Three-dimensional separation techniques: A review

MSc Chemistry

Track analytical sciences

Literature Thesis

Three-dimensional separation techniques:

A review

By

Ahmet Taskale

12344087

April 2020

12 EC

Supervisor:

Mw. Drs. N. Abdulhussain

1

_examiner:

_{dhr. A.Gargano}

2

_examiner:

_{dhr.prof.dr.ir P.J. Schoenmakers}

Van ’t Hoff Institute for Molecular Sciences

Analytical sciences

Contact information

Title: “Three-dimensional separation techniques: A review”


Master literature thesis in Chemistry - Track of Analytical Sciences

Personal information

Name:

Ahmet Taskale

Address:

Mendelssohnstraat 29, 5011 PA

Tel:

+31(0) 624757979

Table of contents

1. Introduction ... 4

1.1 From 1D to 2D ... 4

1.2 From 2D to 3D, dimensionality and orthogonality ... 5

1.3 Time-based and spatial separations ... 6

1.3.1 Hybrid separations ... 6

1.3.2 Hybrid 3D separations ... 7

2. Modulation in MDC ... 3

2.1 Principles of modulation ... 3

2.2 LC modulation interfaces ... 4

2.2.1 Passive modulation: switching valves ... 4

2.2.2 Switching valves with trap columns ... 5

2.2.4 Modulation techniques ... 6

2.3 GC modulation interfaces ... 8

2.3.1 Thermal modulation ... 9

2.3.2 Valve-based modulators ... 10

3. Coupling the third dimension in 3D-LC ... 13

3.1 Physical parameters ... 13

3.2 Choosing the third column dimensions for LC ... 13

4. Coupling the third dimension in 3D-GC ... 16

4.1 Physical parameters ... 16

4.2 Choosing the third column dimensions for GC ... 17

4.3 Stationary phase chemistry ... 19

4.4 Order of the stationary phases ... 20

5. MDC applications ... 21 5.1 Applications in 3D-LC ... 21 5.1.1 On-line 3D-LC ... 21 5.1.2 Off-line 3D-LC ... 23 5.2 Applications in 3D-GC ... 25 5.3 Applications in MDC ... 27 5.3.1 IMS in the 3D ... 27

5.3.2 LC and SFC hyphenated with 2D-GC ... 28

6. Conclusion ... 32

6.1 Future perspective ... 32

1. Introduction

1.1 From 1D to 2D

Multidimensional chromatography (MDC) techniques are widely used and applied in the analysis of complex samples (e.g. proteomics, food constituents and oil products) due to their high resolving power, high selectivity and efficiency, high peak capacity and sometimes for compatibility reasons with respect to the detector. An example for the latter one is the desalting of the first dimension (1_{D) mobile phase coupled to mass spectrometry (MS)}

detection, since salts are not volatile. Even though one-dimensional liquid chromatography (1D-LC) and gas chromatography (1D-GC) deliver excellent selectivity and versatility, robustness and reliability, quantification purposes and the possibility to combine a variety of detectors, they are limited in the resolving power that they can deliver in the separation of complex samples [1][2]. One way to express the resolving power of a separation technique is the peak capacity. The peak capacity of a system, a concept first described by Giddings [3], is the maximum theoretical number of components that can be successfully separated with a given column, set of parameters and resolution [4]. Giddings has also shown that the theoretical peak capacity (nc) is much higher than the number of components that can statistically be separated as singular peaks (p), as can be derived from equation 1 [5].

ln 𝑝 = ln 𝑚 − _𝑛𝑚

)(𝑒𝑞. 1)

Considering a sample contains m = 500 components, a theoretical peak capacity of nc = 500

can be expected. This results in a complete separation of samples containing 181 peaks. In order to separate 95% of the peaks, a peak capacity of nc = 10 000 would be required.

It is true that in the past decade the performance of LC and GC has improved in terms of resolving power and peak capacity. The application of longer columns under high pressure and sub-2 µm particles for LC [6] and the use of ionic liquid (IL) stationary phase columns and the reduction of the injection pulse width for GC [7]–[9], have led to an increase in peak capacities for both techniques. However, these developments add up to the speed of an analysis, rather than high peak capacities or resolving power, as can be seen in Figure 1.

Figure 1: Peak capacity is plotted against separation window and illustrates the changes in performance for columns as a function of particle size for the analysis of peptides. The colored line represents separations on each respective column as a function of change in %B per column volume. Reproduced from Grinias et. al [6].

Samples containing a certain number of analytes or more are being expected to produce fully or partially overlapping peaks, even in the most advanced and efficient 1D separations. In order to reduce peak overlap (higher resolution) and capture more features of a sample in a certain analysis time, the addition of a second dimension (selectivity) is needed. Two-dimensional (2D) chromatography offers more resolving power and higher peak capacities in comparison to 1D chromatography. In 2D chromatography, the effluent of the 1_{D column}

consisting of fractions with analytes are transferred to the second dimension (2_{D) column}

where an additional separation occurs. Transfer of the analyte fractions can be operated on-line or off-on-line using an interface with various modes. In heart-cut 2D chromatography (1_{D –} 2_{D) one or more fractions are collected and subjected to the} 2_{D column. In case of a}

comprehensive 2D chromatography (1_{D ×} 2_{D) the} 1_{D separation is fractionated throughout the}

entire analysis time and each fraction is subjected to the 2_{D column. Selective comprehensive}

2D chromatography is an intermediate of the two modes mentioned above, transferring a series of fractions across one or more regions of interests to the 2_{D separation.}

1.2 From 2D to 3D, dimensionality and orthogonality

Even though 2D chromatography is superior to 1D chromatography when it comes down to resolving power, there are numerous cases in which the resolving power of 2D chromatography is not enough [2][10]. An example is the separation of a mixture of hydrocarbons in the petrochemical industry. Using comprehensive 2D gas chromatography (2D-GC) highly structured chromatograms of petrochemical samples can be obtained, showing separations based on the distributions of the volatility in the first dimension and the polarity of the components in the second dimension. Despite the fact that 2D-GC is a powerful separation technique which provides higher peak capacities thanks to the reduction of the average peak widths to a few sub-seconds in the second dimension, there is a need for an additional selectivity beside volatility and polarity. In order to gain ordered chromatograms, Giddings [11] addressed requirements for a separation system. In this fundamentally approach the sample dimensionality is given as the number of independent variables that must be specified in order to describe all components within a sample. Indeed, the true sample dimensionality of complex samples like crude oil, food constituents and life science samples far exceeds the number of selectivities or separation dimensions that can be applied. Therefore, only the most relevant sample dimensions which are desired are chosen with a matching number of selectivities or separation dimensions. Ideally, the system dimensionality is the number of separation steps required to separate the sample and is equal to the number of sample dimensions. In the case above, comprehensive 2D-GC fails to separate the naphthenics from the aromatics since aromaticity needs to be added as a selectivity beside volatility and polarity. By adding a third dimension, the number of selectivity matches the number of sample dimensions. The addition of an LC method based on aromaticity has been done by Edam. et al [12]. The resulting three-dimensional (3D) chromatography system provides more resolving power than 1D and 2D chromatography, capable of separating the naphthenics from the aromatics. With the increasing demands of speed and peak capacity (resolution) for analysis of complex samples, the need for multiple dimensions has increased. To fully utilize the increasing resolving power per dimension (2D, 3D), all separation dimensions have to be orthogonal [13]. The high peak capacities that come along with the combination of two or more separation dimensions requires that these dimensions have to be different from each other. The choice of separation mechanisms in each dimension directly

affects the peak capacity of the system applied and in order to gain the maximum effective peak capacity, the separation dimensions have to be chromatographically orthogonal. For example, a mixture of di- and trisaccharide’s show variation in monomeric units and the number of hydroxyl groups, leaving us with two sample dimensions. In order to meet the requirement of orthogonality between separation mechanisms, hydrophilic interaction liquid chromatography (HILIC) and reversed-phase liquid chromatography (RPLC) in the first and second dimension would be an appropriate choice of selectivities.

