*2.2. Intervention Design*

This study was a registered (ACTRN12614000159651) randomised controlled crossover intervention, which was approved by the CSIRO Human Research Ethics Committee. Participants were randomised to daily intake of 50 g WPI or 50 g SPI (control) for eight weeks. Following an intervening 16-week washout phase, participants crossed over to the alternative supplement for the final eight weeks. The randomisation scheme was generated by the clinical trials manager using the online resource website: http://www.randomization.com. Both products were packaged in matching opaque containers with a numeric allocation number to ensure blinding of participants, researchers and clinical trial staff.

Participants were asked to maintain their habitual diet during the intervention trial but refrain from eating foods high in vitamin B12 (i.e., foods with vitamin B12 levels > 4.9 μg/100 g) and also refrain from taking supplements containing vitamin B12/choline/antioxidants). Dietary restrictions included limiting consumption of vegetables or pulses (2 servings/day), fruits or juices (3 servings/day), black tea or coffee (2 cups/day), chocolate (50 g/day), wine (200 mL/day) and/or beer (375 mL/day) in order to avoid an excessive intake of antioxidants to minimise variation between subjects and groups. Participants completed 3-day food and beverage intake records at baseline and endpoint to monitor their adherence to these restrictions and were supported throughout the study by a dietitian.

The WPI or SPI was consumed daily as 25 g blended in 200 mL fruit juice or water twice a day so that the total daily intake was 50 g WPI or 50 g SPI each day. The WPI and SPI were provided in 25 g sachets (two per participant for each day of the intervention phases). A compliance checklist for use of the WPI and SPI sachets was completed by the participants.

During the washout period, participants were allowed to return to their habitual dietary habits and were required to avoid consumption of supplements containing vitamin B12 or foods fortified with vitamin B12. Serum B12 was checked mid-way during the washout period. Those exceeding 350 pmol/L serum B12 concentration were required to restrict intake of foods rich in B12 until they achieved a < 350 pmol/L serum concentration before starting the second treatment phase.

#### *2.3. Nutritional Profile of WPI and SPI*

The nutritional composition of WPI and SPI is shown in Table 2. According to the manufacturer's product data, 100 g of WPI and SPI contributed 6 and 0 μg of vitamin B12, respectively. Independent analysis by PathWest Laboratory Medicine WA (Nedlands, Western Australia, Australia) showed that WPI contained 5.64 μg of vitamin B12 per 100 g, whereas the amount of vitamin B12 in SPI was less than 0.34 μg/100 g.


**Table 2.** Nutritional composition of WPI and SPI as per manufacturer's product specifications.


**Table 2.** *Cont.*

#### *2.4. Biochemical Outcome Measures*

Blood samples were collected at the beginning and end of each of the intervention phases and half-way through the washout phase, that is, at 0, 8, 16, 24 and 32 weeks, within the clinical research unit. Blood samples were used fresh or biobanked frozen at −80 ◦C depending on the requirements of each assay. The biochemical outcome measures included in the present analyses were: serum B12, active B12 (holotranscobalamin), methylmalonic acid (MMA), plasma total homocysteine levels (HcY) and serum folate. Complete descriptions of these measures and associated laboratory processing and results have been described elsewhere [31].

#### *2.5. Tests of Cognitive Function*

The CSIRO Cognitive Assessment Battery (C-CAB) comprised of 10 individual cognitive tasks that measure cognitive domains including processing speed, reaction time/attention, reasoning speed, verbal and numeric working memory, immediate and delayed word memory. The tasks utilised obtain latency/response time and accuracy data with performance outcomes for each measure reflecting both of these components. The test battery was programmed to run in the Inquisit version 4 environment [32] and the battery took approximately 45 minutes to complete. Participants completed the C-CAB at each of their clinic visits. The tests used to measure performance on each of the cognitive domains are now described in the order completed.

#### 2.5.1. Immediate Word Memory

Participants were presented with a list of 15 words obtained from the MRC Psycholinguistic Database [33], which included one-to-two syllable words with high concreteness (Mean = 572.81, SD = 35.92) and meaningfulness (Mean = 681.87, SD = 83.31) ratings. Prior to presentation, participants were instructed to remember the words and were informed they will be tested immediately, as well as after approximately 20 minutes. Words were presented in the middle of the computer, one at a time, with a display time of 2-seconds and an inter-stimulus interval (ISI) of 500-millseconds (msec). Parallel word lists were used for alternate testing sessions to remove practice effects. Immediately after presentation, participants were tested using a 30 words list. The list included the original 15 words plus 15 new distractors. Words were shown one-at-a-time in the centre of the computer screen with a 500 msec ISI between the participant's response and subsequent word stimulus. Participants indicated whether the word was from the initial word list or not, by pressing 'Yes' and

'No' response keys. Raw outcome measures included mean response latency—time between stimulus onset and participant's response—and accuracy measures.

## 2.5.2. Processing Speed

The first task assessing processing speed was adapted from previous research [34]. A code table was presented at the top of the computer screen and comprised nine symbols arranged horizontally, to which nine digits, presented directly beneath them, were paired. For each item, one symbol was presented in the centre of the computer screen and participants responded by left clicking the mouse on its corresponding digit in a 3 × 3 numerical response grid positioned at the bottom of the screen. The raw outcome measure was the number of items correctly completed in 90 s.

