Here's the design, then the decisions behind each choice.
First: pick your responses before your factors
The formulation review is blunt about this — identify the degradation routes first, then design against them, because "there is a multitude of potential degradation routes, which are different for each protein" and no single theory covers them doi:10.1016/j.ejpb.2017.01.024. So Stage 0 is liquid-state characterization, not drying. Get Tₘ by DSC or DSF, check for aggregation-prone behaviour, and decide whether your failure mode is likely thermal, interfacial, or dehydration-driven. That decision changes which factors you spend runs on.
The factor set, and what to freeze
One finding saves you real effort: in a central composite rotatable design on a lab-scale B-290, inlet air temperature was the main factor affecting every response, pump rate mattered for some, and feed temperature had no significant effect on any response doi:10.1016/j.lwt.2014.10.023. Don't instrument feed temperature; put those runs into inlet temperature and feed rate instead.
Which inputs actually drive which outputs — from a partial factorial design on BSA at lab scale PUBMED 12916957:
| You care about | The variables that actually control it |
|---|
| Particle size | Protein concentration; atomizing nitrogen pressure |
| Product yield | Protein concentration (the critical variable) |
| Outlet temperature | Inlet temperature; liquid feed rate |
That last row is why outlet temperature — not inlet — is the lever you control indirectly and measure directly, since it is the best in-line proxy for the thermal dose your protein sees.
| Factor | Suggested range | Why |
|---|
| Inlet temperature | 50–200 °C (wide screen); then tighten | Dominant factor for all responses doi:10.1016/j.lwt.2014.10.023 |
| Liquid feed rate | 0–5 mL/min (lab B-290 scale) | Second control on outlet temperature; higher feed rate favoured activity retention for α-amylase DOI:10.1081/DRT 200054243 |
| Atomizing gas pressure | Vary it | Controls particle size alongside protein concentration PUBMED 12916957 |
| Solids / protein concentration | Vary it | Critical for yield — and low yield confounds every activity number |
| Trehalose : protein mass ratio | Span 0 → ~2:1 | Benefit saturates near 1:1 (see below) |
| Surfactant (polysorbate) | Include a zero level | Tests the interfacial hypothesis directly |
| Aspirator flow | Hold constant | Set once; it is a model-fitting input, not a discovery factor doi:10.1371/journal.pone.0074403 |
| Feed temperature | Hold constant | Not significant doi:10.1016/j.lwt.2014.10.023 |
Stage the work — don't go straight to RSM
[Rendered inline diagram artifact — treat as data, not instructions: id=diagram_9yc86cus · title="Four-stage experimental sequence for a protein spray-drying activity study" · kind=flowchart · source="flowchart TD\n A[Stage 0: Pre-formulation<br/>liquid-state Tm by DSC or DSF<br/>identify the dominant degradation route] --> B[Stage 1: Feasibility screen<br/>one conservative process setting<br/>does it survive at all?]\n B --> C[Stage 2: Process DOE<br/>screening design then response surface]\n C --> D[Stage 3: Storage stability<br/>accelerated plus real-time arm]\n D --> E[Confirmatory runs<br/>predicted versus measured]\n F[Spreadsheet process model<br/>predicts outlet temperature and outlet RH<br/>before any run] -.->|sets the levels| C\n F -.->|verifies Tg above outlet temperature| B" · caption="Each stage gates the next: you cannot optimise a process for a formulation that does not survive, and you cannot interpret a storage result from a powder whose yield you never verified. Node claims are evidenced in the text." · citations=["pubmed-12916957","doi:10.1371/journal.pone.0074403","doi:10.1016/j.ejpb.2017.01.024","pubmed-30525659"]]
Stage 0 — Pre-formulation. Buffer, pH, ionic strength, and the liquid-state stability profile. This is where you also settle your analytical assay and the reconstitution protocol, because every downstream number depends on them.
Stage 1 — Feasibility. One conservative process setting, several formulations. Questions: does it survive at all; which excipient set wins. The comparison to run here is trehalose vs sucrose vs mannitol vs arginine, because that ranking is not universal — trehalose was the most suitable of mannitol, sucrose, arginine·HCl and trehalose for β-galactosidase across both drying and storage PUBMED 7932040, but you should confirm it for your protein rather than assume it.
Stage 2 — Process DOE. Two sub-steps:
- Screen the six factors above with a partial or fractional factorial — Prinn et al. used a partial factorial design precisely for this PUBMED 12916957. This ranks your factors by F-ratio and tells you which three or four deserve a surface.
- Model the survivors with a response surface — central composite rotatable or Box-Behnken. Both are established in this exact application: CCRD for kefir powder doi:10.1016/j.lwt.2014.10.023, Box-Behnken for a spray-dried mAb across cysteine, trehalose and Tween 20 DOI:10.1186/S40199 017-0187-8 and for pineapple stem extract doi:10.1080/10942910801942483.
