At present, ITDD is primarily utilized clinically for the treatment of three conditions: (1) chronic non-malignant pain; (2) muscle spasticity; and (3) cancer related pain (Shah and Padalia, 2019). Subcutaneous ports, electronic pumps or single dose injections through the interspinous ligaments in the L3–L5 region are common administration methods for ITDD (Belov et al., 2021). ITDD has been considered a viable option in solving the brain drug delivery problem as it allows for bypassing of the BBB to directly access the CNS (Brumback, 1988; Verma et al., 2020). It works by delivering the drug directly into the cerebrospinal fluid (CSF) within the spinal subarachnoid space (SAS). This route has been found to potentially reduce the required dose 100–300X compared to oral administration, which may lead to lower drug toxicity and reduced side effects while retaining a therapeutic dose to targets of interest within the CNS (Simpson and Jones, 2008; Soderquist and Mahoney, 2010; Calias et al., 2014; Gulur et al., 2014). Furthermore, ITDD is often considered less invasive compared to other drug delivery methods such as intracerebroventricular and intraparenchymal administrations (Belov et al., 2021) and has shown to potentially be more effective (Calias et al., 2012).

ITDD has been investigated in many research studies. A study by Whiteside et al. investigated the efficacy of hyperbaric solutions as spinal anesthesia via the intrathecal route (Whiteside et al., 2001). Previous studies have shown that lumbar intrathecal (IT) administration into the CSF reached the hypothalamus in baboons and dogs (LeBel et al., 1999; McCarthy et al., 2002). A study by Munoz-Rojas et al. found that a patient with Hunter syndrome who received IT injections of loranidase showed improved walk test distance and pulmonary parameters, decreased numbing and tingling, increased stability, and a decreased need for pain medication (Munoz-Rojas et al., 2008). Another study by Muenzer et al. showed that, after 6 months, mean CSF glycosaminoglycans reduced in patients with Hunter syndrome after intrathecal treatment of idursulfase-IT (Muenzer et al., 2016). The use of ITDD for cancer patients has also been investigated (Smith et al., 2002; Rauck et al., 2003). The benefits associated with ITDD compared to conventional pain medication administration methods allowed patients to undergo more aggressive chemotherapy/radiation treatment and henceforth increased patient life expectancy (Smith et al., 2002; Deer et al., 2011). The Food and Drug Administration approved morphine, ziconotide, and baclofen for use via the intrathecal route (Bottros and Christo, 2014). In 2016, nusinersen became the first approved drug to treat spinal muscular atrophy, and it is administered intrathecally as a 5 mL dose within the lumbar SAS (Claborn et al., 2019; Neil and Bisaccia, 2019; Li, 2020). Currently, there are several clinical trials that are investigating intrathecal chemotherapy (Qian, 2016a,b, 2021), stem cell therapy (Kurtzberg, 2014; Staff, 2017; Rong, 2018a,b,c; Lu, 2020; Prodromos, 2021), gene therapy (Sehgal, 2021), and others (McCarthy and Charlesworth, 2017; NIH, 2021).

A thorough understanding of CSF flow dynamics may lead to improved detection and treatment of CNS disorders (Khani et al., 2019, 2020b). Solute transport in the CSF has been shown to play an important role in ensuring drug distribution to the target site (Jose et al., 2013). CSF is a clear fluid that resides in the SAS of the brain and spine (Martin and Heidari Pahlavian, 2019). It has several important purposes, including acting as a “shock absorber” to stabilize intracranial pressure (Yildiz et al., 2017; Martin and Heidari Pahlavian, 2019), providing protection and suspension of neural tissue (Khani et al., 2019), and maintaining immunological and biochemical homeostasis (Engelhardt and Coisne, 2011). The importance of CSF dynamics in CNS disorders has been investigated in several studies (Hårdemark et al., 1988; Klimo et al., 2004; Freedman et al., 2005; Hatterer et al., 2008; Zetterberg et al., 2013; Simon and Iliff, 2016). It is believed that CSF is produced primarily within the ventricles and absorbed at the arachnoid granulations on the surface of the superior sagittal sinus (del Bigio, 2010). CSF pulsates in sync with intracranial cardiac and respiratory cycles and has approximately zero net flow (Martin and Heidari Pahlavian, 2019; Khani et al., 2020b).

