Fusionae_case Study 2

Rapid, Automated Development of a Pharmaceutical Small Molecule Separation Using A Two Phase Method Screening and Optimization Approach Introduction This work, conducted in collaboration with Pfizer Inc., Ann Arbor, describes the use of Fusion AE for Galaxie in a two-phase rapid screening and optimization experiment designed to optimize the HPLC separation of a complex mixture of small organic molecules typically found in pharmaceutical products. Phase 1 - Rapid Screening - consisted of a rapid screen utilizing a novel Trend Response data analysis algorithm designed to to identify the correct column, mobile phase and approximate gradient conditions needed to separate a complex mixture of two API’s and several known impurities. Phase 2 – Optimization - comprised a method optimization experiment that identified the run conditions that gave the best results in terms of resolution and assay time using the column and mobile phase identified from the Phase 1 screen. Overall method robustness was also determined using a novel robustness calculation. The results of this work are presented below.

Materials and Methods Instrument – 1100 HPLC with Diode Array detector (Agilent Inc.). Model 500 Column Valve Module (Varian Inc.) Columns – 150 x 4.6 mm Gemini C18, Synergi Fusion RP, Luna C18, (Phenomenex Inc.), Pursuit DiPhenyl (Varian Inc.), Sunfire C18 (Waters Inc.) Buffer System – Aqueous: Ammonium Acetate Buffer (pH5), Potassium Phosphate (pH 2.5 & 7). Organic: Acetonitrile System Parameters Included as Experiment variables – Flow Rate, Gradient Slope, Gradient Time, pH and Column Type Rapid Method Development Platform – Instrument control, chromatogram generation, peak processing: Varian Galaxie chromatography data system (CDS), (Varian Inc.) Statistical experimental design, data analysis, modeling, optimization: Fusion AE for Galaxie (S-Matrix Corp. CA.) Experimental Method – Study factors for the Phase 1 Rapid Screening experiment were varied according to a model-robust screening design generated by Fusion AE, which constructed the 38 run design as a set of ready-to-run methods and the corresponding sequence in the CDS. The experiment was run overnight on the HPLC under Galaxie CDS control. Peak results were imported from the CDS into Fusion AE, using three trend response variables viz. Total number of peaks and total number of resolved peaks (R>1.5 & R>2.0), for automated data analysis.

Figure 1. Phase 1 – Rapid Screening. Experimental parameters (study variables) including, Gradient Slope, Gradient Time Mobile Phase pH and Column Type were entered into a standardized template. Experiment Variable

Range or Level Setting

Gradient Time (min)

15.0 — 40.0

pH

2.5, 5.0, 6.5

Column Type

Gemini C18 Synergi Fusion RP Luna C18 Pursuit DiPhenyl Sunfire C18

Gradient Slope (% Organic)

5.0 — 95.0

Organic Solvent Type

Acetonitrile

Figure 2. Completing the Fusion experimental design template involved setting the upper and lower bound values for the gradient time, target pH range, the specific columns to be screened, and the desired organic solvent type and percentage

Figure 3. A Model-robust Screening type of design was used to efficiently screen large ranges of the variables and quantify their effects on method performance.

Figure 4. Software generated a statistical experimental design. The variables included process and mixture types. Fusion AE therefore selected a mixture process algorithm design.

Figure 5. A Galaxie sequence, comprising 38 individual automatically generated methods based on the experimental design parameters was built by the Fusion AE software within the Galaxie CDS. This sequence was started by user and all lines were run in “inject and forget” mode.

Figure 6. Peak results data were automatically imported from the Galaxie CDS using file-less data transfer.

Experimental Method cont..– Study factors for the Phase 2 Optimization experiment were varied according to a model-robust optimization design generated by Fusion AE, which constructed a 14 run design as a set of ready-to-run methods and the corresponding sequence in the CDS. The experiment was run overnight on the HPLC under Galaxie CDS control. Peak results were imported from the CDS into Fusion AE, using a file-less exchange module, for automated analysis. Optimization solution searches were conducted with the Fusion AE numerical and graphical optimizers using the following goals: USP Resolution >= 2.5 Cp Resolution Robustness >=1.25

Results and Discussion Table 1 Phase 1 – Rapid Screening experiment design generated from the template along with the Trend Response results computed directly from the chromatogram data.

