TR E2 (21), E3 (22), and E4 (23) mice [n = 51 (10.51 ± 0.12 months of age); E2: n =15 (10.65 ± 0.22 months of age; n = 6 males and n = 9 females); E3: n = 17 (10.47 ± 0.24 months of age; n = 8 males and n = 9 females); E4: n = 19 (10.43 ± 0.19 months of age; n = 10 males and n = 9 females)] expressing human apoE under control of the mouse apoE promoter on a C57BL/6J background were used in this study. Homozygous breeding of the mice was used to generate the experimental mice for this study. Throughout testing, all the mice were singly housed. Animals were maintained on a 12:00 h light/dark schedule (lights on at 06:00). Laboratory chow (PicoLab Rodent diet 20, # 5053; PMI Nutrition International, St. Louis, MO, USA) and water were provided ad libitum. Behavioral testing took place during the light cycle. All procedures complied with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and with IACUC approval at Oregon Health & Sciences University. Experimenters were blinded to the genotype, sex, and treatment of the mice.

TS100 millimeter-scale (7.5 × 7.5 × 4.2 mm) CubiSensTM wireless sensors (CubeWorks, Ann Arbor, MI), packaged in bio-compatible epoxy and coated with parylene, were implanted in the abdomen for accurate, real-time temperature measurement. The TS100 is capable of transmitting up to 100 m in distance, lasts up to 2 years in sensing operation, and allows measuring circadian body temperature in individual mice. The sensors were sterilized using the Cidex solution (CubeWorks, Ann Arbor, MI). A heating pad and bead sterilizer were used for the surgeries.

For the surgery, the mice were anesthetized with isoflurane (4% for induction of the anesthesia and 1%–3% for maintenance of the anesthesia). Lidocaine (7 mg/kg of 0.5%) was injected subcutaneously around the incision site, immediately prior to the aseptic preparation of the abdomen. To close the abdominal cavity, 4-0 undyed, unbraided, monofilament sutures were used. To close the skin, 9-mm AUTOCLIP stainless steel clips were used. For pain control, meloxicam (10 mg/kg) was administered orally prior to the induction of anesthesia and every 24 h for two additional days. The mice were treated and behaviorally tested starting 2 weeks after the surgeries. Body temperatures were acquired and analyzed for the first week of behavioral testing.

The VLPs were generated to express the SARS-CoV-2 nucleocapsid (N), membrane (M), and envelope (E) structural proteins together with S, allowing us to perform SARS-CoV-2-related studies without replicating virus and outside a BSL-3 facility. Plasmid expression vectors encoding each of the SARS-CoV-2 N, M, E, and S proteins were constructed by standard cloning methods, using synthesized codon-optimized sequences. M, E, N, and S plasmids were transfected at a ratio of 5 µg:1 µg:5 µg:1 µg into suspension-grown Expi293F (25 mL at 3×10⁶ cells/mL) cells using ExpiFectamine reagent (ThermoFisher). Cells were allowed to grow for 4–6 days post-transfection before harvesting. Cells were separated from culture supernatant by centrifugation at 2,000 rpm for 15 min, and culture supernatant was passed through a 0.45-µm filter. VLPs were further concentrated and purified by ultracentrifugation through a 20% sorbitol cushion at 30,000 rpm for 2 h. The pellet was resuspended in 1/100 original volume of phosphate-buffered saline (PBS). Viral proteins S and N were detected by Western blotting ( Figures 1A, B ), and the VLP structure was assessed by transmission electron microscopy of negatively stained samples ( Figure 1C ). Total protein content will be assessed by BCA assay.

For two subsequent weeks, mice were injected daily (weekdays 1–5), each morning 1 h prior to the first behavioral test of that day, with virus-like particles (VLPs) (1 μg/mouse) or vehicle, intraperitoneally, in a volume of 100 μL. The dose was selected based on a preliminary study showing that that dose showed a robust threefold increase in plasma corticosterone levels 1 h following i.p. injection, while a lower dose of 0.3 μg did not.

Body weights were taken prior to the surgery and at the time of the grip strength tests, on day 4 of weeks 1 and 2. The body weight ratio for both weeks was calculated as outcome measure and defined as: (body weight at the time of the grip strength test − body weight prior to the surgery)/(body weight prior to the surgery).