In this review, a summary of the potential, applications and drawbacks with recently developed 3D separations methods will be discussed. 3D separations can offer extremely high peak capacities but come with their own set of limitations. Different combinations of analytical techniques such as LC, GC, supercritical fluid chromatography (SFC), Ion-mobility spectrometry (IMS) and hyphenation of MDC techniques will be investigated. More focus and details will be given on the modulation interface between different dimensions and the progress made in the past decade. Finally, 3D systems that show high potential for the fast and efficient analysis of proteins, hydrocarbon mixtures and other common complex samples will be discussed and summarized. In addition, this review will provide some future perspectives on the development of spatial chromatography.

1.3 Time-based and spatial separations

Separation obtained by chromatography is either made in a time-space (time-based) or in a physical space (spatial) [14]. Obviously, separation in the time domain is achieved with an analytical column as the separation space. Based on this, separations carried out by LC and GC are considered time-based chromatography (t_{D ×} t_{D ×} t_{D). Spatial chromatography (}x_{D ×} x_{D ×} x_{D) is carried out in the space dimension, also benefitting from a medium exhibiting a}

selectivity. In contrast with time-based chromatography, spatial chromatography relies on the time given to proceed. For the given duration, the analytes move to different locations on the medium, resulting in distribution of different fractions spread across the medium. The medium usually consists of a thin layer of a packing material, a monolith or any other porous material (e.g. paper) coated between two solid plates, also called thin-layer chromatography (TLC).

1.3.1 Hybrid separations

When considering multidimensional separations, the coupling of multiple time-based separations is certainly possible and considerable work has been done over the last 50 years to improve the separation performance of time-based chromatography [10]. The coupling of two spatial-based separations is possible as well, which is done by a sheet of paper combining two successive separations, also known as 2D-TLC [15] shown in Figure 2 below.

Figure 2: (a) Principle of 2D-TLC of a mixture containing five components with mobile phases A and B. (b) Schematics of a 2D separation on one plate of four samples containing the same mixture. Reproduced from [16].

Hybrid coupling like the combination of a time-based separation followed by a spatial separation (t_{D ×} x_{D) has not been done nor described in the literature. This would utilize a} 1_D

separation on an analytical column followed by a 2_{D separation carried out on a flat spatial}

column or bed. The concept of the coupling of these two dimensions is illustrated in Figure 3. The coupling of a spatial separation followed by a time-based separation (x_{D ×} t_{D) allows the}

separation of a mixture in space on a flat medium in the 1_{D followed by a time-based}

separation carried out on a chromatographic bed in the 2_{D, being monitored and recorded by}

a detector [14]. Indeed, the coupling between these two separation dimensions requires the capability of transferring analytes separated in the space dimension onto an analytical column. One way to utilize this coupling is the use of an x_{D ×} x_{D instrument [14][17]. The}

instrument, shown in Figure 4, consists of a steel block with four grooves each one parallel to one side of the square bed cavity. The injection is done manually with a syringe through a septum, which is placed near the corner end of a groove. Thereafter, the mobile phase is pumped into one of the grooves, flowing through the bed and elution is carried out along one side of the steel block. The stream of the mobile phase delivered by the pump in the 1_{D is}

switched off after separation is achieved and the stream of the 2_{D mobile phase is switched}

on, which is in a perpendicular direction [14].

Figure 4: Schematic of a x_{D ×} x_{D instrument. Left: lower half.}

Right: upper half. Reproduced from [14].

1.3.2 Hybrid 3D separations

As for hybrid 3D systems, there are eight combinations of possible time- and space-based separations. These eight combinations are shown in Table 1 below.

The design of systems like t_{D ×} x_{D ×} t_{D and} t_{D ×} t_{D ×} x_{D has not been done nor described in the}

literature. The main reason for this is that the transfer of analytes from a spatial separation to a time-based separation and back will surely cause many difficulties and are considered to be not realistic [18]. t_{D ×} t_{D ×} x_D, t_{D ×} x_{D ×} x_{D and} x_{D ×} t_{D ×} t_{D systems are considered to be not}

practical. The former two systems would require a great number of devices capable of space-based separations, while the latter one would struggle with the transfer of analytes from a spatial separation into an analytical column. Fully time-based and spatial separation systems (t_{D ×} t_{D ×} t_{D and} x_{D ×} x_{D ×} x_{D) [19][20] were investigated, since these systems come in next as}

continuation of t_{D ×} t_{D and} x_{D ×} x_{D systems. It is known that for a} x_{D ×} x_{D ×} x_{D system the most}

2_{D Spatial Space column}

1_{D Time space column}

Flow thr ough the

2D

Figure 3: Concept of the coupling of an analytical column to a flat column or bed.

difficult part is to implement the detection [18]. However, adding a time-based separation (x_D

× x_{D ×} t_{D) will make the detection of analytes certainly possible. Wouters et. al [21] developed}

a microfluidic device for spatial 3D-LC, demonstrating the addition of a time-based separation after a x_{D ×} x_{D separation, resulting in a} x_{D ×} x_{D ×} t_{D system. These systems are in the early}

stage of development and no other work is found in the literature. From a future perspective, spatial 3D-LC holds the key for resolving truly complex mixtures. The reader is referred to section 6.1 for more reading about spatial 3D-LC and its potential.

Table 1: Overview of possible combinations of time-based and spatial separations. Reproduced from [18].

3D system Comments Investigated

x_{D ×} x_{D ×} x_{D Viable Yes} t_{D ×} x_{D ×} t_{D Not realistic No} t_{D ×} t_{D ×} t_{D Viable Yes} x_{D ×} x_{D ×} t_{D Practical Yes} x_{D ×} t_{D ×} t_{D Not practical No} x_{D ×} t_{D ×} x_{D Not realistic No} t_{D ×} t_{D ×} x_{D Not practical No} t_{D ×} x_{D ×} x_{D Not practical No}

2. Modulation in MDC

In this review, only time-based MDC separation methods are summarized. In time-based MDC, the columns are connected by means of a transfer interface, which is called the modulator. As mentioned before, a modulator transfers the effluent of the 1_{D and} 2_{D column consisting of}

fractions with analytes to the 2_{D and} 3_{D column where an additional separation occurs. There}

are various types of modulation interfaces for LC and GC, which will be presented in the next section (2.1).

2.1 Principles of modulation

According to Bahaghighat et. al [22], there are four parameters that evaluate the performance of a modulator. These are the duty cycle, modulation period, the resulting peak capacity and the injection pulse width.