Several modeling studies have been conducted to assess intrathecal solute transport in the spinal CSF. In 1996, Myers used an idealized 3D elliptical geometry to investigate the impact of injection flow rate, catheter size, and catheter angle (Myers, 1996). An early study by Tangen et al. (2015) used a patient specific SAS model to investigate the effects of spinal microanatomy on flow patterns and stirring effects. A later study by Tangen et al. (2017) investigated the effect of injection volume and CSF pulsations. Hsu et al. (2012) assessed the impacts of CSF pulsations on ITDD using a 2D geometric model from anatomical images. Haga et al. (2017) and Pizzichelli et al. (2017) investigated the effect of catheter position and angle, tissue permeability, and injection flow rates. Kuttler et al. investigated the impact of a slow vs. fast bolus (Kuttler et al., 2010).

A precise understanding of the impact of ITDD injection parameters on pharmacokinetics may help further optimize intrathecal solute transport and provide drug distribution to the target site of pharmacologic action (Hocking and Wildsmith, 2004; Kuttler et al., 2010; Tangen et al., 2017). Therefore, the goal of the present study was to investigate the impact of several lumbar puncture (LP)-based injection parameters and the injection device types on intrathecal solute transport to the brain in a subject-specific 3D anatomically detailed human in vitro model of the SAS. To our knowledge, this is the first study to investigate a combination of ITDD infusion parameters that include injection volumes and rates, injection location, and injection device. The results of this study provide potential guidelines for intrathecal administration of drugs and further optimize ITDD injection protocols to improve solute transport to the brain for the treatment of CNS disorders.

The development of the subject-specific 3D model used in this study has been previously described by Sass et al. (2017) and Khani et al. (2020b). In brief, a healthy 23-year-old female underwent a high-resolution T2 weighted magnetic resonance (MR) imaging sequence that was used to quantify subject-specific CSF space geometry 5656. The model was developed by combining the result of the high-resolution MR images with 31 pairs of anatomically realistic dorsal and ventral nerve roots, thecal sac, and filum terminale (Sass et al., 2017; Khani et al., 2020b). Stereolithography was used to print the model as a 2 mm thick transparent shell in three parts: cranial, upper thoracic, and lower spine, to avoid exceeding the limit of the 3D printer (Khani et al., 2020b). Once combined, the model was 76 cm in length and had a total CSF space volume of ~330 mL. A detailed description of the subject-specific model, including the connectivity of cervical and lumbar nerve roots, the ventricular system (lateral, third, and fourth ventricles), cerebellum, basal cistern, and the cortical SAS can be found in Khani et al. (2022).

The in vitro system layout previously used by our group has been briefly described by Khani et al. (2020b, 2022). In brief, the following system modifications were made in this study: (a) the addition of two infusion pumps for implementation of the bolus injection and flush, (b) CSF waveform verification, (c) utilization of two cameras to image the system, and (d) implementation of an algorithm for image stitching from the two cameras. A block diagram of the in vitro system configuration is shown in Figure 1.

Imaging configuration, calibration, and post-processing was applied as previously described by Khani et al. (2020a,b). In brief, an imaging camera was used to quantify axial distribution of fluorescein tracer concentration over time. The imaging system utilized two high-resolution cameras with one focused on the brain and the other focused on the spine (Figure 1). Each camera was optimized to attain maximum dynamic range within its respective imaging locations by modification of exposure time. The capturing of images was triggered by cardiac systole to obtain consistent phase acquisition during each CSF flow oscillation. A mask was applied to each camera image to define the model edges for the brain and spine. A manually draw ROI was then specified at the craniocervical junction to delimit the brain only and spine only. To calibrate the system, the model was filled with known molar concentrations of fluorescein spanning the range of dilution within the in vitro model and imaged so that the raw pixel intensities could be converted to molar concentrations. The collected images were used to interpolate pixel intensity to molar concentrations for each location and time. To analyze both images from the brain and spine camera, the images were stitched (Figure 2).