The second task was a computerised string search task adapted from the ETS Kit of Factor Referenced Tests [35]. In this task, participants were required to decide whether two number strings, presented side-by-side, were identical and press the 'Yes' or 'No' key on the keyboard accordingly. The raw outcome measure was the number of items completed correctly in 90 s.

#### 2.5.3. Reaction Time/Attention

Two-choice and four-choice reaction time tasks were used to measure this domain and were adapted from our previous studies [34]. To begin each trial, the participant pressed the 'H' key on a response keypad. This action triggered a cue symbol (+) of approximately 40 px which appeared in the centre of the computer screen together with two or four black bordered empty target squares (100 px) positioned 150 px each side of the cue (left and right for two-choice, with addition of top and bottom square for four-choice). After a variable ISI ranging from 1000 msec to 2500 msec, a black arrow was presented in one of the squares (arrow direction was congruen<sup>t</sup> with location; pointing left, right, up or down). Participants responded by pressing the corresponding response arrow in the response keypad as quickly as possible (response keys were positioned left, right, above and below the 'H' key). Practice phases required participants to complete four and eight trials correctly. Mean response latencies for each task were subsequently estimated over 30 valid trials defined as: 1) response was correct; 2) response latency was not less than 150 ms; and 3) response latency was not greater than mean latency + 3.5 times the intra-individual standard deviation of latencies obtained during the second phase of practice trials in each task.

#### 2.5.4. Verbal Working Memory

This was measured using a letter memory task adapted from previous studies [36]. In this task, participants were initially instructed to memorise a set of five letters presented sequentially on the computer screen for 2000 msec each with an ISI of 500 msec. Following a warning probe of 3000 msec, 31 test trials commenced and participants were required to indicate whether the test letter displayed was in the original learned set or not. In the second phase of this task, participants were required to remember a new set of eight letters which did not include the original five, followed by 31 test trials as per above. Raw outcome measures included mean response latency and accuracy.

#### 2.5.5. Numerical Working Memory

This was measured using a number memory task adapted from previous studies [36] and was operationally equivalent to the verbal working memory task described above. However, in this task, participants were instructed to memorise sets of numbers (either five or eight) and to make yes/no decisions about numbers presented subsequently in the test phase. Raw outcome measures included mean response latency and accuracy.

#### 2.5.6. Delayed Word Memory

Approximately 20 minutes into the cognitive battery, participants were tested again on the 15 words presented at the start of the session. The test consisted of 30 trials involving the original 15 words, plus 15 new distracters not used in the immediate recognition test. All other presentation parameters and scoring parameters were as per the immediate recognition test described above.

## 2.5.7. Reasoning Speed

Two tasks adapted from the odd-man-out reaction time paradigm were used to assess this domain [37]. The tests used are more complex than simple reaction time because they assess the ability to make reasoned decisions concerning distal relationships between stimuli [38]. In the first version, trials consisted of eight circles arranged horizontally with the total stimulus set being 640 px wide and 84 px high. In each set, three target circles were illuminated red and two targets were closer together in relation to the third target. Participants indicated to which side-left or right-of the two closer targets the odd-man-out (i.e., third target) was located. Participants completed two practice phases consisting of six trials each. The test-phase consisted of 40 trials and raw outcome measures included mean response latency and accuracy.

The second task is procedurally equivalent to the first task. However, in this version, trials consisted of eight mixed shapes (squares, diamonds and triangles) arranged horizontally. The target stimuli were triangles as opposed to illuminated circles and participants indicated to which side—left or right—of the two closer triangles the odd-man-out (i.e., third triangle) is located. Outcomes measures were as per above.

#### *2.6. Sample Size, Data Processing and Analysis*

The required sample size for this study was estimated using GPower. Results showed a minimum of *n* = 74 participants were required for a repeated-measures model assuming a small within-between interaction effect (η<sup>2</sup> = 0.02) with 90% power. For crossover designs, as used herein, this figure is halved, requiring a minimum of *n* = 37 participants. The number of recruited participants who completed this study was *n* = 44, thus satisfying statistical power requirements.

Transformations were undertaken to normalise distributions of cognitive tasks for analysis purposes. In line with previous studies [39], work rates representing the number of items completed per second were calculated. Then, these work rates were adjusted for accuracy (work rate × % correct). Finally, performance was averaged across tasks measuring the same domain (e.g., reaction time/attention) to arrive at a composite measure.

Cognitive data were analysed at the Intention-to-Treat level and therefore included all participants who commenced the trial (whether or not they withdrew). Change in cognitive function was assessed using linear mixed effects models—capable of handling missing data—and the repeated nature of the measurements were accounted for by permitting random slopes and intercepts for each participant for each treatment. Fixed effects included time, treatment and time by treatment interactions. An unstructured covariance matrix was specified for random effects and all models controlled for participant's sex, age and the order of assigned treatments. The assumption of normally distributed residuals was assessed and was satisfied for all models.

The final analyses assessed the relationship between changes in vitamin B12 and related parameters occurring during the WPI intervention period only and corresponding changes in cognitive function using first-order correlations adjusted for sex and age covariates. The calculation of change scores was thus based only on those who attended final follow-up (*n* = 44) and who therefore had complete data.