Table 2. Regression analysis results for the Total Peaks trend response. F-Ratio

Lower 95% Confidence Limit

Upper 95% Confidence Limit

---

---

1,324.10

1,507.08

-5.4545

<+/- 0.0001

29.7514

-910.91

-414.61

105.27

-4.3952

0.0001

19.3178

-677.67

-247.69

-113.54

55.15

-2.0589

0.0483

4.2389

-226.16

-0.92

(Grad. Δt)*Column 2

442.20

168.63

2.6223

0.0136

6.8762

97.81

786.60

(Grad. Δt)*Column 3

901.09

170.96

5.2707

<+/- 0.0001

27.7807

551.94

1,250.24

(Grad. Δt)*Column 4

781.87

164.97

4.7396

<+/- 0.0001

22.4635

444.96

1,118.77

(Grad. Δt)*Column 5

1,075.53

163.78

6.5667

<+/- 0.0001

43.1221

741.04

1,410.02

Parameter Name

Coefficient Value

Coefficient Standard Error

Coefficie nt t Statistic

P-Value

Constant

1,415.59

44.80

---

Gradient Δt

-662.76

121.51

Column 4

-462.68

(Grad. Δt)*pH

Table 2 contains two important results. Firstly, all equation (study parameter effect) terms were found to be statistically significant. This can be seen from the significance test values associated with each term in the table (P-Value less than 0.0500, F-Ratio value > 4.0000, zero outside the 95% confidence interval). Secondly, all study parameters are represented in the equation in a form related to the nature of their effects (nonlinear, interaction, etc). In fact, as expected, a ranking of the effect coefficients identifies the largest effect as due to changing columns.

Table 3. Regression analysis results for the Resolved Peaks (> 1.50) trend response.

F-Ratio

Lower 95% Confidence Limit

Upper 95% Confidence Limit

---

---

8.88

9.37

-6.2877

<+/0.0001

39.5346

-2.66

-1.36

0.28

-4.5450

0.0001

20.6568

-1.82

-0.69

-0.38

0.15

-2.6404

0.0130

6.9715

-0.68

-0.09

(Grad. Δt)*Column 2

1.90

0.44

4.2800

0.0002

18.3185

0.99

2.80

(Grad. Δt)*Column 3

3.53

0.45

7.8523

<+/0.0001

61.6582

2.61

4.45

(Grad. Δt)*Column 4

2.56

0.43

5.8987

<+/0.0001

34.7952

1.67

3.45

(Grad. Δt)*Column 5

3.51

0.43

8.1422

<+/0.0001

66.2962

2.63

4.39

Parameter Name

Coeffici ent Value

Coefficient Standard Error

Coefficie nt t Statistic

P-Value

Constant

9.13

0.12

---

Gradient Δt

-2.01

0.32

Column 4

-1.26

(Grad. Δt)*pH

Table 3 shows that all the equation (study parameter effect) terms for the resolution trend responses are statistically significant, and all the study parameters are represented in the equation in a form related to the nature of their effects (nonlinear, interaction, etc). These results show that a predictive equation has been developed for Fusion AE which accurately and quantitatively relates the study parameter effects to a second key aspect of compound separation – the separation of each compound from all other compounds to the extent required.

Once the software derived the equations from the Trend Response data sets, these equations were linked to a numerical algorithm that identified the study parameter settings that maximized both responses. In this study the Fusion AE automated optimization analysis immediately identified the column type, pH, and gradient conditions that should be used in the second phase of the method development workflow. These results are presented in Table 4 below.

Table 4. Phase 1 – Rapid screening experiment Automated Optimizer results. Parameter Name

Optimizer Result Level Setting

Gradient Time

40.0

pH

2.5

Column

Column 3

At this point is is important to remember that in practice, the Trend Response approach will not always yield the optimum HPLC method (instrument parameter settings) in a single experiment, and indeed it is not meant to. The Trend Response approach is part of a phased workflow in which the trend responses enable the experimenter to identify the best settings of parameters such as Column Type and pH; parameters that normally have the greatest effect on separation and therefore cause the most inherent data loss. Once these settings have been identified, these parameters can then held constant in a second experiment to designed optimize the HPLC instrument method.