The parameters investigated included (a) bolus injection rate and volume, (b) flush volume, rate, and timing, (c) infusion device as either LP needle or lumbar catheter, and (d) injection location within the lumbar spine (Table 2). The baseline bolus injection volume and rate were set to 5 and 5 mL/min, respectively (1HUM1), a value based on the protocol administered for nusinersen in spinal muscular atrophy (Neil and Bisaccia, 2019). The 1HUM1 condition was utilized as a standard for which to make comparisons across experiments.

To study the effect of bolus injection volume, the bolus injection rate was held constant at 5 mL/min and the bolus volume increased by four times from 5 to 20 mL (1HUM1–2HUM1). Injection bolus rate was observed by increasing the rate from 5 to 13.3 mL/min while maintaining a constant 20 mL bolus volume (2HUM1–2HUM2). To study the effect of flush volume, a 12 mL flush was added to the baseline case with the flush administered subsequent to the initial bolus injection (3HUM1–3HUM3). The impact of flush rate was investigated by increasing the rate from 2.5 to 5 mL/min (1HUM2–1HUM3). The effect of injection device type was observed by switching the 3.5″ 22-gauge spinal needle (~0.4 mm ID, Jorgensen Labs, J-529 H, SNM1018-046) in the baseline case to a 3.5 Fr (0.6 mm ID × 1.1 mm OD) rounded tip polyurethane catheter (Access Technologies, CNC-3.5PR-36″) (1HUM1–3HUM1). Injection location was compared by moving the catheter tip two vertebral levels from L3–L4 to L1–L2 (3HUM3–4HUM2). The impact of a delayed flush was also compared by injecting a 15 mL flush immediately after the bolus injection and subsequent to a 15-min delay following the bolus injection (5HUM1–5HUM4).

To verify repeatability of experimental results three repetitions for each experiment were conducted with an average time delay of 6 days between each repetition. A detailed comparison of repetitions was performed by the following correlation analysis: the standard deviation of the slice average concentration for all three repetitions at each z-location and time was calculated and plotted as a spatial temporal plot to visualize the location of variability between repetitions. Additionally, an array of the slice average concentration mean at each z-location and time for all three repetitions was generated and each individual repetition of an experiment was subtracted from this mean and plotted as a Bland-Altman plot in terms of spatial temporal differences of concentration and location over 3-h post-injection. The standard deviation and the 95% confidence interval (CI) of the mean of the repetitions was calculated. Furthermore, the error as a percent of dynamic range (% DR) was calculated as:

where C_max represents the maximum concentration. Ideally, the % DR is <5%.

To quantify the intrathecal solute transport to the brain at 3-h post-injection, the percent injected dose to the brain (%ID) was calculated as:

The brain portion of the model was defined as the anatomic region superior to the craniocervical junction located at model position defined as 0 cm. Solute exposure to the brain was calculated as the trapezoidal integral value of the tracer concentration over 3-h and is hereon referred to as the area under the curve (AUC).

The mean, standard deviation, and 95% CI in terms of %ID was calculated for each experiment. To assess the significance of a given parameter, the pooled variance was calculated using the following analysis. First, a variance check was conducted in STATGRAPHICS, and the variances were found to be equal. The common standard deviation was then estimated by the pooled standard deviation. The populations were assumed to be independent and normal, such to allow the assumption that the population followed a t-distribution with n₁ + n₂-2 degrees of freedom. The (1–a) 100% confidence interval for m₁-m₂ for pooled variances could be calculated as: x1¯-x2¯ ± tα2 sp1n1+ 1n2 , where x₁ is the mean from sample 1, x₂ is the mean from sample 2, tα2 comes from a t-distribution with n₁ + n₂ – 2 degrees of freedom, s_p is the pooled standard deviation, and n₁ and n₂ correspond to the sample sizes of population 1 and 2, respectively. A one-way analysis of variance in MATLAB was conducted to test significant differences between parameters. An a = 0.05 was used to indicate significance.

The CSF oscillatory waveforms were verified using the Transonic flow meter with agreement between the input of the in vitro model and the ideal waveform (Figure 3). A stopwatch and bucket test was used to calculate the error as percent of average CSF production rate and was found to be <0.02%. Strong repeatability of the slice average concentration was seen between experiment repetitions (Table 3). The greatest standard deviation and 95% CI observed was 1.68 and 3.29 micromoles, respectively, and error as a %DR did not exceed 6.87% for all experiments conducted (Table 3). Agreement of solute distribution with respect to time and location across repetitions can be observed as a spatial temporal plot in Figure 4A. The location with greatest variance for repeated experiments was near the injection location (Figure 4B).