The duty cycle of a modulator is referred to as the fraction of analyte of a preceding dimension that is transferred to the following dimension. A modulator capable of completely transferring the effluent from a preceding dimension to the next one has a duty cycle of 1.0 [22], [23]. Considering a full duty cycle, all thermal modulators have a duty cycle of 1.0 since the columns in such a setup are serially connected and the preceding column effluent is in its entirety being transferred to the next one, providing improvement in sensitivity. Valve-based modulators typically have a duty cycle of ≤1.0 based on the design of the modulation interface. In comprehensive mode a duty cycle of 1.0 is achievable, if the modulation interface is capable of sampling the loops with the effluent from a preceding dimension and completing the following dimension separation before the other loop is filled. However, selective comprehensive and heart-cutting systems are considered to have a duty cycle of ≤1.0, since not all of the preceding dimension effluent is subjected to the following dimension.

The modulation period Pm is referred to as the amount of time needed to sample the

preceding dimension effluent and automatically defines the run time on the following dimension. For example, if the Pm is set at 30 s this means that every 30 s a fraction of the

effluent is sampled. In order to keep sampling at 30 s intervals, the following separation dimension has to be completed within 30 s. The modulation period should be selected in such a way, that it is complementary to the peak width at base (Wb) of the preceding dimension.

Selection of the right modulation period is achieved by providing the right modulation ratio MR [24]. For LC and GC, the modulation ratio is the preceding dimension peak width divided

by the modulation period, as shown in equation 2. 𝑀1 =

𝑊₃

𝑃₅(𝑒𝑞. 2)

The modulation period should be selected in such a way, that the modulation ratio has a value between 2 and 4 [25]. This means that a peak has to be modulated multiple times, since a MR

value of <2 results in under sampling of the preceding dimension separation. Under sampling results in reduction of peak capacity, precision and accuracy [26]. Over sampling, (MR >4) is

not harmful for the preceding dimension separation. However, the peak capacity of the following dimension is reduced, since the following dimension separation space is occupied by more peaks with a higher value of MR.

The peak capacity (nc), as explained before, is referred to as the total analysis time (t) of one

𝑛₎ = 8 𝑡 𝑊₃

8 (𝑒𝑞. 3)

For a 3D system, the peak capacity is given by equation 4: 𝑛_),;< = 𝑛= _{) 𝑛} ) > _𝑛 ) ; _{(𝑒𝑞. 4)}

Equation 4 can be rewritten in order to assist in evaluations of modulation interface performance and the effect on the optimization of peak capacity, resulting in equation 5. However, further consideration based on orthogonality may be needed.

𝑛_),;< = = 𝑡 𝑊₃ = × 𝑃₅ = 𝑊₃ > × 𝑃₅ > 𝑊₃ ; (𝑒𝑞. 5)

In order to maximize peak capacity, the MDC system has to produce efficient separations per dimension, aided by very narrow peak widths in each dimension. The 1_{D peak widths can be}

narrowed down by injecting narrow pulses on the 1_{D column. Reducing the following}

dimension column peak widths relies on the injection pulses from the modulation interfaces. Hence, the resulting peak capacity of a 3D system relies heavily upon the performance of the modulator.

Every modulator, no matter LC or GC, operates on the principle of trapping or isolating a fraction of a preceding dimension and re-injecting that fraction in the following dimension. As will be explained in the following section, thermal modulators trap analytes in a cold region until they are released by rapid heating. When a modulation period of 4 s is applied and the time needed to release those trapped analytes is 40 ms (injected pulse width), the analytes are concentrated 100-fold [22]. Hence, there are two parameters affecting the injection pulse width. First one being the physical dimensions of the region where the analytes are trapped and the time needed to release the trapped analytes using rapid heating.

2.2 LC modulation interfaces

The modulation interface between separation dimensions connecting columns is one of the most important features of any MDC system. When it comes down to LC, 2D- and 3D-LC have always been implemented by using one or more switching valves as modulation interface. These switching valves are usually provided with either sampling loops or trap columns. Besides switching valves, temperature manipulation is a recent upcoming alternative for modulation, which has changed the conventional method for performing MDLC [28], [29].

2.2.1 Passive modulation: switching valves

Conventional LC systems can easily be upgraded to an MDLC system with the addition and installation of pumps and a software which can control the switching valves. There are numerous types of switching valves used, varying from a single valve setup to multiple switching valve configurations. The most common and popular valves used are the 8-, 10- and 12-port two-position switching valves [30]. The 8-port valves, also called the 4-port duo valves, are nowadays gaining popularity in application like the 10-port valves. The reason behind this is that these valves function with a symmetric flow path in both co-current and counter-current, which result in high speed switching in comparison to the older generation 8-port valves [31]. However, one downside of the new generation 4-port duo valves is the lack of

flexibility in instrumental configuration. It is true that combining multiple 4-port valves in an instrumental configuration allows the use of parallel dimensions, but 6-port valves seem to be more popular and more used in multiple valve setups in comparison to multiple 4-port valve setups due to their flexibility in configuration [32]. One advantage of the 4-port duo valve is the lower pressure impact upon switching on the following dimension column. Talus et. al [33] reported that the pressure caused by 10-port valves upon switching had a higher value than for the 4-port duo valves. On top of that, results showed that faster pneumatically actuated valve ports create lower pressure impact upon switching in comparison to electronically actuated valve ports.

2.2.2 Switching valves with trap columns

As switching valves are known to be connecting separation dimensions using two or more loops, integrating trap columns is another way of employing a modulator interface. The primary advantage of switching valves with trap columns replacing sample loops is the function of the focusing mechanism. By adding a focusing mechanism, one fundamental criteria of modulation is being ensured and that is transferring analyte fractions from one dimensions to the next one without contributing to band broadening [34]. Implementing a trap column is simple, since the conventional sample loops are replaced by trap columns. Normally, the trap column consists of the same stationary phase as the following dimension, in order to allow the analytes that are retained to be desorbed by the mobile phase that is used in the following dimension. One of the huge advantages of switching valves with trap columns is that solvent incompatibility is reduced. One example is the combination of RPLC and hydrophobic interaction chromatography (HIC). HIC is known to have salts in the mobile phase, since protein analysis is the major application in this field. By implementing RPLC in a preceding dimension with a gradient increasing in organic phase, the fractions introduced in the following dimension will therefore contain a high percentage organic phase which is not favorable for proteins. However, integrating a trap column solves this problem by focusing the analytes on the trap column and elute the analytes by pumping the mobile phase of the following dimension. Beside focusing the analyte and reducing solvent incompatibility, another advantage is that the modulation volume doesn’t depend on the loop size anymore, but rather on the volume of the trap columns used. On top of that, trap columns of different internal diameters can be coupled. One downside is the trap column robustness and discrimination, since complex samples will most probably have some analytes showing no interaction with the trap column and other analytes showing very strong interactions. Trap columns are mostly used in bio-analysis like proteomics and metabolomics to minimize band broadening, solvent incompatibility issues and dilution [35], [36]. The latter one is to reduce the solvent strength of the preceding dimension, enhancing trapping on the trap column and therefore improving sensitivity. Other applications of trap columns in bio-analysis is on-line protein reduction and protein digestion [35], [36]. After trapping the analytes, a mobile phase containing reduction or digestion reagents are pumped through the column starting the reaction followed by elution with a mobile phase with a different composition.

