Microglia-containing brain assembloids model early HIV infection, cell by cell

What the work claims

This is a primary model-building and single-cell characterization study. Its central claim is that a cerebral assembloid with co-developed microglia reproduces the cell-type selectivity of HIV brain infection: when the R5-tropic, macrophage-tropic strain NL-AD8 is added to the culture medium, intact proviral DNA and spliced viral RNA appear only in microglia-containing assembloids, never in organoids without microglia, and HIV protein colocalizes specifically with the tagged microglial lineage. Around that claim sit three supporting results: microglia at physiologically relevant densities accelerate neuronal maturation in uninfected assembloids; infected microglia adopt a reactive, interferon-high, pro-inflammatory state that signals outward to neurons and astrocytes; and uninfected bystander microglia, plus neurotrophic ligands from other cells, mount a measurable counter-program aimed at repair.¹

How it works

The construction is the contribution as much as the infection result. Rather than adding finished microglia to finished organoids, the authors combine CD34-positive, CD45-positive myeloid precursors with neural progenitors at a 7:3 ratio on day 1, in the presence of IL-34, TGF-beta1, and M-CSF, so microglia differentiate alongside the neural tissue they will inhabit. A permanently activated nuclear tdTomato reporter in the myeloid lineage makes every microglial cell traceable. By day 15, IBA1-positive microglia make up about 30 percent of cells and disperse to roughly 10 to 15 percent by day 25, bracketing the 5 to 20 percent microglial fraction of the human brain. Single-cell RNA sequencing maps the microglia onto human brain atlas references with homeostatic marker expression, though with reduced P2RY12 and TMEM119, an immaturity the authors attribute to the absence of vasculature and in vivo cues.

Infection is introduced as free virus in the medium, mimicking how HIV seeds the brain across the blood-brain barrier during peripheral infection. Digital PCR shows intact proviral DNA rising by three days post-infection in microglia-containing assembloids only, spliced viral transcripts peaking at four days and declining by five as infected cells die or, the authors propose, enter latency; control organoids without microglia show no virus at all, and HIV Tat protein colocalizes with tdTomato-positive cells. Single-cell analysis then splits microglia into three states: HIV RNA-positive cells loaded with interferon-stimulated genes (MX1, XAF1, IFITM3, OAS3, IRF9) and pro-inflammatory ligands (IL1B, TNF, TNFSF10) while losing homeostatic markers (P2RY13, CX3CR1, TGFBR1); bystander microglia in the same infected assembloids sitting in between, with higher TREM2, VEGFA, and the immunoregulatory ligands SERPING1 and SIRPA; and uninfected microglia from control assembloids at the homeostatic pole.

Neurons respond to this without dying. At three days post-infection they upregulate MHC class I and II antigen-presentation genes (CD74, B2M, HLA-B and HLA-C), endoplasmic reticulum stress and protein-quality-control programs, and anti-apoptotic and neurotrophic genes (PTN, MDK, PRDX2). Ligand-receptor inference (NicheNet) assigns the stress side to microglia-derived TNF-family and IL18 signaling and the protective side to BDNF, VEGFA, and FGF2 from microglia and other cells. The same analysis in uninfected assembloids shows why microglia matter even before infection: microglia-bearing assembloids have more neurons, enriched in synaptic assembly and neurotransmission genes (RELN, NEGR1, ROBO1, FOXP2), with predicted microglia-to-neuron synaptogenic signaling through neurexin, neuroligin, neuregulin-ERBB4, and NEO1 pairs.¹

Where a skeptic should push

The single most load-bearing assumption is that this system models HIV's selective microglial tropism in the adult human brain. The assembloid is an embryonic-like construct: microglia carry reduced mature markers precisely because the tissue lacks vasculature, oxygenation gradients, and developmental timing, and the neurons are young. HIV in the real brain enters a mature, myelinated organ in people whose immune systems are under years of pressure from systemic infection. A three-to-six-day infection window in fetal-like tissue captures early seeding biology plausibly, but nothing here demonstrates that latency, reactivation, or chronic neuroinflammation will behave the same way, and the paper's own comparison to published late-stage models shows its neurons sit transcriptionally closer to homeostasis than neurons in those systems.