Impact of injection parameters on solute transport to the brain are shown in Table 4. From the conducted protocols, flush volume was the most important factor leading to increased solute delivery to the brain, followed by flush rate, bolus injection volume, location, device, and bolus injection rate (Figure 5). By increasing the flush volume from 0 to 12 mL (3HUM1–3HUM3), solute transport increased to the brain by 3.9% (95% CI 1.6–6.1%ID, p = 0.009). Additional flush volume protocols were conducted, and similar trends were observed. An increase of flush rate from 2.5 to 5 mL/min (1HUM2–1HUM3) increased 1.6%ID (95% CI 0.2–2.9%ID, p = 0.038). Using a catheter increased solute transport to the brain by 1.5%ID (95% CI 0.3–2.7%ID, p = 0.026). By increasing the bolus injection volume from 5 to 20 mL (2HUM1–2HUM2) increased solute transport to the brain by 1.3% (95% CI −0.7–3.2%ID). Moving the catheter tip from L3–L4 to L1–L2 (3HUM3–4HUM2) increased by 1.2% (95% CI−1.9–4.4%ID). An additional location experiment was conducted, and similar trends were observed. An increase in bolus injection rate had the least impact on solute delivery to the brain; an increase from 5 to 13.3 mL/min (2HUM1–2HUM2) showed a decrease in %ID to the brain by 0.6% (95% CI −2.3–1.2%ID). Similar results were observed when comparing the average AUC in the brain at 3-h (Table 4). An increase of 1.2%ID (95% CI −3.2–5.6%ID) was seen with a delayed flush. The bolus injection volume, type of device, location of injection, and bolus injection rate were not statistically significant. In addition, %ID observed in the protocols relating to a lumbar catheter (3HUM1 without flush-−4HUM2 with flush) ranged from 6.0 to 11.6%, while the %ID observed in the protocols relating to a LP needle (1HUM1 with lower bolus injection volume and rate-−2HUM2 with higher bolus injection volume and rate) ranged 4.9–6.5% (p < 0.0001). Supplementary Figures 1–8 show the average spatial temporal tracer concentration and AUC trends for each experiment described in Table 1.

Across all experiments, a ~2 log difference in regional AUC was present between the spine compared to the brain at a time point 3-h after injection (Figure 6). AUC values near the injection site were greatest for all experiments and AUC values around the brain dropped precipitously with a minimum amount located near the superior aspect of the brain. Protocols with the greatest solute transport to the brain displayed the lower AUC values in the lumbar region at 1- and 3-h post-injection (Figure 6). For all cases, the solute distribution differences decreased over time (Figure 7). In 1HUM1 at 10-min, the solute spread to approximately the thoracic region (T8/T9); in 4HUM2 at 10-min, the solute started to spread to the cervical region (C7). However, at 3-h post-injection, the distribution of the solute appeared similar with both cases spreading around the brain, to some degree.

Across all experiments conducted in this study, the goal was to maximize the amount of solute transport to the brain. Regardless of protocol, all LP simulations resulted in limited distribution to the brain. Indeed, the range of %ID observed in the brain at 3-h post-injection was 5.0–11.6% across the protocols tested. Even with several parametric delivery injections analyzed, this indicates a relatively poor brain delivery efficiency of lumbar ITDD, with most of the injected solute remaining in the lumbar spine 3-h post-injection. In the case of spinal cord injury or disease, this may be a desirable result as the target is within the spinal region. In cases of targeting the brain, the efficiency of solute transport observed in this study was, on average, 7.2%ID and dosing decisions would need to account for this limited distribution. Parametric changes in %ID induced by flush volume, rate, and timing, bolus volume and rate, location of lumbar injection, and type of lumbar device did modify solute transport to the brain by −0.6–3.9%ID. While these changes are small, the degree of change in terms of percent from standard LP injection protocol is notable. Indeed, if 5% of the total solute is delivered to the brain, a parametric change in %ID of 3.9% equates to a 78% increase in solute transport to the brain. In this context, one can see how readily optimization of solute transport to the brain can be attained. This demonstrates a potential in which a relatively modest change in injection parameter(s) can make substantial percentwise improvements in brain delivery.