Table 5. Experiment design generated from the modified template along with the Resolution response results imported directly from the CDS for the Phase 2 Optimization experiment chromatograms. Resolution results were imported for four critical peak pairs (12, 2-3, 5-6, and 9-10), as the compounds corresponding to the other sample peaks were well resolved in all experiment chromatograms.

Figure 7. A trellis of four resolution response surface graphs illustrating the changes in resolution of four critical peak pairs (1-2, 2-3, 5-6, and 9-10) as a function of changing the pump flow rate (X axis) and the final percent organic (Y axis).

Figure 8. Response Overlay graph showing multiple response goals from the Phase 2 Optimization experiment overlaid on one graph. Resolution goals (Maximize, all Lower Bounds = 2.5) for all four critical peak pairs in the DOE-based are displayed.

Figure 9 Response Overlay graph for Phase 2 Optimization experiment with additional overlays of method Robustness Cp goals (Maximize, all Lower Bounds = 1.25) defined for all peak pairs having predicted mean resolution values below 4.00. responses. The unshaded region in this final overlay graph represents the level setting combinations of the study factors that exceed the defined goals for both mean performance and robustness.

Figure 10. Chromatogram obtained by injecting a test sample on the HPLC set at the optimum method parameter settings identified in the Phase 1 and 2 Fusion AE experiments. The final method conditions are defined below. It is noteworthy that the total experimental work required to obtain this final method consisted of two multifactor statistically designed experiments, both of which were carried out overnight in fully automated inject-and- forget (walk-away) mode. Phase 1 – Column/Solvent Screening Column Type – Column 3 pH – 2.5 Gradient Time – 40 minutes Phase 2 – Method Optimization Pump Flow Rate – 0.67 mL/min Final % Organic – 70 %

Conclusions Chromatographic analytical method development work normally begins with selection of the analytical column, the pH, and the organic solvent type. A major risk of using a trial-and-error based one-factor-at-a-time (OFAT) approach is that it provides no ability to visualize or understand the interaction effects usually present among these key instrument parameters. In addition, this approach often results in significant inherent data loss in key chromatographic performance indicators such as compound resolution due to the large amount of peak exchange and compound co-elution common in these experiment data sets.

This inherent loss makes it difficult or impossible to quantitatively analyze and model these data sets, reducing the analysis to a pick-the-winner strategy based solely on visual inspection of the chromatograms. The 2 phased Quality by Design based approach described here, uses statistical experimental design coupled with automatically computed Trend Responses™. This new practice successfully overcomes these problems to provide a rigorous and quantitative methodology for column/solvent screening without the need for difficult and laborious peak tracking implemented in a fully automated HPLC experimentation platform. The Phase 1 – Rapid Screening experiment identified the correct analytical column, pH, and organic solvent type. Once these instrument parameters had been identified, the Phase 2 Optimization experiment involved manipulating the remaining important instrument parameters to obtain a method that met all the performance goals including overall method robustness. The novel Quality-by-Design based methodology used for the Phase 2 experiment combined Design of Experiments methodology with a Monte Carlo simulation to successfully integrate quantitative robustness metrics into the method optimization process resulting a the development of an analytical HPLC method capable of separating four critical peak pairs simultaneously.

Authors Patrick H. Lukulay, Ph.D., Manager, Drug Quality Control and Training USP Drug Quality and Information U.S. Pharmacopeia, Rockville, MD Richard Verseput, President, S-Matrix Corporation. Eureka, CA

Acknowlegements The authors are grateful to Dr. Graham Shelver, Varian, Inc. for providing hardware, software, and expertise in support of the live experimental work conducted to prove out the Quality-by-Design approach to method development presented in this paper. The authors also want to thank Dr. Raymond Lau, Wyeth Consumer Healthcare, and Dr. Gary Guo and Mr. Robert Munger, Amgen, Inc. for the experimental work done in their labs which supported refinement of the phase 1 and phase 2 rapid development experiment templates.