2.2.4 Modulation techniques

The mode of an MDC system (e.g. comprehensive) refers to what or what part of the effluent from a preceding dimension is being transferred to the following dimension. The modulation interface, described in the previous section, is a device connecting dimensions and transferring effluents containing analyte fractions. The modulation technique is referred to as how the transfer of effluents is done. Modulation techniques can be divided into two categories: passive modulation and active modulation. Passive modulation transfers the effluent from a preceding dimension, unmodified in its entirety, to the following dimension. Active modulation refers to modified parts of the preceding dimension before being subjected to the following dimension.

Figure 5: Schematic overview of a passive modulation interface using a 4-port duo switching valve.

Passive modulation, shown in Figure 5, is by far the most used and most common modulation technique in MDC. Switching valves like the 8-, 10-, 12-port two position valves are used to connect the dimensions using two or more sampling loops. On the left side in figure 5, the first-dimension effluent is sampled by the loop. After this the valve switches and the content of the loop is subjected the second dimension. Right at the exact time of injection, the other loop is connected to the first-dimension column repeating this whole cycle.

Stationary-phase-assisted modulation (SPAM), first introduced by Vonk et. al [37], is currently the most popular active modulation technique as an alternative to the passive modulation described above. The principle of this modulation technique is introduced in section 2.2.2, which utilizes trapping columns rather than sampling loops. The latter one is used in passive modulation. A schematic overview of a SPAM interface is shown in Figure 6 below.

There are several advantages and disadvantages of a SPAM interface. One of the advantages is (i) the reduction in solvent incompatibility. The analytes are focused on a trapping column, which essentially removes the mobile phase from the effluent of the preceding dimension. This is especially important when there are solvent incompatibility issues between connected dimensions. By the removal of the preceding mobile phase, more compatible preceding dimension solvent is thus subjected to the following dimension, relative to passive modulation. The removal of the mobile phase also leads to an (ii) improvement of detection sensitivity. Since the analytes are retained on the trapping column and the mobile phase is

removed, the dilution factor is significantly reduced. Furthermore, (iii) a decrease in the following dimension injection volume is obtained. With the removal of the mobile phase of the effluent, the total injection volume in the following dimension is proportionally reduced. This results in the reduction of the total analysis time, because shorter columns can be used in the following dimensions. Last but not least, (iv) independency of dimensions in terms of analysis times. One of the major drawbacks in comprehensive mode with passive modulation is the completion of the analysis of the preceding effluent in the following separation dimension before the other sample loop is filled with the preceding dimension effluent. By utilizing SPAM, sample loops can be filled over and over again as long as the trapping column doesn’t get overloaded.

Figure 6: Schematic overview of a SPAM interface. As shown, instead of using two sampling loops in passive modulation, SPAM utilizes two trapping columns to retain and focus analytes from a preceding dimension. In order to enhance retention of analytes, SPAM interfaces employ a dilution flow to reduce the solvent strength of the preceding effluent.

One of the disadvantages of SPAM is that all analytes within a sample have to be retained by the trapping columns. Complex samples containing analytes with different sample dimensions have to be retained by a trapping column exhibiting one selectivity. For a separation of a sample containing both hydrophobic and hydrophilic analytes, the appropriate mobile phase and stationary phase must be selected. This can be extremely challenging, considering a sample with a large number of sample dimensions. On top of that, the robustness of the trapping columns also shows a drawback. These columns are usually fragile and have to withstand the extreme conditions of the following dimensions, like the high backpressure (1300 bar). These disadvantages are considered to be the major bottleneck for the implementation of SPAM in high-throughput MDC systems.

Active-solvent modulation (ASM), first introduced by Stoll et. al [38], was developed to solve solvent incompatibility issues. ASM is just like SPAM an active modulation technique, in which parts of the effluent a preceding dimension are modified before being subjected to the following dimension column. As shown in Figure 7 below, parts A and C are similar to figure 3 in passive modulation. Parts B and D, however, refer to a situation where parts of the following dimension mobile phase coming from the pump is used to remove the content of the loop and at the same time mixing the loop volume with the same mobile phase belonging to the

following dimension. This weakens the strength of the mobile phase, enhancing retention on the following dimension column.

Figure 7: Schematic overview of ASM.

Just like SPAM, ASM comes with its advantages and disadvantages. One advantage is the (i) dilution of incompatible solvents, since ASM is designed to dilute the preceding dimension effluent with a weaker following dimension mobile phase composition in order to (ii) enhance retention in the following dimension column and prevent breakthrough. In comparison to SPAM, ASM is (iii) robust and easy to apply. SPAM struggles with robustness and premature elution, since a trapping column provides insufficient selectivity for a sample containing multiple sample dimensions.

Unlike SPAM, ASM is dependent of the volume of the sample loops. The sample loop has to be large enough to sample to analyte fractions in its entirety inside the sample loop and at the same time be fast enough to inject its content before the other loop is filled.

The performance and the possibility of coupling LC modulation interfaces in 3D-LC setups will be discussed in an upcoming section.

2.3 GC modulation interfaces

Modulators in 3D-GC can be divided into two categories. The first one is the thermal modulators [39]–[42], which refers to any modulator that utilizes temperature control to trap

analytes from a preceding dimension (2_{D) and injecting the trapped analyte fractions into the}

following dimension (3_{D). The second one is the valve based modulators [30-31], which refers}

to any modulator that utilizes gas flow to control and trap parts of the preceding dimension. Valve based modulators are also referred to as pneumatic modulation or flow modulation. As there is some difference in terminology, the author of this review has kept the classification as simple as possible by dividing valve-based modulators into two subcategories. Differential flow modulation (including diaphragm valve based) and flow diversion (including Deans’ switching).

2.3.1 Thermal modulation

Thermal modulation, introduced by Liu et. al [45], is the most applied MDGC modulation interface. It utilizes low temperatures to trap and focus analytes from a preceding dimension before subjecting it to the following dimension column by rapid heating. Thermal modulators can be divided into three classes: resistively heated trap, heated sweeper and cryogenic focusing. The latter one is often divided into longitudinal movable trap and jet trap. Resistively heated trap and the heated sweeper were popular during the early 1990’s. However, this popularity has shifted towards the cryogenic modulation since the late 1990’s. It’s true that cryogen-based jet modulators have proven to be consistent and reliable, but the cost of employing these modulators are often considered too high. Recent developments and innovations have been shifted towards the design of an alternative, one providing more simplicity and offering low costs. For that matter, the development of cryogen-free thermal modulators was needed.

