Mechanistic modeling explains the production dynamics of recombinant adeno-associated virus with the baculovirus expression vector system

Abstract

Current manufacturing processes for recombinant adeno-associated viruses (rAAVs) have less-than-desired yields and produce significant amounts of empty capsids. The increasing demand and the high cost of goods for rAAV-based gene therapies motivate development of more efficient manufacturing processes. Recently, the US Food and Drug Administration (FDA) approved the first rAAV-based gene therapy product manufactured in the baculovirus expression vector system (BEVS), a technology that demonstrated production of high titers of full capsids. This work presents a first mechanistic model describing the key extracellular and intracellular phenomena occurring during baculovirus infection and rAAV maturation in the BEVS. The model predictions are successfully validated for in-house and literature experimental measurements of the vector genome and of structural and non-structural proteins collected during rAAV manufacturing in the BEVS with the TwoBac and ThreeBac constructs. A model-based analysis of the process is carried out to identify the bottlenecks that limit full capsid formation. Vector genome amplification is found to be the limiting step for rAAV production in Sf9 cells using either the TwoBac or ThreeBac system. In turn, vector genome amplification is hindered by limiting Rep78 levels. Transgene and non-essential baculovirus protein expression in the insect cell during rAAV manufacturing also negatively influences the rAAV production yields.

Keywords: AAV-based gene therapy; baculovirus expression vector system; biomanufacturing; gene therapy; mechanistic modeling; recombinant adeno-associated virus; viral vector; viral vector manufacturing.

Graphical abstract

Figure 1

TwoBac and ThreeBac constructs (a) TwoBac delivers the genetic information for rAAV production through two BVs: goiBV and repcapBV. goiBV carries the ITR/GOI cassette, while repcapBV contains a Rep cassette from which both Rep52 and Rep78 are expressed and a Cap cassette from which the structural proteins are expressed. In ThreeBac, the same goiBV as in TwoBac is used, but Rep and Cap are delivered through two separate BVs: repBV and capBV, respectively. In repBV, Rep proteins are expressed through separate Rep52 and Rep78 cassettes. The promoters in all cassettes in TwoBac and ThreeBac are the strong A. californica very late promoters polh and p10, except for the Rep78 cassette in repBV, which contains the O. pseudotsugata weak immediate-early promoter ΔIE1 instead. (B) TwoBac: BV infection model. Uninfected cells duplicate with kinetic constant $\mu$ and present low death kinetics ($k_{death.T}$). Virions in the medium bind uninfected cells with kinetic constant $k_{bind0}$. Infected cells are infected by additional virus with decreased binding kinetics ($k_{bind}$). Coinfection from repcapBV and goiBV is necessary for rAAV production. Infected cells do not duplicate and experience an accelerated death kinetics ($k_{death,I}$). Budded virions are released in the very late infection stage with rate $k_{rel}$. Virions in the medium slowly degrade ($k_{d,V}$ ). (C) TwoBac: intracellular reaction-transport network. Receptor-bound BV is transported into the nucleus ($k_{\mathit{int}}$). Rerouting to lysosomes leads to degradation of several ($\eta$) BVs. In the late stage of the infection, the viral DNA replicates ($k_{repl}$) in the nucleus. Transcription from the promoters used in TwoBac (polh and p10) occurs during the very late infection stage. The two promoters present slightly different transcription kinetics ($k_{p10}$ and $k_{polh}$). Rep52 and Rep78 are synthetized ($k_{transl}$) from the Rep transcript, while empty rAAV capsids form from the structural proteins translated from the Cap transcript. The transgene is expressed through the p10 promoter. Vector genome amplification ($k_{ampl}$) originates from the ITR/GOI cassette of goiBV. Rep78 and Rep52 play a role in vector genome amplification ($K_{ampl}$) and encapsidation ($k_{encaps}$), respectively. Non-negligible degradation is registered for Rep proteins ($k_{d,rep}$), for mRNA ($k_{d,mRNA}$), and for the non-encapsidated vector genome ($k_{d,DNA}$). Dashed lines indicate that reactants are not consumed in the reaction. (B) and (C) were created with BioRender (https://biorender.com/).