Mice were behaviorally tested as follows ( Figure 2 ). In the morning of days 1 and 2 of both weeks, mice were tested for measures of activity, measures of anxiety, and spatial habituation in the open field. In the morning of days 3 and 4 of both weeks, the mice were tested for object recognition. In the morning of day 5 of both weeks, mice were tested for spontaneous alternation in the Y maze. In the afternoon of days 2 and 3 of both weeks, the mice were tested for sensorimotor function on the rotarod. In the afternoon of day 4 of both weeks, the mice were tested for grip strength. In the afternoon of day 5 of both weeks, the mice were tested for depressive-like behavior in the forced swim test. The behavioral tests were performed as described below.

The mice were put in an open field enclosure (16 × 16 inches, Kinder Scientific, Poway, CA) for 10 min on two subsequent days. On day 3, the open field contained two identical objects for a 15-min trial. The next day, one object was replaced with a novel object for a 15-min trial. Between trial, the arenas and objects were cleaned with 0.5% acetic acid. Interaction within a 2-cm proximity with the object was coded as object exploration by hand scoring videos acquired with Noldus Ethovision software (version 17, Wageningen, The Netherlands). A discrimination index was defined as the time spent exploring the familiar object subtracted from the time exploring the novel object, and dividing the resulting number by the total time spent exploring both objects. A positive discrimination index indicates a preferential exploration of the novel object. A negative discrimination index indicated a preferential exploration of the familiar object. Different objects were used during the first and second week of testing.

The outcome measures in the open field analyzed were as follows (1): distance moved in the open field in the absence and presence of objects, an activity measure (2); the difference in the distance moved in the open field over days, habituation to the open field, a cognitive measure (3); time spent in the center of the open field, an anxiety measure; and (4) the discrimination index, a cognitive measure.

A Harvard Apparatus (Holliston, MA) grip strength meter for mice was positioned horizontally. The mice were allowed to grasp the metal grid and pulled backwards in the horizontal plane. The force applied to the grid was recorded as the peak tension. Three measurements were conducted at 1-min intervals (28). The peak grip strength for each mouse was recorded. In addition, we calculated the relative grip strength as the ratio of grip strength to body weight, as previously described (29). The outcome measures in the grip strength test were the peak grip strength and the ratio of grip strength to body weight.

Activity levels and hippocampus-dependent spontaneous alternations were assessed in week 1 in the Y maze from Harvard Apparatus (Panlab, Holliston, MA, United States). This Y maze is smaller and distinct [raised sides and made of non-reflective opaque gray plastic (30 cm × 6 cm × 15 cm)] from the one used during the second week of testing. For the second week of testing, we used the Y-shaped maze from O’ Hara & Co., Ltd. (Tokyo, Japan) that had raised sides (3.8 cm bottom width, 12.55 cm top width, and 12.55 cm height) with plastic, opaque gray arms (37.98 cm length). The maze was cleaned with 0.5% acetic acid between trials. Performance was assessed during a 5-min trial. Performance was recorded using the Noldus Ethovision software and hand scoring was used to assess the number of arm entries and the percent spontaneous alternations. The outcome measures in the Y maze were total arm entries, an activity measure, and percent spontaneous alternations, a cognitive measure.

The rotarod test (rod diameter: 3 cm, elevated: 45 cm; Rotamex-5, Columbus Instruments, Columbus, OH, USA) was used to assess sensorimotor function. The rotation speed started at 5 rpm and accelerated 1.0 rpm every 3 s. Fall latency (s) was recorded. For both weeks of testing, mice received three subsequent trials on two subsequent days. The outcome measure in the rotarod test used was the mean fall latency of each mouse for each day.

To assess depressive-like behavior, mice were placed for 6 min in a container with water (water height: 15 cm; container diameter: 16–20 cm; 25°C) not allowing the mouse’s tail to touch the bottom. Immobility, defined as cessation of limb movements except for minor involuntary movements of the hind limbs or those movements necessary to stay afloat, was scored manually by an observer blinded to genotype and test history using a sampling technique every 5 s during the trial. The data are expressed as the percentage of immobility (number of immobility observations divided by the total number of observations) during the last 4 min (=48 observations) of the test, as previously described (30).

Following the forced sim test, the mice were euthanized by cervical dislocation. The brain was quickly removed and the hippocampi were dissected in ice-cold PBS and stored at −80°C for analysis. RNA isolated from hippocampi was analyzed by qRT-PCR for expression of the inflammatory mediators TNF-α, IFN-γ, IL-4, and CCL11. RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. Relative expression of cytokines was determined by qRT-PCR using gene-specific primer-probe sets (ThermoFisher) and normalized to β-actin mRNA expression using the ΔΔCt method (31).