There are a number of commercially available cryogen-free thermal modulators. The ZX2 thermal modulator of ZOEX Corporation uses a heat exchanger to employ a two-stage loop modulator, which is capable of modulating C7+ [22]. The ZX2 utilizes a continuous cold jet flow

to trap analytes and a pulsed hot jet to remobilize or re-inject the trapped analytes. Temperatures of -90 °C and 475 °C is achievable using the cold and the hot jet with a minimum modulation period (Pm) of 1 s. When comparing the cryogen-free thermal modulator to a

liquid nitrogen based thermal modulator, the nitrogen based one is capable of trapping analytes down to C3 with a cold jet reaching temperatures near -190 °C. Indeed, the

temperature range and Pm (1 s) of the cryogen-free thermal modulators is limited for certain

application like analytes with high volatility and the utilization in a fast high-throughput 3D-GC system, since the Pm is too slow for the third dimension. However, in most 2D-GC systems

the cryogen-free based thermal modulator should suffice for a variety of analyses.

The solid-state modulator (SSM), designed by J&X Technologies, is the world’s first commercially available SSM. The SSM utilizes thermoelectric cooling, mica-thermic heating and a movable capillary column, as shown in Figure 8 below.

This modulator works with a two-stage method of trapping and re-injecting analytes. There are two zones in which the analytes are trapped and released. The functionality of this two-stage trapping and releasing is to reduce the breakthrough phenomenon. Modulation is achieved in one of the three modulation columns which comes along with the SSM. The selection of modulation columns connects the following dimension. However, each modulation column has a specific application, which has to be chosen carefully. The columns differ in selectivity like C2-C12, C5-C30 and one specific for mineral oils and aerosols.

Luong et. al [46] demonstrated the use of an SSM with a test mixture containing C6-C24 in a

2D-GC system. Pm of 4 s was used in the second dimension, obtaining peak widths of 120 ms

thermal modulator the utilization of this modulator in a third dimension may provide difficulties since it’s not fast enough (two stage trapping and heating) in terms of Pm.

Cryogen-based thermal modulators, LN2 or CO2, remain popular due to the established and

well-understood concept and performance. The quad jet is the most common design, utilizing two stage cold trapping and heat releasing. The cold and heat jets are paired to form two stages of trapping and releasing in order to prevent breakthrough and desorption difficulties. These types of modulators are considered the simplest form of thermal modulation, because of the absence of moving parts.

Figure 8: (A) Schematic of the SSM with key components. (B) Schematic of a 2D-GC system utilizing the SSM. Reproduced from Bahaghighat et. al [22].

2.3.2 Valve-based modulators

Differential flow modulation interfaces, based on diaphragm valves, were introduced by Bruckner et al. in 1998 [47]. Unlike thermal modulation trapping and releasing analytes with the use of jets, current design differential flow modulation interfaces collect the effluent from the preceding dimension in sample loops, which is thereafter flushed with a carrier gas to the following dimension column.

Originally, the interface design was a 6-port diaphragm valve which transferred effluent with analyte fractions from a preceding dimension to the next one, reaching a Pm of 500 ms. Even

after producing remarkable results regarding peak widths and reproducibility, there were however two major drawbacks. Only 10% of the effluent (low duty cycle) of the preceding dimension was transferred to the following dimension. On top of that, the diaphragm valve could not be utilized above temperatures exceeding 175 °C. In order to overcome the latter issue, the valve face was placed outside the oven. This solution extended the operating range of the diaphragm valve up to 250 °C. To overcome the drawback concerning the low duty cycle of the diaphragm valve-based modulation interface, Seeley et. al [48] employed a sampling loop to collect the effluent from a preceding dimension. By using sample loops a higher amount of effluent was introduced containing more of the analyte fractions, which improved detection sensitivity. On top of that, utilizing sample loops creates the opportunity to reverse the flow during re-injection, creating a narrow injection pulse which reduces peak widths. Overcoming the low duty cycle and low temperature range, these high temperature diaphragm valves have gained in popularity. A schematic overview of a differential flow modulation interface is shown in Figure 9 below. The 6-port valve operates with the collect and inject mode. In the collect mode the effluent from the preceding dimension flows in the sample loop and a portion of it out to the waste. After this, in the inject mode, the valve

actuates with the flow of the carrier gas and transferring the collected effluent to the following dimension column.

Figure 9: Schematic of a differential flow modulation interface showing the collect and inject position. On the left during the loading step the effluent from the preceding dimension fills the channel. On the right the modulation valve is switched and the effluent containing the analyte fractions are subjected to the following dimension column. Reproduced from Giardina et. al [49]

Flow diversion modulation interfaces had a breakthrough with the introduction of the Deans’ switch in 1968 and has gained popularity since then. The principle of operation is similar to the differential flow modulation, however, in the case of flow diversion modulation the multiple column flows are inextricably connected with a flow diversion implementation. Figure 10 shows how a flow diversion modulator interface works, illustrating a Deans’ switch [48]. On the left side, the Deans switch is in the no cut position providing the carrier gas. This ensures that the effluent from column 1 is transferred through the restrictor to detector 1 with the aid of the carrier gas. The restrictor is necessary for a pressure drop and ensures that the column flow matches the flow rate of the carrier gas. On the right side, the Deans switch is in the cut position. The effluent containing analyte fractions from column 1 is redirected to column 2 and detector 2.

Figure 10: (Left) Deans switch in the no cut position. (Right) Deans switch in the cut position. Reproduced from [50].

Another unique valve-based modulation technique was introduced by Cai et. al [51] in 2004, which utilizes partial modulation by creating pulses of carrier gas. This pulse flow modulator alternates between a high (positive) and low (negative) carrier gas flows at a T peek connector between two dimensions. This resulted in high and low analyte concentration pulses eluting from a preceding dimension and subjected to the following one. A Pm of 1 s was applied and

peak widths of 60 ms were obtained in the second-dimension separation. Figure 11 illustrates a schematic of a pulse flow valve modulator.

In most cases the pulse flow valve with the flow of carrier gas is open. This dilutes the effluent coming from a preceding dimension before being subjected to the following one. When the

pulse flow valve is closed, the effluent becomes less diluted and the effluent is higher in concentration. After opening the valve again, the carrier gas ‘cuts’ a portion of the effluent which produces a pulse of effluent containing higher concentration of analyte fractions. Later in 2018, this modulation technique was adjusted using a commercially available pulse flow valve modulator interface by Freye et. al [52]. Modulation periods as fast as 50 ms were reported, showing a 4-fold increase in speed in comparison to the work of Wilson et. al [53], who reported a modulation period of 200 ms. When compared to the fastest thermal modulation, which is 250 ms reported by Fitz. et al [54], a 5-fold increase in modulation period is obtained.

The high-speed operating ranges of the pulse flow modulation interface is very promising and shows potential for the implementation in a high-throughput 3D-GC system. The utilization of these modulation interfaces will be further discussed in Chapter 5, providing the application, advantages and the drawbacks.

Figure 11: (A) Pulse flow valve is open, allowing the carrier gas to dilute the effluent stream. (B) The pulse flow valve is closed, the effluent becomes less diluted. (C) Once opening the pulse flow valve again, a short pulse is created. Reproduced from Gough et. al [55].