Figure 2

Model validation: BV infection dynamics Shown is validation of cumulative extracellular and intracellular BV DNA concentration: model prediction vs. experimental measurements (data are from Figure 2 of Vieira et al.⁴¹).

Figure 3

Model validation: intermediates dynamics (A) mRNA concentration. Data are from Figure 7 (AcNPV-Bgal) of Mitchell-Logean and Murhammer. (B) Total rAAV capsids (TwoBac, rAAV-5). Data are from an in-house experiment. (C) Total rAAV capsids (ThreeBac, rAAV-2). Data are from Figure 5 of Meghrous et al. (D) Rep (TwoBac, rAAV-2). Data are from Figure 2 of Smith et al. (E) Rep (ThreeBac, rAAV-5). Data are from Figure 4 of Urabe et al. (F) Non-encapsidated vector genome. Data are from Figure 3A of Li et al. (G) GFP. Data are from Figure 3B of Li et al.

Figure 4

Model fit to the ThreeBac dataset used for model calibration Shown is the model fit to the ThreeBac dataset (rAAV-2) used for estimation of $k_{transl}$, $K_{transl}$, $k_{ampl}$, $K_{ampl}$, and $k_{encaps}$: model predictions (circles) and experimental data (bars). Error bars refer to experimental data. (A) Total rAAV capsids at 72 hpi. Data are from Figure 3 of Aucoin et al. (B) Filled rAAV capsids at 96 hpi. The reported experimental data are obtained by multiplying the infective viral particle titer reported in Figure 2A of Aucoin et al.⁴⁵ by a factor of 1,000. (C) Filled-to-empty capsid ratio.

Figure 5

Model validation: rAAV production (A) ThreeBac (rAAV-2). Data are from Figure 5 of Mena et al. (B) TwoBac (rAAV-5). Data are from an in-house experiment. (C) TwoBac (rAAV-2). Data are from Table 1 of Smith et al.^, and from Figure 3A of Kurasawa et al. Shown are model predictions (bars) and experimental data (markers). (D) OneBac (rAAV-5). Data are from Figure 4A (bioreactor) of Joshi et al. In all plots, rAAV concentration [#/cell] is reported as rAAV concentration in the system [#/mL] normalized by the cell density at inoculation [cell/mL].

Figure 6

In silico analysis of the process: sample plots Shown is the in silico analysis of the process: sample plots for TwoBac and ThreeBac, generated with MOI = 3 PFU/cell for each BV and cell density at infection equal to 2 × 10⁶ cell mL⁻¹. The 95% confidence interval (CI) of the model prediction for TwoBac is also reported. (A and B) Total concentration in the system: (A) viable coinfected cells and (B) filled rAAV capsids. (C–L) Intracellular concentration in viable coinfected cells: (C) BV bound to the cell surface (for each type of BV of the system), (D) BV DNA in the nucleus (for each type of BV of the system), (E) Rep transcripts, (F) Cap transcript, (G) transgene transcript, (H) GFP, (I) Rep proteins (the Michaelis-Menten [MM] constant for Rep78 limitation to vector genome amplification is reported), (J) empty capsids, (K) non-encapsidated vector genome, and (L) filled capsids. VCC, viable coinfected cell (i.e., a viable cell that has a coinfection capable of rAAV production).

Figure 7

In silico analysis of the process: reaction rates and sensitivity analysis Shown is the in silico analysis of the process: reaction rates and sensitivity analysis, generated with MOI = 3 PFU/cell per BV and cell density at infection equal to 2 × 10⁶ cell mL⁻¹. (A) Dynamic trends of reaction rates for TwoBac (continuous line) and ThreeBac (dashed line) in viable cells presenting productive coinfection. (B) Sensitivity of rAAV production to TwoBac model parameters (Equation 52). (C) Comparison among rates of vector genome amplification, vector genome encapsidation, and rAAV (empty) capsid synthesis in TwoBac simulation.