All behavioral data are reported as mean ± standard error of the mean and were analyzed using SPSS v.22 (IBM, Armonk, NY, USA) or GraphPad v.8 (La Jolla, CA, USA) software. Genotype and treatment were included as factors in analysis of variance (ANOVA), and sex was used as a covariate. In case there were statistical genotype interactions, genotypes were analyzed separately, as indicated. There were no statistical treatment × sex interactions. When sex or an interaction with sex was not significant, we dropped sex as a covariate and reran the analysis. Repeated measures were used when appropriate. As the E2, E3, and E4 mice were of similar age, age was not included as part of the analysis. The 2 weeks of open field testing and the 2 weeks of testing behavioral performance in the open field containing objects were analyzed separately using day as the repeated measure. Because of the strong practice effects in the rotarod tests, we analyzed the four subsequent days of rotarod testing over the 2 weeks using a repeated measures ANOVA. For the other behavioral tests, week was used as the repeated measure in the ANOVA. For the circadian data, based on the pattern of the data, the light and dark periods for the 5 days were analyzed as separate analyses, with the mean body temperature in the light or dark period of each day as the repeated measure. Based on the three-way treatment × genotype × treatment interaction revealed, we next analyzed the VLP- and vehicle-treated genotype-matched group separately and finally performed an analysis of each light and dark period separately, with the hour as the repeated measure. Statistical significance was considered as p < 0.05. When sphericity was violated (Mauchly’s test), Greenhouse–Geisser corrections were used. Mice were tested in separate cohorts, each containing mice of all experimental groups. All researchers were blinded to genotype and treatment, and the code was only broken after the data were analyzed.

To determine the relationships between behavioral performance measures on the different tests in individual mice, a principal components analysis (PCA) was performed. The behavioral measures used for this analysis are indicated in Table 1 . The PCA was performed using SPSS software and using the varimax rotated matrix. Factors with eigenvalues > 1 were considered significant.

For body weights, there was an effect of week [F (1,44) = 10.638, p = 0.002], with a lower body weight ratio in week 2 than in week 1, and there was a trend towards a week × genotype × treatment interaction [F(2,44) = 2.810, p = 0.071] ( Figure 3 ). In E2 mice, there was a trend towards an effect of treatment [F(1,13) = 3.976, p = 0.0676], with a trend towards a lower body weight ratio in VLP- than vehicle-treated E2 mice. In E3 mice, there was only an effect of week [F(1,15) = 8.858, p = 0.0094], with a lower body weight ratio in week 2 than in week 1. Similarly, in E4 mice, there was only an effect of week [F(1,17) = 12.85, p = 0.0023], with a lower body weight ratio in week 2 than in week 1.

When the activity levels in the open field during week 1 were analyzed, there was an effect of day, with lower activity levels on day 2 than day 1 [F(1,44) = 67.636, p < 0.001] ( Figure 4A ). In addition, there was a trend towards an effect of treatment [F(1,44) = 3.057, p = 0.087]. When the activity level in the open field containing objects (days 3 and 4) during week 1 were analyzed, there was an effect of day [F(1,44) = 41.925, p < 0.001] ( Figure 4A ).

When the activity levels in the open field during week 2 were analyzed, there was an effect of treatment [F(1,44) = 10.183, p = 0.003] and a genotype × treatment interaction [F(2,44) = 5.739, p = 0.006] ( Figure 4B ). In E2 mice, there was an effect of day [F(1,13) = 14.41, p = 0.0022] and a trend towards an effect of treatment [F(1,13) = 3.407, p = 0.0878]. In E3 mice, there was only an effect of day [F(1,15) = 25.7, p = 0.0001]. Similarly, in E4 mice, there was only an effect of day [F(1,17) = 36.75, p < 0.0001]. When the activity level in the open field containing objects (days 3 and 4) during week 2 was analyzed, there was an effect of treatment [F(1,44) = 4.588, p = 0.038] ( Figure 4B ).

When time spent in the center of the open field during week 1 was analyzed, there was an effect of day [F(1,44) = 29.993, p < 0.001], with less time spent in the center of the open field on day 2 than day 1 ( Figure 4C ). When time spent in the center of the open field containing objects during week 1 was analyzed, there was an effect of sex [F(1,43) = 4.463, p = 0.040] and an effect of genotype [F(2,43) = 4.839, p = 0.013], with E2 mice spending more time in the center than E3 (p = 0.0155, Tukey’s) and E4 (p = 0.0024, Tukey’s) mice ( Figure 4C ).