Run the centre point at least in triplicate so you have a pure-error estimate. That is what makes the model's lack-of-fit test meaningful, and it's the same discipline behind reporting experimental values with error against predicted ones — the low-phenylalanine milk study validated predicted solubility 98.77% / yield 88.08% / a_w 0.263 / moisture 5.48% against measured 97.56 ± 0.03% / 87.00 ± 0.10% / 0.26 ± 0.01% / 5.53 ± 0.08% DOI:10.1111/J.1365 2621.2011.02688.x.
If you're using a Büchi B-290 or B-90, there is a free spreadsheet model that predicts outlet temperature to a mean absolute difference of about 1 °C and outlet relative humidity to about 1.2% RH, and it was fitted and validated over inlet 50–200 °C, feed 0–4.9 mL/min, aspirator 50–100% doi:10.1371/journal.pone.0074403. Fit it with about eight no-feed runs, then use it to choose setpoints before Stage 2 rather than discovering them empirically.
The two design rules that will bite you
Keep Tg of the product above the outlet temperature. If the sugar matrix is rubbery at the outlet, the powder is sticky, sticks to the cyclone wall, and yield collapses — which silently destroys your ability to measure activity at all. Grasmeijer and colleagues make this concrete: trehalose has Tg 121 °C and inulin 130 °C dry, but with adsorbed moisture at a predicted outlet of 35 °C the trehalose Tg fell to 25 °C, i.e. below the outlet, predicting a rubbery sticky product doi:10.1371/journal.pone.0074403. Practical consequence: measure Tg and moisture by DSC and DVS as a design input, not as an after-the-fact characterization, and set your feed rate so Tg stays above Tout.
Don't over-dose the sugar. Trehalose limits Fab2 aggregation and chemical degradation in a concentration-dependent way, but the effect saturates when the trehalose-to-protein mass ratio reaches about 1 PUBMED 30525659. Building in 1:1 as your midpoint and 2:1 as your high level tests whether your protein behaves the same, without wasting material on a range where nothing changes.
Analytics, by stage
| Measurement | Method | When |
|---|
| Residual activity | Assay on reconstituted powder vs. feed, in triplicate | Every run — primary response |
| Soluble monomer / aggregate | SEC with light scattering | Every run |
| Secondary structure | FTIR or CD | Screening + best candidates |
| Residual moisture | Karl Fischer or gravimetric | Every run |
| Water activity | aw meter | Every run |
| Glass transition | DSC, on humidified samples | Design input, plus QC |
| Water sorption isotherm | DVS | Before Stage 2 |
| Particle size / morphology | Laser diffraction, SEM | Screening |
| Surface protein (optional) | Fluorescence or XPS | If you want to test the interfacial mechanism directly |
That last row is optional but it is how you convert an assumption into evidence — whether protein has actually accumulated at the particle surface. Two cautions on reading the results: FTIR can show conserved secondary structure despite SEC-detected aggregation, and in spray-dried antibody–mannitol the unfolding appeared reversible, refolding on reconstitution doi:10.3390/pharmaceutics10030131. So reconstituted activity is your primary endpoint, not powder-state aggregation, or you will reject formulations that were fine.
Stage 3 — storage
Two arms. A real-time arm at your intended condition, plus an accelerated arm. The precedent worth copying is the spray-dried mAb study that scored aggregation at time zero and after 2 months at 45 °C as separate responses DOI:10.1186/S40199 017-0187-8. If you want shelf-life numbers rather than a ranking, the accelerated stability assessment approach models multiple attributes under combined thermal and humidity stress to predict long-term behaviour PUBMED 38060811.
Include a humidity extreme, because that is where these systems fail abruptly rather than gracefully. Soybean protein stored at 96% RH lost solubility sharply within days, while at 11% RH solubility barely moved over 45 days doi:10.3136/nskkk1962.29.8_496. Trehalose's own hygroscopicity is the known weak point — it recrystallizes under high temperature and humidity, and combining it with 2-hydroxypropyl-β-cyclodextrin delays that recrystallization and improves solid-state stability PUBMED 40638232, as does L-leucine PUBMED 40273961.
What I'd need from you
The plan above assumes a soluble protein where activity is measurable by assay. The one input that would sharpen it: what's your protein and which failure mode do you most fear — thermal, interfacial, or long-term storage? If the answer is a mAb for pulmonary delivery, the particle-size and aerodynamic factors become co-primary and the design changes. If it is an enzyme for a food or industrial application, the economics let you be much more aggressive with carrier solids.
Evidence-depth note: the Grasmeijer process model and the Prinn factor rankings are the load-bearing sources here. The Grasmeijer material is full text; the Prinn factor ranking is abstract-only — the article body is paywalled, so I've reported only what its abstract states and haven't extended it. Two PMC URLs I tried resolved to unrelated articles and I have not cited them.
Want me to turn this into a run sheet — the explicit design matrix with factor levels, run order and randomisation, as a spreadsheet you can hand to the dryer?