Second, the lineage and ligand conclusions are inference-heavy. NicheNet ranks ligand-receptor pairs from expression data; it predicts who could be talking to whom, not who is. Every statement about microglia "driving" neuronal stress or bystanders "mounting repair" rests on correlations plus pathway enrichment, without functional perturbation such as microglial depletion or pathway blockade in this study. The bystander-neuroprotection story is attractive and consistent with patient multi-omics work the authors cite, but it is a hypothesis the model generates, not one it tests.

Third, generalization across donors is thin. The lineage-traced work is built on one genome-edged reporter iPSC line (with a second unedited line used for key immunostaining confirmations), pooled organoids from two donors in the viral-load assays, and roughly forty organoids per condition in the day-18 sequencing. That is respectable for a methods paper but well short of what a drug-screening claim would need: nothing here establishes that the infection rate, the bystander program, or the neuronal stress signature is stable across the human genetic diversity that HIV drug programs must eventually cover. The claimed early latency is likewise an interpretation of declining spliced RNA, not a demonstrated reservoir.

Brain organoid infection models, upgraded

The non-obvious implication for organoid-based drug discovery is negative and structural: without microglia, this infection does not happen at all. Cerebral organoids lacking microglia showed no detectable provirus or viral RNA under identical exposure, which means every screen for anti-HIV brain agents, latency-reversing compounds, or neuroprotective adjuncts run in standard brain organoids has been testing a tissue the virus cannot productively enter. The resident immune compartment is not a refinement; it is the entry ticket. The same logic extends to other brain-tropic viruses and to neuroinflammatory disease modeling generally: if the cell type that first meets the pathogen is missing, the model's negative results are uninterpretable and its positive results may not transfer.

The opportunity is a screening substrate with two separable readouts. Infected microglia and bystander microglia are transcriptionally distinct states in the same dish, so a compound that shifts microglia from the interferon-high reactive pole toward the homeostatic pole, or that amplifies the neuroprotective ligand program, is directly measurable by single-cell readouts already demonstrated here. That is a concrete, mechanistically grounded assay design for the long-sought class of ART adjuncts aimed at HIV-associated neurocognitive impairment, a condition antiretrovirals treat incompletely. The tdTomato lineage trace plus digital PCR proviral quantitation also gives a workable primary readout for reservoir-targeting compounds, with the caveat that true latency remains to be shown.

The threat is overclaiming the platform. The ligand-receptor maps are predictions; the neuroprotection is compensatory expression, not rescued function; and there is no electrophysiology, no synaptic physiology, and no behavioral or in vivo validation in this paper. A screening program built on this model would be wise to treat the single-cell signatures as candidate biomarkers to validate with perturbation, and to demand donor diversity before translating a hit. The deeper caution generalizes: assembloids are better at revealing which cell type matters than at proving how, and the distance from a beautiful interaction map to a druggable mechanism is where most of these models quietly fail.

The bottom line

Established: co-developed microglia integrate into cerebral assembloids at physiological densities, support neuronal maturation, and are the exclusive productive target of R5-tropic HIV in this system; infected microglia broadcast interferon and pro-inflammatory signals while bystander microglia retain a homeostatic, neuroprotective-leaning profile; and neurons respond with stress and repair programs rather than death at this early stage. Hypothesis: that manipulating this microglial balance, dampening reactive signaling or reinforcing the protective one, would slow HIV-associated neurocognitive impairment in patients on suppressive therapy. What would confirm it: perturbation experiments in the assembloid (microglial depletion, pathway blockade) showing the neuronal stress signature is causally microglia-dependent, demonstration of genuine latency and pharmacological reactivation, and replication across a donor panel large enough to trust for screening. What would break it: donor variability that swamps the reactive-versus-bystander distinction, or evidence that the fetal-like microglial state in these constructs responds to HIV unlike adult human microglia in vivo.