When time spent in the center of the open field during week 2 was analyzed, there was an effect of sex [F(1,43) = 7.839, p = 0.008] and a trend towards an effect of genotype [F(2,43) = 2.708, p = 0.078] ( Figure 4D ). When time spent in the center of the open field containing objects during week 2 was analyzed, there was a day × sex interaction [F(1,43) = 4.637, p = 0.037], an effect of genotype [F(2,43) = 4.477, p = 0.017], with E2 mice spending more time in the center than E4 mice (p = 0.0333, Tukey’s) and a trend towards spending more time in the center than E3 mice (p = 0.0928, Tukey’s), and a trend towards an effect of sex [F(1,43) = 3.860, p = 0.056].

Next, object recognition was assessed. During week 1, vehicle-treated E2 (p = 0.0531, paired t-test) and E3 mice (p = 0.0783, paired t-test) showed a trend towards exploring the novel object more than the familiar one ( Figure 4E ). In contrast, VLP-treated E2 and E3 mice and vehicle- and VLP-treated E4 mice did not. During week 2, only vehicle-treated E3 mice explored the novel object more than the familiar one (p = 0.0176, paired t-test) ( Figure 4F ).

When performance on the rotarod was analyzed, there was an effect of day [F(1,44) = 62.701, p < 0.001] with improved performance with training. There were no effects of genotype or treatment ( Figure 5 ).

For grip strength, there was a week × sex [F(1,44) = 5.25, p = 0.027] and a week × genotype [F(2,44) = 4.747, p = 0.014] interaction. While there was no genotype or treatment effect in week 1 ( Figure 6A ) or week 2 ( Figure 6B ), the grip strength in week 2 was lower than that in week 1 in E2 (t = 3.547, p = 0.0036, paired t-test) and E3 (t = 4.049, p = 0.0009, paired t-test), but not E4 mice ( Figure 6C ).

When spontaneous alternation was assessed in the Y maze, there was only an effect of week [F(1,44) = 4.263, p = 0.045], with lower spontaneous alternation in week 2 than in week 1 ( Figures 6D, E ). We recognize that this might be due to the larger Y maze used in week 2 than in week 1. When activity levels were analyzed in the Y maze, there was an effect of sex [F(1,43) = 6.433, p = 0.015] and a trend towards a week × genotype interaction [F(2,43) = 3.024, p = 0.059] ( Figures 6F, G ).

When depressive-like behavior was tested in the forced swim test, there was an effect of week [F(1,44) = 8.981, p = 0.004], with more depressive-like behavior in week 2 than in week 1, but no effect of genotype or treatment ( Figures 6H, I ).

The circadian body temperatures during the first week of behavioral testing are illustrated in Figure 7 . Based on the pattern of the data observed, the light and dark periods were analyzed as separate analyses, with the mean body temperature in the light or dark period of each day as the repeated measure.

During the light periods, there was only an effect of day [F(2.771,110.855) = 22.958, p < 0.001, Greenhouse–Geisser correction]. However, during the dark periods, there was an effect of day [F(2.495,99.780) = 22.958, p = 0.005, Greenhouse–Geisser correction] and a day × genotype × treatment interaction [F(4.989,99.780) = 2.518, p = 0.034, Greenhouse–Geisser correction], with the E2 mice more affected by VLP treatment than E3 or E4 mice. In general, in E2 mice ( Figure 7A ), body temperatures were lower in VLP- than in vehicle-treated mice, with the most profound effect seen in the dark period on day 4. In contrast, in E3 ( Figure 7B ) and E4 ( Figure 7C ) mice, a more subtle higher body temperature in VLP- than in vehicle-treated mice was seen.

Based on this three-way interaction, we also analyzed each treatment group separately. In the vehicle treatment group, no significant effects or trends were seen. However, in the VLP treatment group, there was a trend towards an effect of day [F(3,57) = 2.668, p = 0.056] and a trend towards a day × genotype interaction [F(6,57) = 2.159, p = 0.060].

We also analyzed each light and dark period separately, using the hour as the repeated measure.

During the light period, on day 1, there was only an effect of hour [F(5.238,183.317) = 22.958, p < 0.001, Greenhouse–Geisser correction]. However, during the light period on day 2, there was an effect of hour [F(6.385,197.920) = 22.958, p < 0.001, Greenhouse–Geisser correction] and a hour × genotype interaction [F(12.769,183.317) = 1.970, p = 0.026, Greenhouse–Geisser correction]. During the light period on day 3, there was only an effect of hour [F(3.501,108.519) = 8.609, p < 0.001, Greenhouse–Geisser correction]. During the light period on day 4, there was an effect of hour [F(4.920,142.685) = 13.965, p < 0.001, Greenhouse–Geisser correction] and a trend towards an hour × genotype interaction [F(9.840,142.685) = 1.832, p = 0.061, Greenhouse–Geisser correction]. During the light period on day 5, there was an effect of hour [F(2.072,64.239) = 1.832, p < 0.001, Greenhouse–Geisser correction] and a trend towards an effect of genotype [F(2,31) = 3.120, p = 0.058].