3. Coupling the third dimension in 3D-LC

3.1 Physical parameters

For a comprehensive 3D system, the condition of being able to subject the entire 1_{D and} 2_D

effluent to the following dimension separation has to be met. This technically means that the

3_{D analysis time must be even shorter than the} 2_{D analysis time in order to complete the}

separation of the preceding dimension effluent. On top of that, the 2_{D flow rate has to be}

selected in such a way that it meets the requirements of the second modulation interface and that of the 3_{D separation. Also, the modulator must be capable of storing all of the effluent}

from the 2_{D column in the sampling loop (modulation period) for the duration of the} 3_{D run}

time. These requirements are applicable when considering 3D-LC systems with modulation interfaces utilizing sample loops, like the switching valves for LC. Physical parameters, like column dimensions for LC (internal diameter, length and particle size), 2_{D and} 3_{D flow rates}

and the volume of the loops, are heavily constrained.

In the following sections, guidelines will be presented for selecting the optimal physical parameters for 3D-LC systems to cope with the constraints. It needs to be mentioned that these guidelines are based on principles for the optimization of comprehensive 2D-LC systems worked out by Pirok et. al [56].

3.2 Choosing the third column dimensions for LC

According to Schoenmakers et. al [57], choosing the proper column dimensions first requires the determination of the maximum 1_{D analysis time and the pressure drop. The} 1_{D separation}

should be slow using a shallow gradient and exhibit the highest separation power, making use of all the space that is available. The 2_{D, however, should be fast and efficient and the} 3_D

should be even faster, more efficient and also compatible with the chosen detector. The 3_D

column is usually shorter compared to the preceding dimension to ensure fast separations (<50 mm). This is very favorable, since short columns are suitable for gradients with high volume ratios. The volume ratio is the duration of the gradient time divided by the column dwell time (tG/to). Short columns with high volume ratios are able to maintain short cycle times

and need less time for column equilibration after a gradient is completed [56]. The selection of the 2_{D and} 3_{D column internal diameters heavily relies on the injection band broadening in}

the 3_{D separation and the dilution factor, which is introduced with the implementation of}

consecutive wider columns. The effect of column diameters on the peak capacity and the total analysis time is shown in Figure 12 below.

Figure 12: Pareto-optimality chart of the 2_{D theoretical peak capacity versus the total analysis time column diameter ratios, which is given by} 2_d

The theoretical peak capacity shown above is representative for a comprehensive 2D-LC system. However, the principle of the highest peak production rate (nc/tanal), achievable with

a certain diameter ratio obtained by the 1_{D and} 2_{D (}2_dc/1_{dc), is also applicable for} 2_{D and} 3_D

(3_dc/2_{dc). The pareto chart in Figure 12 shows that the highest peak production rate is obtained}

with a ratio between 4 and 7. This means that that the 3_{D columns needs to be wider than the}

2_{D column [56]. For example, a correct choice of column diameters would be 0.3 mm, 1 mm}

and 4.6 mm with 150 mm, 100 mm and 50 mm column lengths in the consecutive dimensions. The idea behind wider internal diameters is that these columns allow separation at low volumetric flow rates and keeping proper linear velocities at the same time. Naturally, this will result in small volumes per collected fraction, which on its turn reduces the 3_{D injection}

volumes. On top of that, using wider columns in the 3_{D separation allows the injection of larger}

volumes (increase in column volume) without altering the separation. Furthermore, wider 3_D

columns can operate at flow rates much higher than smaller columns, which also reduces the delay of the gradient used in that dimension.

The selection of the particle sizes in each dimension has to be made in such a way that the particles are resistant against the severe pressure. The 3_{D separation is accompanied by fast}

gradients and high linear velocities. Therefore, the chosen dimension column needs to be robust and needs to deliver high linear velocities and high efficiency. Currently, sub-2 µm particles with ultra-high pressure liquid chromatography (UHPLC) setups are used, since these systems reduce band broadening and exhibit shorter dwell times [58]. According to Pirok et. al [56], smaller particles will improve the peak capacity for fast separations. As can be seen in Figure 13, the utilization of smaller particles results in higher peak capacities for a fast gradient time. However, smaller particles seem to benefit from high peak capacities up to a certain gradient time.

Like mentioned before, the utilization of smaller particles indeed take advantage from the high efficiencies and from the reduced influence of high velocities, which introduces high back pressures. Therefore, besides robustness, backpressure is an essential parameter in considering the 3_{D column in a 3D-LC system.}

ΔP = ΦηLuH

𝑑_J> (eq. 6)

The back pressure is calculated by equation 6 above, with the column flow resistance parameter (Φ), viscosity of the mobile phase (η), length of the column (L), the flow velocity (𝑢0) and the particle size (dp).

Pirok et. al [56] also stated that difficult separations with high plate numbers (N) (N ≥ 100 000) and long analysis times (> 100 min) in the 1_{D are obtained with large particle diameters (≥5}

µm). On the other hand, sub-2 µm particles seem to perform better in simple (N ≤ 25 000) and fast (<1 min) separations. This verifies the implementation of particles decreasing in size per consecutive dimension.

The guidelines described above, concerning the implementation of wider columns and smaller particles in consecutive dimensions, are meant for on-line comprehensive 3D-LC systems. To the best of the authors knowledge, no work has been done utilizing an on-line comprehensive 3D-LC system with analytical columns in each dimension.

Figure 13: The maximum theoretical peak capacity as a function of the gradient time for HPLC separation of peptides. Reproduced from Pirok et. al [56].

One other important aspect is the modulator-loop size. The mobile phase, experiencing pressure applied by the pumps in order to push the mobile phase through narrow tubing and channels, creates a parabolic flow profile. The velocity in the center of a parabolic flow profile is twice as high as the average velocity. Analyte molecules eluting in these faster streamlines could end up being lost, considering a sample loop with a volume of 30 µl and a flow of 10 µl/min. In order to prevent analyte loss, the volume of the sample loop should be double the volume of the collected preceding dimension effluent [56].

Choosing the right modulation time is connected to the holdup time in the preceding dimension. For a 2_{D separation, the time allowed is approximately} ½_{√N times shorter than the}

holdup time in the 1_{D (}1_{t0) [56]. However, gradient elution is mostly utilized in the} 1_{D and the}

peaks of interest usually do not elute near t0. For this matter, a factor of 0.15 √N suffices. For

example, if 1_{N = 15 000 and the holdup time in the} 1_{D is 6 min (}1_{t0), using the factor given}

above an ideal modulation period of 20 s is obtained. Indeed, practicing this guideline to estimate the right modulation time for the 3_{D in a comprehensive 3D-LC system is not}

recommended. Gradient times in separations lasting 20 s are considerably short and therefore applying a factor of 0.15 √N in order to determine the modulation period for the 3_{D is not}

feasible. This implies that the 1_{D run time needs to be very long or the} 3_{D separation time}

needs to be extremely fast, unless a stop-flow method is utilized. According to Schoenmakers et. al [57], the 3_{D separation time is approximately 0.5 s or less corresponding to a} 1_{D analysis}

time of 8 hours. Therefore, it is not possible to achieve a reasonable total analysis time for an on-line comprehensive system with the addition of a 3_{D separation. However, applying this}

factor to an off-line (time decoupled) 3D-LC system or a 2D-LC system hyphenated with IMS, is certainly viable.

Indeed, knowing how to couple the third dimension is not enough to answer the analytical question. Issues like the selection of selectivity’s, compatibility of mobile phases between dimensions and solutions to circumvent solvent incompatibility are of utmost importance. For more in depth reading about these topics, the reader is referred to the excellent work of Pirok et. al [56][10].