Flushing the injection device may help rinse out residual drug volume and ensure device integrity and is therefore often part of ITDD protocols (Hemley et al., 2012; Chung et al., 2016; Slavc et al., 2018). Our results showed that an increase in flush volume, within the lumbar spine after initial bolus injection, significantly (p = 0.009) increased solute distribution to the brain (Figure 5). Tangen et al. utilized a computational model to investigate two protocols related to flush (Tangen et al., 2017). Their first protocol, with a 5 mL bolus injection and a 5 mL flush, showed high drug concentration spread between the C3 and T5 region; their second protocol, with a 5 mL bolus injection and a 10 mL flush at the same injection rate, showed spread to the upper cervical spine and the brain parenchyma, improving drug delivery to the brain within 1-h post-injection (Tangen et al., 2017). Similar protocols were conducted for our in vitro study relating to flush volume. For both protocols conducted in our study with the 5 mL flush, the solute spread reached the C1–C4 levels at 1-h post-injection, and for a higher flush volume of 15 mL, the solute reached the cranial base within 1-h post-injection; a value in agreement with the Tangen study.

The potential benefit of a lumbar spine flush has been documented in several pre-clinical studies (Hinderer et al., 2014). A study by Wolf et al. utilized a variety of imaging methods to track neuraxial exposure following an IT lumbar bolus injection in rats (Wolf et al., 2016). A protocol consisting of a 30 μL bolus followed by a 40 μL saline flush was tested. Immediately following injection, the bolus filled the spinal SAS and reached the cranial CSF spaces within 2-h, which was interpreted to improve solute distribution to the brain compared to without a flush. To our knowledge, multiple flushes and flush timing has not been non-clinically or clinically investigated. However, repeated bolus injections have been shown to have a therapeutic effect in context of intrathecal pain and spasticity therapeutics: improved functional scores, lower 24-h opioid dose, and less dose escalation (McRoberts et al., 2017). Additionally, multiple bolus doses showed reduced potential for intrathecal fibrosis in dogs when compared to continuous infusion, though this may be accounted for by a drug specific effect and may not hold true for all drugs (Hildebrand et al., 2019). Repeated bolus injections, and therefore perhaps flushes, may also allow lower injection volumes and more rapid CSF pressure recovery post-injection.

Increasing lumbar spine bolus injection rate did not increase solute transport to the brain (Figure 5, Table 4). It has been hypothesized that a greater bolus injection rate may increase solute spread due to potential to increase turbulence at the needle tip that could improve diffusion and mixing of the injected solute (Eisenach et al., 2003; Buchser et al., 2004). Similarly, within intrathecal drug delivery, investigators have considered that increasing an acute bolus injection rate may increase solute transport to the brain. In the present study, we increased the bolus injection rate from 5 to 13.3 mL/min and did not observe any statistically significant differences between the rates. The difference in findings may be a result of differing flow regimes. Typically, intrathecal drug delivery has been used to treat chronic pain and, in those cases, pain medication is often administered chronically in slow doses over the course of several hours or even days on the order of 0.01–0.02 mL/min (Buchser et al., 2004; Flack and Bernards, 2010). This is a value 200–300X slower than that imposed in our acute dosing study that ranged from 5 to 13 mL/min, and thus may not be directly comparable for the scenarios analyzed.

On average, the results for the lumbar puncture catheter experiments delivered approximately twice as much solute to the brain compared to the results of the lumbar puncture needle experiments (p < 0.0001). When the catheter was oriented inferior to superior with the equivalent protocol, a statistically significant (p = 0.026) increase of 1.3 %ID to the brain was quantified. We hypothesized that this is a result of cranial-directed flow velocities originating from the catheter tip during drug injection. To test, the catheter tip was oriented orthogonal to the neuroaxis. Thus, to test the potential impact of device orientation, an additional experiment was conducted with the LP catheter at the same location as the LP needle but oriented in the anterior to posterior direction. This experiment showed no significant difference between the LP needle and catheter. These findings agree with a computational study conducted by Pizzichelli et al. (Pizzichelli et al., 2017) that found catheter angle and position could impact drug penetration to the spinal cord.

