperature. The solid crystals were filtered and dried prior to use.
Nuclear magnetic resonance (NMR) and X-ray diffraction (XRD) analyses of DTG, RPV and CAB are shown in Supplemental Figure 5 and 6 respectively.

Estimating a plausible range for the impact of a weekly formulation on pre-exposure prophylaxis (PrEP) efficacy
To estimate a plausible effect size of a weekly oral option as an alternative to daily oral PrEP, we leveraged a meta-analysis of the difference in medication possession ratio (MPR), defined as the number of days of supply within the year that the patient has been prescribed, or proportion of days covered rate (PDR) for once-daily versus once-weekly oral dosing of any chronic medication for non-communicable diseases in non-trial settings 3 . The reported odds ratio of 1.90 (95% CI: 1.81-2.00) for MPR above 80% was translated to a mean difference (MD) of 15.5% (95% CI: 13.6-17.1%) using random effects meta-regression 4 of the studies investigated in 3 .
Medication possession ratio is less meaningful in a clinical trial setting where medication refills are procedurally linked to follow-up visits and substantial effort is made to maximize trial retention. However, other measures of adherence, such as routine or unannounced pill counts, may still be variable among the retained population and provide insight into the relationship between adherence and plasma drug levels. We examined the correlation between pill count rate (PCR) and detectable levels of tenofovir, a component of oral PrEP, in the Partners PrEP study 5 . PCR had an 84% sensitivity to identify subjects with detectable tenofovir in plasma. We thus multiplied the estimated mean difference in adherence by 84% for a plausible mean difference in detectable drug levels between daily and weekly PrEP. It should be noted that PCR is a more conservative estimate of adherence (MPR≥PDR≥PCR) but these three estimates become more similar at high levels of adherence (91% and 90% adherence by PCR in emtricitabine (FTC)/ TDF and TDF arms respectively in Partners PrEP).
Finally, the relationship between detectable drug levels and PrEP efficacy was estimated in a regression across clinical trials 6,7 . Strata of adherence within the trials were not considered to avoid the potential for confounding. A linear regression of PrEP efficacy versus the fraction of subjects with detectable plasma tenofovir in the VOICE, FEM-PrEP, iPrEx, iPrEx-OLE, Partners PrEP, and Partners Demonstration Studies 8 yielded a regression coefficient of 1.54% (R 2 =0.94) increase in efficacy for a 1% increase in the fraction with detectable drug concentrations in plasma.
After combining these estimates, we arrived at plausible range of 20.1% (95% CI: 17.6-22.1%) increase in efficacy for weekly PrEP as compared to daily PrEP. However, because weekly PrEP will likely involve different drugs and may influence efficacy and coverage in unexpected ways, a wider range of efficacy and coverage assumptions were tested using a mathematical model.

Structure, calibration and implementation of the epidemiological model
The stochastic individual-based network model EMOD v2.5 9 , calibrated to HIV epidemic trends in South Africa 10 , was used to evaluate population-level impact of daily and weekly oral PrEP over time horizon of 20 years starting in 2017.
EMOD-HIV is an individual-based model that simulates transmission of HIV using an explicitly defined network of heterosexual relationships that are formed and dissolved according to age-and risk-dependent preference patterns 11 . Mother-to-child transmission is also included in the model. The modeled population was initiated in 1960 and projected forward with age-specific fertility rates and age/gender specific (non-AIDS) mortality rates obtained from UN World Population Prospects. The model was calibrated to match retrospective estimates of age-stratified, national-level prevalence, incidence, and ART coverage from four nationally representative HIV surveys in South Africa. The age patterns of sexual mixing were configured to match those observed in the rural, HIV-hyperendemic province of KwaZulu-Natal, South Africa 12 . Recently, a validation study showed that self-reported partner ages in this setting are relatively accurate, with 72% of self-reported estimates falling within two years of the partner's actual date of birth 13 . Further, the transmission patterns observed in EMOD are consistent with those revealed in a recent phylogenetic analysis of the age/gender patterns of HIV transmission in this setting 14 .
Transmission rates within relationships were assumed to vary according to HIV disease stage, male circumcision, condom usage, co-infections, and antiretroviral therapy 15,16 with the latter reducing infectiousness by 92% within six months of treatment initiation-a conservative estimate based on observational data of serodiscordant couples in which outside partnerships could have contributed to HIV acquisition 17 . All scenarios included male medical circumcision 18 with an assumed 60% reduction in acquisition risk.
We configured the EMOD health care system module to follow trends in antiretroviral therapy (ART) expansion in South Africa. Diagnosis was assumed to be accessible via voluntary counselling and testing (VCT), antenatal and infant testing, symptom-driven testing, and low level of couples testing. The model includes loss to follow-up between diagnosis and staging, between staging and linkage to ART or pre-ART care, and during ART or pre-ART care 19,20 , Two assumptions about future ART coverage were compared: one keeping present-day ART coverage levels of 60%, and another more optimistically assuming that the UNAIDS 90-90-90 targets will be achieved through expanded HIV testing, improved linkage to ART, and reduced lost-to-followup, such that in the model by 2032 90% of all individuals ever tested positive would be on ART.
Population-level impact of switching to weekly formulation for oral PrEP was assessed in terms of additional infections averted compared to daily PrEP. We assumed that a weekly formulation would increase efficacy from a baseline efficacy level of 50% by some positive increment, which was varied in sensitivity analysis. Efficacy was implemented as the reduction in the per-exposure probability of infection. In addition, we explored the possibility of an increase in coverage arising from greater acceptability of weekly oral PrEP, from baseline levels of 30% coverage to up to 60% coverage. PrEP was assumed to be available to high-risk populations, both genders, ages 15-29.

Statistical analyses
One-way ANOVA followed by a post-hoc Bonferroni test was used to compare flexural strength of Elastollan ® R6000 to PCL and PLA, and to compare the Elastollan ® R6000-Elastollan ® 1185 interface to other polymer interfaces. The effect of SGF on flexural strength of Elastollan ® R6000 and Elastollan ® R6000-Elastollan ® 1185 interface and differences in pharmacokinetics of sustained and immediate release dosage form was tested using two-sample t-test. Results of the funnel tests were evaluated using paired two-sample t-test. Variance in the groups being compared was generally similar. A p-value of less than 0.05 was considered statistically significant.