4. Coupling the third dimension in 3D-GC

A comprehensive 3D-GC system consists of (i) an injector, (ii) a secondary oven, (iii) two modulators (see section 2.3 ‘GC modulation interfaces’), (iv) a detector and indeed (v) three stationary phase columns. In this section, the reader will be provided with knowledge about how to couple the 3_{D column, the variety in stationary phase chemistry and the order of the}

stationary phases.

4.1 Physical parameters

As for 1D- and 2D-GC, the choice of the injector and the injection-mode in 3D-GC depends on the sample characteristics and the analytical requirements [59]. These characteristics include thermal stability, the boiling point range and the risk of potentially soiling the injector (also known as ‘dirty samples’). As for the analytical requirements, these include the limit of detection (LOD) which refers to bulk or trace analysis, the robustness and last but not least the ease of operation of the required method. In Table 2, the reader is provided with a summary of commonly used injectors in 3D-GC, showing the main characteristics [60]. Table 2: Characteristics of commonly used injectors for the potential implementation in a 3D-GC setup. Reproduced from Daviau et. al [60].

Hot split Hot splitless PTV split PTV splitless Cool on-column Type of analysis Bulk Trace Bulk or labile Trace or labile Trace or labile

Representative

sampling Poor Poor Moderate to high Moderate to high High Inertness Medium Low Medium Medium High

Robustness for

dirty samples Moderate Moderate High High Low Sample volume

flexibility medium medium High Medium including large volume injection

Medium to high including large volume injection A typical 3D-GC system utilizes a temperature program and a dual-oven setup [2], [55], [61]– [70] in which the 2_{D and} 3_{D column is placed in the secondary oven and the} 1_{D column in the}

main oven. By doing so, the system is more flexible in adjusting or optimizing the temperature in the secondary oven and therefore more flexible in adjusting the 2_{D and} 3_{D retention.}

Another reason for the implementation of a secondary oven is the ability to rapidly heating the 3_{D column. The separation in the} 3_{D column injected by the modulator has to be}

completed before another modulation pulse is injected. In order to prevent this overlap of different modulation cycles, a secondary oven is installed [71]. For more reading about the principles of modulation and different modulation interfaces, the reader is referred to section 2.1 and 2.3.

A temperature program is used in a 3D-GC setup to resolve complex samples showing a broad range of retention of analytes. In order to reduce the total analysis time, a temperature program is used during the 1_{D separation, which decreases the retention times of the later}

eluting analytes. Indeed, one might think that a reduction in total analysis time is at the expense of loss in resolution. However, at a retention factor (k) of >2, only a little gain in resolution is obtained at the expense of increasing analysis time as it is derived from equation 7 [59]. The resolution R and the retention factor k (equation 8) are expressed by:

𝑅_QR = 1 4 S 𝐿 𝐻_Q V (𝑟_QR− 1)𝑘_Q (𝑘_Q+ 1) Z (𝑒𝑞. 7) 𝑘 = 𝑡\− 𝑡H 𝑡_H (𝑒𝑞. 8)

Where L is the length of the column, H is the plate height, r refers to the selectivity factor given by ki/kj and k is the retention factor. Even when the condition k> 2 is not met, there is

still room for optimizing the resolution by keeping the plate height H as low as possible, which depends on the linear gas velocity. Increasing the length of the column is less preferable, (which adds a gain in resolution with the square root of L) because an increase in L, increases the total analysis time and the pressure drop across the column. Hence, the utilization of a temperature program aids in the development of a fast and high-throughput 3D-GC system, without showing a significant loss in resolution.

4.2 Choosing the third column dimensions for GC

The choice of the 3_{D column length, internal diameter and stationary phase film thickness (df),}

should be based on the application requirements [59]. The latter one includes the total analysis time, the resulting peak capacity of the system (nc,3D) and the required column

loadability for all dimensions. The condition of modulating a peak 2.5-3 times [56] has to be met, which also implies that the 3_{D separation has to be completed to prevent wrap-around.}

Furthermore, an optimal average linear gas velocity (u) should be used, providing the minimum plate height (Hmin) and maximum number of theoretical plates (Nmax) at the same

time. The maximum number of theoretical plates and the experimental value of theoretical plates (N) can be calculated by equations 9 and 10, respectively:

𝑁_{5_8} = 𝐿 𝐻_5R` (𝑒𝑞. 9) 𝑁 = b 𝑡_c 𝜎e > = 5.54 b 𝑡c 𝑊_H.fe (𝑒𝑞. 10)

where tr is the retention time and σ is the standard deviation of the peak of interest. As

mentioned before, increasing the length of the column is not preferable. Minimizing the plate height H, however, can be achieved by adjusting the parameters that describe H in equation 11, also known as the Golay equation:

𝐻 = 2𝐷5 𝑢 + 𝑓(𝑘) 𝑑₎>_𝑢 𝐷₅ + 𝑔(𝑘) 𝑑_j>_𝑢 𝐷_j (𝑒𝑞. 11)

Where Dm is the diffusion coefficient of the analyte of interest in the mobile phase, dc is the

internal diameter of the column and Df is the diffusion coefficient of the analyte in the

stationary phase. The f(k) parameter describes the mobile-phase flow profile and the g(k) parameter the stationary-phase mass transfer [73], which are described by equation 12 and 13:

𝑓(𝑘) = 1 + 6𝑘 + 11𝑘>

96(1 + 𝑘)> (𝑒𝑞. 12) 𝑔(𝑘) =

2𝑘

3(1 + 𝑘)> (𝑒𝑞. 13)

To make comparisons between columns of different sizes, internal diameters and film thicknesses easier, the following dimensionless parameters (equation 14, 15 and 16) are introduced:

ℎ = 𝐻 𝑑₎ (𝑒𝑞. 14) 𝑣 = 𝑢𝑑₎ 𝐷₅ (𝑒𝑞. 15) 𝛿j = 𝑑j 𝑑₎ S 𝐷₅ 𝐷₎ (𝑒𝑞. 16)

where h is the dimensionless plate height, v is the dimensionless gas velocity and δf is the dimensionless film thickness. From here on, the Golay equation can be simplified and rewritten as:

ℎ = 2

𝑣+ 𝑓(𝑘)𝑣 + 𝑔(𝑘)𝛿j>𝑣 (𝑒𝑞. 17)

Based on equation 17, the f(k) and g(k) factors should ideally be kept low since these contribute to reduction in plate height. It is known that in capillary chromatography (open columns) the plate height usually increases rapidly with increasing k [73]. For these reasons, low retention factors are recommended, because it eventually adds up to the plate height. By utilizing temperature programs, as recommended earlier, the total analysis time and therefore the retention factor of the analytes are reduced. It is clear in Figures 14 and 15 that a large δf value results in large reduced plate height values and it is also clear that the slope of the h-v curve significantly increases with increasing retention factor k.