During the dark period, on day 1, there was only an effect of hour [F(6.971,223.069) = 5.847, p < 0.001, Greenhouse–Geisser correction]. During the dark period, on day 2, there was only an effect of hour [F(6.323,227.621) = 5.378, p < 0.001, Greenhouse–Geisser correction]. During the dark period, on day 3, there was only an effect of hour [F(7.293,269.855) = 9.508, p < 0.001, Greenhouse–Geisser correction]. During the dark period, on day 4, there was an effect of hour [F(5.745,172.353) = 7.596, p < 0.001, Greenhouse–Geisser correction].

Nine factors were identified with eigenvalues < 1.0 and that explained a total of 81.3% of the variance among the behavioral measures ( Table 2 ). Distance in the open field on 7 out of 8 days, distance moved in the center of the open field in 6 out of 8 days, entries and spontaneous alternation in the Y maze in week 2, and the body weight ratio in week 2 all loaded on Factor 1, indicating a common underlying ability being assessed by all these behavioral measures. The directions of the component loadings in Factor 1 were such that increasing values of the factor indicate higher activity measures and increased cognitive performance in the Y maze and an increased body weight ratio in week 2.

Distance moved in the center of the open field in 6 out of 8 days, entries in Y maze in week 2, and fall latency on all 4 days in the rotarod test all loaded on Factor 2. The directions of the component loading in Factor 2 were such that reduced activity levels in the center of the open field and in week 2 and increased activity levels in the Y maze in week 2 indicate increased sensorimotor performance in the rotarod test.

Distance in the open field in 3 out of 4 days in the open field in week 2 and percent immobility in the forced swim test in week 2 loaded on Factor 3. The directions of the component loading in Factor 3 were such that decreased activity in the open field in week 2 indicates increased depressive-like behavior in the forced swim test in week 2.

Performance on all day of the rotarod test exclusively loaded on Factor 4.

Entries and spontaneous alternation in the Y maze in week 1 and grip strengths in weeks 1 and 2 loaded on Factor 5. The direction of the component loading was such that increased activity levels and increased cognitive performance in the Y maze in week 1 indicate increased grip strength in both weeks.

Activity levels in the center of the open field days 1 and 2 in week 1 and body weight ratios in both weeks loaded on Factor 6. The direction of the component loading was such that increased activity levels in the center of the open field in week 1 indicate increased body ratios in both weeks.

The activity levels in the center of the open field on day 1 of open field testing in week 1, the discrimination index in the object recognition test in week 2, and body weight ratios in both weeks loaded on Factor 7. The direction of the component loading was such that increased activity in the center of the open field in the first day of open field testing in week 1 indicate better cognitive performance in the object recognition test but lower body weight ratios in both weeks.

Depressive-like behavior in the forced swim test in week 1 and grip strength in week 2 loaded on Factor 8. The direction of the component loading was such that increased depressive-like behavior in the forced swim test indicate reduced grip strength in week 2.

The discrimination index in the object recognition test in week 2, depressive-like behavior in the forced swim test in week 1, and grip strength in week 1 loaded on Factor 9. The direction of the component loadings was such that more depressive-like behavior in the forced swim test in week 1 and more grip strength in week 1 indicate better cognitive performance in the object recognition test in week 2.

Acute SARS-CoV-2 pathology is largely driven by inflammatory factors that mediate damage to the lungs. Notably, longer-term cognitive sequelae (“brain fog”) are also associated with dysregulated cytokine expression in serum and CSF (24). We examined mRNA expression of several inflammatory cytokines in the hippocampi of VLP/saline-treated mice. Mice were euthanized after the last behavioral test (on the day of the last VLP injection). RNA isolated from hippocampi was analyzed by qRT-PCR for expression of the inflammatory mediators TNF-α, IFN-γ, and IL-4. We also analyzed expression of CCL11, which has been implicated in cognitive dysfunction post-COVID in experimentally infected mice and human patients (24) ( Figure 8 ). Although no significant changes were observed for TNF-α, IFN-γ, and IL-4, CCL11 expression was elevated in VLP-treated animals of the E2 genotype. Thus, expression of TNF-α, which was reported to be induced in the brains of SARS-CoV-2-infected mice at 7 dpi, but not at 7 weeks (24), was not significantly changed in any genotype by VLP treatment.