Research has indicated that subject-specific CSF flow can be a factor leading to changes in CSF solute transport within humans (Edeklev et al., 2019; Eide et al., 2021; Halvorsen et al., 2021; Ringstad and Eide, 2021). Thus, because the current study was formulated based on subject-specific CSF flow boundary conditions, we expect results to be representative in vivo flow phenomena, but not identical. A study by Verma et al. quantified CSF-brain molecular exchange, neuraxial spread, and CSF-peripheral clearance in 15 healthy human volunteers after intrathecal injection of artificial CSF and Technetium-99 DPTA and observed signal translocation within the cranial cisterns and the brain parenchyma by 3-h post-injection (Verma et al., 2020). Another study measured glymphatic flow in a single 55-year-old male and found traces of gadobutrol delivered intrathecally in the cisterna magna between 1- and 3-h post-injection (Watts et al., 2019). A study by Ringstad et al. observed contrast of gadobutrol in the foramen magna after 20-min and a later study by the group observed contrast in the cerebellum between 2- and 4-h post IT administration in eight healthy volunteers (Ringstad et al., 2017, 2018). In the present study, we observed solute transport within the cranial region between 1- and 3- h post lumbar intrathecal injection, indicating spatial-temporal agreement with these in vivo studies.

It should be noted that our model was constructed with a rigid material while the human spinal anatomy, in specific, the thecal sac, is deformable. As such, the in vivo CSF flow pulsation phase and amplitude vary along the neuroaxis with nearly zero flow pulsation at the spinal termination. Thus, we expect that our model results lack accuracy in predicting solute dispersion near the spinal termination, for example, in the case of spinal anesthetics that are aimed to remain near the lumbar spine. The primary focus of this study was to model spatial-temporal solute distribution to the brain. In this context, our spatial-temporal CSF solute distribution show similarity to several in vivo studies in the literature as described above. Additionally, our study parametrically investigated changes in %ID to the brain and neuraxial AUC distributions rather than baseline magnitude of drug concentrations. As such, we expect the parametric impact of the different injection parameters to be similar even with moderate changes in CSF flow dynamics and/or geometry. However, future research should be applied to understand exactly what degree thecal sac deformation can alter solute transport to the brain.

There are several in vitro modeling limitations that have been previously described by our group (Sass et al., 2017). A primary limitation of the study was that the subject-specific 3D model used in this study neglects to assess drug absorption into the tissues of the CNS. Our approach was therefore to quantify early solute transport within a short time scale post-injection. Depending on the solute absorption, the results can vary substantially and could be modeled in future work by integrating molecular dynamics of the solute and the CNS. All experiments were conducted at room temperature in a rigid model without physiological feedback, which would result in slightly different molecular diffusion and viscosity compared to a human body. In this respect, the addition of a CSF pressure monitoring system in a deforming model could be incorporated for future work. Furthermore, drug specific kinetics have been shown to play an important role in the rate of drug dispersion and tissue uptake (Tangen et al., 2017). This study consisted of a generalized model of solute transport using an aqueous solution of fluorescein to represent a small molecule drug. Additionally, studies have shown that the pulsation rate has a significant effect on solute transport (Hsu et al., 2012; Tangen et al., 2017). A single, consistent, and idealized waveform was used in this study for which to make parametric comparisons across injection scenarios. Previously published studies have shown that the effects of CSF amplitude and frequency oscillations are important factors in solute transport (Khani et al., 2022); however, the primary objective of this study was to investigate the parametric changes of injection protocols. This subject-specific model also utilized semi-idealized geometry that may not fully represent diseased cases, such as syringomyelia or Chiari malformation (Sass et al., 2017). Future studies should investigate solute transport within the CSF for diseased cases. Further, the introduction of a cranio-cervical pulsation, specifically within the highly vascularized cerebellar area would allow for further resemblance to in vivo (Watts et al., 2019; Khani et al., 2020a). Longer duration in vitro studies could be conducted to further investigate the agreement between trends observed in vitro and in vivo; in vivo studies have tracked drug dispersion up to 80-h (Watts et al., 2019) and in vitro studies up to 24-h (Khani et al., 2020a). Lastly, subject-specific in vitro vs. in vivo CSF transport has not been fully validated in this study.