According to Mommers et. al [59], the film thickness df should not exceed 0.001dc. Based on

Figures 14 and 15, the δf value should best be kept at 0.3 or lower with retention factors around 1. As for the question on how to choose the proper film thickness for the 3_{D column,}

the analyst should consider the mass loadability of all three columns. Thicker films (high δf) are more favorable in the 1_{D since this increases the} 1_{D retention times and peak widths, which}

on its turn is favorable for the modulation performance. The condition of modulating a peak 2.5-3 times is easier when peaks are broad and at the same time reduces the chance of mass-overloading in the following dimension. Furthermore, thicker films are preferable for the analysis of volatiles and low molecular weight (MW) analytes. For example, a correct choice of film thickness per consecutive dimension would be 0.30 µm, 0.18 µm and 0.10 µm [62], [64], [70]. Indeed, there are numerous other combinations possible. The rationale for this example is that the film needs to be thicker in preceding dimensions with respect to the Figure 15: h-v curve showing the influence of the stationary phase

film thickness at k = 1. Lower values for stationary film thickness are preferable. Reproduced from Pirok, B.W.J. [57]

Figure 14: h-v curve showing the influence of the retention factor k. The plate height rapidly increases with increasing k. Reproduced from Pirok, B.W.J. [57].

modulation performance and mass-loadability of the following dimensions, as mentioned before.

As for the length of the columns, shorter columns with thinner films will decrease the retention times and therefore the elution temperature of analytes. This allows the analysis of less volatile and less thermally stable analytes and the use of stationary phases with low thermal stability. Columns with small internal diameter and thinner films provide low plate heights and therefore high efficiencies at high average linear gas velocities, making it ideal for fast separations. However, the drawbacks of these columns include lower column loadability, high pressures drops and a limitation in dynamic range [58-60]. In order to keep up with the modulation cycle, the 3_{D separation needs to be faster than the} 2_{D separation. Naturally, the} 3_{D column should be shorter than the preceding dimension. Furthermore, the internal}

diameter dc of the 3D column should be smaller than the preceding dimension, ensuring speed

and efficiency. Unlike LC, where wider internal diameters are utilized, GC has the advantage of gases which are non-viscous and show high diffusion. For example, for the 1_D, 2_{D and} 3_D

columns, respectively (length, internal diameter and film thickness): 30 m, 300 µm, 0.30 µm; 3.5 m, 180 µm, 0.18 µm; 1 m, 100 µm, 0.1 µm would be a proper choice of column dimensions. Indeed, the example given above is suitable for a high-throughput comprehensive system, when speed of an analysis has the highest priority. The main advantage of utilizing narrow internal diameters per consecutive dimension, is the peak production rate. However, when high resolution is more important, the use of columns with thicker films (1_{df >} 2_{df >} 3_{df) and}

wider internal diameters per consecutive dimension is recommended [59]. 4.3 Stationary phase chemistry

Reaction mechanisms in GC arises from polar and non-polar interactions between the analytes and the stationary phase. The interaction of analytes occurs through a phenomenon called ‘van der Waals’ forces, which is reduced by higher temperatures. The ‘van der Waals’ interactions can be divided into the following: dipole-dipole (Keesom), dipole-induced dipole (Debye) and induced dipole-induced dipole (London dispersion) interactions [61-62].

The use of non-polar stationary phases in GC exhibit induced dipole-induced dipole interactions, also called dispersive interactions. Naturally, there will be retention between a non-polar stationary phase and an analyte exhibiting dispersive interactions, since dispersive interactions are non-polar. It is well known that dispersive interaction forces increase with increasing sizes of analytes. Therefore, the boiling point and the retention time of an analyte increases with the size of the analyte.

The polarity of a molecule is determined by the functional groups and the structure of the molecule itself. Logically, a stationary phase exhibiting polar interactions with a molecule will be more retained in comparison to a non-polar molecular of similar MW. Dipole-dipole interactions are caused by electronegative atoms in a molecule. These electronegative atoms exhibit a permanent dipole and are attracted to each other (dipole-dipole interactions). The permanent dipole of a molecule is expressed in the dipole moment. In case of stationary phase chemistry, the stronger the dipole moments of the stationary phase molecule groups and the dipole moments of the analyte molecules, the higher the retention times.

Hydrogen bonding is another type of interaction causing chromatographic retention. Molecules containing atoms like nitrogen and oxygen serve as proton acceptors. For example, the proton of a carboxylic acid attached to the oxygen atom can bound to another oxygen or nitrogen atom in a different molecule. In this case, the molecule donating the proton (hydrogen) is the donor and the molecule receiving the proton the acceptor. The donating and

accepting molecules will have strong interactions called hydrogen bonding, as explained before. Hence, the selection of the 3_{D column should be based on the possibility of molecular}

interactions between the analyte of interest and the stationary phase chemistry.

Ionic liquid stationary phases, also called liquid salts, are a relatively new type of commercially available columns. Ionic liquid stationary phases are highly polar, show high thermal stability and are known for their low vapor pressures and a broad range of applications [79]. For these reasons, ionic liquids are capable of being utilized at higher temperatures without showing significant column bleeding. Nonionic polar stationary phases show relatively high column bleeding at high temperatures in comparison to ionic liquids, which makes ionic liquids ideal as an alternative. However, despite its high thermal stability, ionic liquid stationary phases are not as mature yet as the conventional GC columns, like the dimethylpolysiloxane (DMPS) and polyethylene glycol (PEG) columns [80]. Unlike the latter ones, ionic liquid stationary phases require further research on column coating, stability of the film and stationary phase chemistry. Despite the lack of further exploration of ionic liquids, there are several cases in which these stationary phases are utilized in 3D-GC setups and examples are increasing in the literature [65], [66], [69].

4.4 Order of the stationary phases

One other important aspect to consider, is the order of the stationary phases in a setup. The

1_{D column needs to provide high resolution and also operates with a temperature program.}

The following dimensions are usually shorter and have smaller internal diameters, which logically provide lower plate numbers. The 3_{D separation is the fastest and operates under}

isothermal conditions, since temperature programming cannot be utilized in these short time frames. As for the selection of the order of stationary phases, the analyst must consider the molecular interactions (section 4.3) between the analyte and the stationary phase chemistries with respect to the fundamental aspects of a 3D-GC setup described above. Furthermore, the extent to which the order of the stationary phases will affect the retention, resolution and separation overall, needs to be taken in consideration as well.

For example, a proper choice of order of stationary phases in 1_D, 2_{D and} 3_{D would be,}

respectively: non-polar, medium polar and polar [55], [62], [64], [67], [70]. Examples of stationary phase chemistries corresponding to these would be a 100% DMPS, 50% diphenyl dimethylpolysiloxane and a PEG column. The rationale for this choice of order is that this column-set provides more information on the sample composition. By using a non-polar stationary phase in 1_{D separation, information on the boiling point will be obtained. The}

following dimensions exhibiting polar interactions will provide polarity ranges and therefore group-type information on both the sample and the matrix analytes. If needed, this information can be used for a better and more tuned set of stationary phase chemistries. It is recommended to use the least polar stationary phases, which provide optimal separation, retention, resolution and orthogonality. Polar columns lack robustness, thermal stability and are prone to suffer from bleed lines in the 2_{D and} 3_{D chromatogram. Ionic liquid stationary}

phases may be considered as alternatives.

There are several examples of utilization of polar stationary phases in the 1_{D [68], [81]. It is}

noteworthy to mention that these guidelines are meant to aid the analyst in the choice of the column-set. However, procedures can deviate from guidelines depending on the analytical question.